DE10296675T5 - Virtual networking system and method in a processing system - Google Patents

Abstract

A method for emulating a switched local Ethernet network in a platform with a plurality of computer processors, a switch fabric and point-to-point connections to the processors, comprising:
Providing Ethernet driver emulation logic for execution in at least two computer processors;
Providing switching emulation logic for execution in at least one of the computer processors;
Establish a virtual interface between the switch emulation logic and each computer processor in which an Ethernet driver emulation logic is executed to enable software communication therebetween, each virtual interface via the switch fabric a software communication path from a computer processor to one other computer processor defined;
Establishing a virtual interface between each computer processor in which ethernet driver emulation logic is executed and each other computer processor in which ethernet driver emulation logic is executed; in which,
if the virtual interface between one computer processor and another works in such a way that it fulfills predetermined criteria, the ethernet driver emulation logic of the one computer processor via a virtual interface which defines a software communication path therebetween from point to point with .. ,

Description

background 1. Field of the Invention

The present invention relates on computer systems for Businesses and application service providers, and more specifically on processing systems with virtualized communication networks.

2. Discussion of the related area

In the current corporate IT and Application service provider environments need to be staffed by several Information technology functions (IT functions) (electrical, networking personnel etc.) in the use of processing and networking resources participate. Because of planning and other difficulties with the Coordination of activities several departments can consequently use a new computer server Take weeks or months. This lengthy, manual process elevated both personnel and equipment costs and delays the Start applications.

Since it is difficult to predict how Much processing power requires applications, managers see Moreover typically too much computing power. As a result data center EDP equipment is often underused or underused.

Finally, if further processing power is required than originally was intended the different IT functions to coordinate the activities again deploy more or improved servers, they with the communication and to connect storage networks etc. This task is increasing difficult while the systems are getting bigger.

The use is also problematic. For example when using 24 conventional Servers over 100 discrete connections may be required to to configure the entire system. The management of these cables is an ongoing challenge, and each represents a failure point Try to reduce the risk of failure by adding Can mitigate redundancy double the wiring, which exacerbates the problem while it the complexity and increases the cost.

The provision of high availability is a difficult and costly goal with today's technology. General must for every primary server a failover server can be used. Also are common complex management software and specialist services required.

Generally it is impossible to an older server discontinue processing power or update CPUs. Instead, scaling processor capacity and / or required the transition to the architecture of the next generation of a provider frequently a "forklift update", i. H. adding more Hardware / software systems, new connections and the like need.

Hence there is a need for one System and a process to create a platform for corporate and ASP-EDP, which deal with the above disadvantages.

Summary

The present invention provides a platform and method for computer processing in which virtual processing area networks configured and used can be.

According to one aspect of the invention are a method and system for emulating a mediated local Ethernet network created. There will be several computer processors and a switch fabric and point-to-point connections to the Processors created. A virtual interface logic provides virtual interfaces over the Switch fabric and the point-to-point connections. Any virtual Interface defined via the switch fabric a software communication path from one computer processor to another computer processor. In Ethernet driver emulation logic is used for at least two computer processors executed while switch emulation logic in at least one of the computer processors accomplished becomes. The switching emulation logic creates a virtual interface between the switch emulation logic and each computer processor, by using an ethernet driver emulation logic accomplished to enable software communication in between. Moreover receives a message from one of the virtual interfaces to one Computer processor in which the Ethernet driver emulation logic is executed being assigned addressing information in response to the message sends a message to another computer processor in which Ethernet driver emulation logic is running. Also poses a virtual interface between each computer processor, in which an ethernet driver emulation logic is executed, and any other computer processor that contains an ethernet driver emulation logic is performed, forth. If the virtual interface works satisfactorily, the Ethernet driver emulation logic communicates via a virtual interface that defines a software communication path in between, from point to point with another computer processor in the emulated Ethernet network while them over the switching emulation logic communicates from point to point, if the virtual interface is not working satisfactorily.

According to a further aspect of the invention, a method and a system for implementing an address resolution protocol (ARP) created. An EDP platform has a plurality of processors that are connected by an underlying physical network. A logic executable in one of the processors defines a topology of an Ethernet network to be emulated in the EDP platform. The topology includes processor nodes and a switching node. Logic executable in one of the processors assigns a set of processors from the plurality as processors to act as the processor nodes. A logic executable in one of the processors assigns virtual processor addresses to each processor node of the emulated Ethernet network. A logic executable in one of the processors assigns virtual interfaces via the underlying physical network in order to provide direct software communication from each processor node to every other processor node. Each virtual interface has a corresponding identifier. Each processor node has ARP request logic that transmits an ARP request to the switch node, the ARP request containing an IP address. The switching node contains ARP request broadcast logic that transmits the ARP request to all other processor nodes in the emulated Ethernet network. Each processor node has ARP response logic that determines whether it is the processor node associated with the IP address in an ARP request and, if so, an ARP response at the switch node that outputs the virtual Contains the MAC address of the processor node that is assigned to the IP address. The switching node contains ARP response logic that receives the ARP response and changes it to include a virtual interface identifier for the ARP request node.

According to another aspect of Invention will be a platform and method for computer processing created that support processor failover. Several Computer processors are connected to an internal communication network. About the internal network, a virtual local communication network is defined and manufactured. Every computer processor in the virtual local Communication network has a corresponding virtual MAC address, the virtual local area network communication among a lot of computer processors allows while it those processors from the majority that are not defined in the Amount are excluding. It becomes a virtual storage room with a defined correspondence to the address space of the filing network defined and manufactured. In response to a computer processor failure, a Computer processor from the majority assigned to the failed Processor replaced. The MAC address of the failed processor is assigned to the processor that replaces the failed processor. The virtual storage room and the defined correspondence of the failed Processors are assigned to the processor that is the failed processor replaced. The virtual local area network is restored so that it contains the processor which replaces the failed processor while the failed processor is excluded.

According to another aspect of Invention will be a system and method of providing of a service addressed by an IP address. At least two computer processors each contain logic that provides the service. Cluster logic is receiving Request messages for the service. The message has the IP address. The cluster logic distributes the requirements to one of the at least two computer processors, who has logic to provide the service.

According to another aspect of Invention contains one computer processing platform several Computer processors connected to an internal communication network are. At least one control node is in communication with one external communication network and with an external filing network that has an external filing address space. The at least one control node is connected to the internal network and communicates with the Most of the computer processors. The configuration logic defines and manufactures: a virtual processing network with a corresponding one Set of computer processors from the majority of processors, a virtual local communication network that is a communication among the set of computer processors allows, but those processors excludes from the majority that are not in the defined amount, and one virtual storage room with a defined correspondence to the Filing network address space.

Summary the drawing

In the drawing:

1 Fig. 10 is a system diagram illustrating an embodiment of the invention;

2A - C 14 are diagrams illustrating communication links established in accordance with an embodiment of the invention;

3A - B 14 are diagrams illustrating the networking software architecture of certain embodiments of the invention;

4A - C 14 are flowcharts illustrating driver logic in accordance with certain embodiments of the invention;

5 illustrates service clusters according to certain embodiments of the invention;

6 illustrates the filing software architecture of certain embodiments of the invention;

7 illustrates the processor-side filing logic of certain embodiments of the Er invention;

8th illustrates filing address mapping logic of certain embodiments of the invention; and

9 illustrates the cluster management logic of certain embodiments of the invention.

Full description

Preferred embodiments of the invention create a processing platform, from virtual systems to configuration commands can be used. The platform creates a big one Pool of processors that make up a subset through software instructions selected and can be configured so that they form a virtualized network of computers ("processing area network" or "processor cluster") that are used can be so that there is a given set of applications or serves a customer. The virtualized processing area network (PAN) can then be used to create custom applications such as web-based Run server applications. Virtualization can virtualize local area networks (LANs) or include virtualization of the I / O repository. As a result of that Such a platform can be created using processing resources quickly and easily over Configuration commands, e.g. B. by an administrator, via software are used instead of a physical server, a cabling network and filing connections, delivering power to each server, etc.

Overview of the Platform and their behavior

As in 1 a preferred hardware platform is shown 100 a lot of processing nodes 105a-n that have a fast connection 110a , b with a switch fabric 115a, b , are connected. In addition, the switch fabric 115a, b , with at least one control node 120a, b in communication with an external IP network 125 (or with another data communication network), as well as with a storage area network (SAN) 130 connected. A z. B. Remote management application 135 can over the IP network 125 access one or more of the control nodes to help configure the platform 100 and to help with the use of virtual PANs.

According to certain embodiments, there are approximately 24 processing nodes in a single chassis 105a-n , two control nodes 120 and two switch fabrics 115a, b , contained and connected to each other with a fixed, pre-wired mesh network of point-to-point connections (PzP connections). Every processing node 105 is a circuit board that has one or more (e.g. 4) processors 106j-l , one or more network cards (NICs) 107 and contains local memory (which is, for example, larger than 4 GB) and which contains, among other things, a certain BIOS firmware for booting and for initialization. For the processors 106 there is no local plate; instead, the entire tray, including the tray required for the page change, is provided by the SAN trays 130 treated.

Each control node 120 is a single board that contains one or more (e.g., 4) processors, local memory, and local disk storage that contains independent copies of the boot image and initial file system used to run the operating system software for the processing nodes 105 and for the control nodes 106 to boot contains. Each control node communicates with the SAN 130 over 100 megabyte / s adapter cards 128 with a second fiber optic channel that with the fiber optic channel connections 122 . 124 are connected, and via an external network interface 129 with single or multi-gigabit ethernet NICs using the gigabit ethernet connections 121 . 123 connected to the Internet (or any other external network) 125 , (Many other techniques and hardware can be used for SAN and external network connectivity.) Each control node contains a slow Ethernet port (not shown) as a dedicated management port, instead of web-based remote management about the management application 135 can be used.

The switch fabric is made up of one or more 30-port Giganet switches 115 such as switch NIC-CLAN 1000 and switch clan 5300, with the various processing and control nodes using corresponding NICs to communicate with such a fabric module. Giganet switch fabrics have the semantics of a non-broadcast multiple access network (NBMA network). All communication between the nodes takes place via a switch fabric. Each connection is as a serial connection between a NIC 107 and a port in the switch fabric 115 ausebifdet. Each connection works at 112 megabytes / second.

In some embodiments, multiple cabinets or chassis connected together so that they have larger platforms form. Moreover the configuration may differ in further embodiments; for example, redundant ones Connections, switches and control nodes can be omitted.

According to software control, the platform supports several simultaneous and independent processing area networks (PANs). Each PAN is configured via software commands so that it has a corresponding subset of the processors 106 owns, which can communicate via a virtual local area network, which emu via the PzP mesh network is gated. Each PAN is also configured to have a corresponding virtual I / O subsystem. No physical use or physical wiring is required to manufacture a PAN. According to certain preferred embodiments, software logic that executes in the processor nodes and / or in the control nodes emulates the semantics of a switched Ethernet; additional software logic that executes in the processor nodes and / or in the control nodes provides the functionality of a virtual filing subsystem that follows SCSI semantics and provides independent I / O address spaces for each PAN.

network architecture

Certain preferred embodiments allow an administrator to build virtual, emulated LANs using virtual components, interfaces, and connections. Each of the virtual LANs can go to the platform 100 be internal and private, or multiple processors can be formed into a processor cluster that is externally visible as a single IP address.

Although the physical is based lying network is a PzP mesh network, they emulate in this way generated virtual networks according to certain embodiments a mediated ethernet network. The virtual network uses IEEE MAC addresses and the processing nodes support IETF-ARP processing to IP addresses identify and assign MAC addresses. Hence answer a given processor node consistent to an ARP request, equal to, whether the ARP request from an internal to the platform or external node has come.

2A shows an exemplary network arrangement that can be modeled or emulated. A first subnet 202 is formed by the processing nodes PN ₁ , PN ₂ and PN _k , via the switch 206 can communicate with each other. A second subnet 204 is formed by the processing nodes PN _k and PN _m , via the switch 208 can communicate with each other. According to the semantics of a switched Ethernet, a node in a subnet can communicate directly with another node in the subnet; for example, PN ₁ can send a message to PN ₂ . Semantics also allow one node to communicate with a set of the other nodes; for example, PN ₁ can send a broadcast message to other nodes. Since PN _{m is} in another subnet, the processing nodes PN ₁ and PN ₂ cannot communicate directly with PN _m . In order for PN ₁ and PN ₂ to communicate with PN _m , network software from a higher layer would have to be used, which has a more comprehensive understanding of both subnets. Although not shown in the figure, a given switch may communicate with another switch or the like via an "uplink". As is clear from the description given, the need for such uplinks is different from their need if the switches are physical. More precisely, as they are virtual and modeled in software, the exchanges can scale horizontally as wide as necessary. (In contrast, physical switches have a fixed number of physical ports, and the uplinks are sometimes required to provide horizontal scalability.)

2 B shows exemplary software communication paths and software communication logic, which according to certain embodiments for modeling the subnets 202 and 204 out 2A is used. The communication paths 212 connect the processing nodes PN ₁ , PN ₂ , PN _k and PN _m , more precisely their corresponding processor-side network communication logic 210 , and also connect the processing nodes to the control nodes. (Although PN _k is shown as the only instance of the logic for reasons of clarity, PN _{k can have} several instances of the corresponding processor logic, for example one per subnet.) According to preferred embodiments, the management logic and the control node logic are responsible for the structure, the management and dismantling of the communication paths. The individual processing nodes must not set up such paths.

As explained in detail below, the processor logic and the control node logic together emulate the semantics of a switched Ethernet via these communication paths. For example, the control nodes have a virtual switching logic on the control node side 214 that emulates part (but not necessarily all) of an Ethernet switch's semantics, while processor logic includes logic that emulates part (but not necessarily all) of an Ethernet driver's semantics.

A processor node can be in a subnet via a corresponding virtual interface 212 communicate directly with another. Likewise, a processor node can communicate with the control node logic via a separate virtual interface. In certain embodiments, the underlying switch fabric and associated logic (e.g., switch fabric manager logic, not shown) provide the ability to build such virtual interfaces (VIs) over the point-to-point mesh network and to carry out their management. In addition, these virtual interfaces can be constructed in a reliable, redundant manner, whereby they are referred to here as RVIs. Since the choice between a VI versus an RVI largely depends on the level of reliability that the system desires at the expense of system resources, the terms virtual interface (VI) and verver casual virtual interface (RVI) used interchangeably at points of this description.

If the node PN ₁ with common reference to the 2A - B to communicate with the node PN ₂ , it usually does this via the virtual interface 212 _1-2 , However, preferred embodiments enable communication between PN ₁ and PN ₂ to take place via the switching emulation logic if, for example, the VI 212 _1-2 not working satisfactorily. In this case, a message about the VI 212 _{1 operator206} and about the VI 212 _{Mediation 206-2} be sent. If the PN ₁ sends a message to other nodes in the subnet 202 broadcast or collective, it does so by sending the message via the virtual interface 212 _{1 operator206} to the logic on the control node side 214 sends. The logic on the control node side 214 then emulates the broadcast or multicast functionality by cloning the message and sending it to the other relevant nodes using the relevant VIs. The same or analog VIs can be used to carry other messages that require node logic.

For example, it contains the control node side Logic, as described below, logic that the address resolution protocol (ARP) supports where The VIs are used to provide ARP responses and requests to be transmitted to the control node. Although the description above is just a VI between processor logic and control logic, many embodiments use several such Links. Moreover allows the architecture, although the figures have symmetry in the software communication paths suggests, actually asymmetrical communication. For example, the packages for communication-clustered Services as discussed below are routed through the control node. However, the back communication done directly between the nodes.

It is noted that it looks like the net 2A there is no mechanism for communication between the nodes PN ₂ and PN _m . In addition, such that the management and the establishment of the communication paths (instead of via the processing nodes) is carried out centrally by the processing nodes, and the defined subnetwork connectivity cannot be infringed by a processor.

2C shows the exemplary physical connections of certain embodiments that the subnets of FIG 2A and B realize. Each instance of the processing network logic communicates more precisely 210 via a PzP connection 216 the wiring 110 with the switch fabric 115 , The control node also has several instances of the switching logic 214 , each via a PzP connection 216 communicates with the switch fabric. As will be described below, the virtual interfaces are included 2 B the logic to convey information about these physical connections.

In order to create and configure such networks, an administrator defines the network topology of a PAN, whereby he (e.g. via a utility in the management software 135 ) MAC address assignments of the various nodes specified. The MAC address is virtual, identifies a virtual interface, and is not tied to any specific physical node. In certain embodiments, the MAC address follows the IEEE 48-bit address format, but in which the content is a "locally organized" bit (set to 1), the serial number of the control node 120 , in which the virtual interface was originally defined (further below) and contains a count of a permanent sequence counter in the control node, which is held in the NVRAM in the control node. These MACs are used (as is conventional) to place the nodes in one layer 2 to identify. For example, these MACs are included in the ARP response when responding to ARP requests (either from a node internal to the PAN or from a node in an external network).

The control node-side network logic maintains data structures that contain information that reflects the connectivity of the LAN (for example, which nodes can communicate with which other nodes). The control node logic also maps VI (or RVI) maps and assigns them to the defined MAC addresses and maps them to VIs or (RVIs) and maps them between the control nodes as well as between the control nodes and the processing nodes. In the example 2A logic arranges the VIs 212 out 2 B and assigns them. (In some embodiments, the naming of the VIs and RVIs is a consequence of the switching fabric and manager logic used for the switch fabric.) As each processor boots, the BIOS-based boot logic initializes each processor 106 of the knot 105 , including a VI 212 builds up to (or discovers) the control node logic. The processor node then receives relevant data connection information from the control node, such as the MAC address of the processor node and the MAC identities of other devices in the same data connection configuration. Each processor then registers its IP address with the control node, which then binds the IP address to the node and to an RVI (for example to the RVI to which the registration has arrived). In this way, the control node can bind IP addresses for each virtual MAC for every node in a subnet. In addition to the above, the processor node receives the RVI or VI related information for its connections to other nodes or to the control node network logic.

Thus, the various processor nodes should have their connectivity to the layer after booting and initialization 2 , the data link layer, understand. As explained below, the connectivity of the layer 3 (IP) and become more specific the assignments of the layer 3 to the shift 2 determined as a result of the address resolution protocol during normal processing of the processors.

3A shows the processor-side network logic in detail 210 , while 3B the logic of networking on the control node side 310 certain embodiments. The processor-side logic 210 contains the IP stack 305 , the virtual network driver 310 who have favourited ARP logic 350 who have favourited RCLAN Layer 315 and the redundant Giganet drivers 320a, b , The logic on the control node side 310 contains the redundant Giganet drivers 325a, b who have favourited RCLAN Layer 330 who have favourited Proxy Logic 360 for virtual clusters, the virtual LAN server 335 who have favourited ARP Server Logic 355 , the virtual LAN proxy 340 and the physical LAN drivers 345 ,

IP stack

The IP stack 305 is the communication protocol stack that is associated with that from the processing nodes 106 operating system used (e.g. Linux) is provided. The IP stack represents an interface of the layer 3 ready for applications while the operating system is in a processor 106 that communicates with the simulated Ethernet network. The IP stack provides information packets for the virtual Ethernet layer 310 in conjunction with a Layer 3 IP address as the destination for this packet. In addition to certain embodiments avoiding checksum calculations and checksum logic, the logic of the IP stack is conventional.

Virtual ethernet driver

The virtual ethernet driver 310 appears for the IP stack 305 like a "real" ethernet driver. In this regard, the virtual ethernet driver receives 310 IP packets or datagrams from the IP stack to the subsequent broadcast on the network and it receives packet information from the network for delivery to the stack as an IP packet.

The stack builds the MAC header on. The "normal" Ethernet code can be used in the stack become. The virtual Ethernet driver receives the packet in which the MAC header already set up and the correct MAC address already in the starting block is.

Essentially and as in the 4A - C is shown 405 the virtual ethernet driver 310 outgoing IP datagrams from the queue so that the packet can be sent on the network. The standard ARP logic of the IP stack is used. As explained below, the driver intercepts all ARP packets entering and leaving the system to change them in such a way that the correct information arrives in the ARP tables of each node. Normal ARP logic places the correct MAC address in the data link header of the outbound packet before the packet is queued for the Ethernet driver. The driver then examines only the data link header and destination MAC to determine how to send the packet. The driver does not directly manipulate the ARP table (apart from the occasional cancellation of the ARP entries).

The driver 310 certainly 415 whether the ARP logic 350 Has MAC address information (below) associated with the IP address in the queued packet. If the ARP logic 350 the information has, the information is used to send the packet accordingly 420 , If the ARP logic 350 the driver does not possess the information, the driver must determine this information, which information is provided in certain preferred embodiments, such as in connection with the 4B - C is discussed as a result of implementing the ARP protocol.

If the ARP logic 350 Having the MAC address information, the driver analyzes that from the ARP logic 350 information returned to determine where and how to send the package. More specifically, the driver looks at the address to determine whether the MAC address is in a valid format or in a special invalid format. For example, in one embodiment, internal nodes (ie PAN nodes internal to the platform) are signaled by a combination of the setting of the locally organized bit, the collective send bit and a further predetermined bit pattern in the first byte of the MAC address. The overarching pattern is a pattern that is very unlikely to be a valid pattern.

If the one returned by the ARP logic MAC address in a valid Format is the IP address assigned to this MAC address for one Node that is at least external to the relevant subnet and in preferred embodiments externally to the platform is. To deliver such a package, the Drive the packet a TLV header (type length value header). thereupon the logic sends the packet a previously built VI on the control nodes. Then treated the control node, as far as appropriate, the rest of the transmission.

If that's from the ARP logic 350 returned MAC address information is in a special invalid format, the invalid format signals that the IP addressed node is for an internal node, the information in the MAC address information being used to help identify the VI (or RVI), the that directly connects two processing nodes. For example, the ARP table entry can hold information that the RVI to be used 212 , e.g. B. the 212 _1-2 , identify to send the packet to another processing node. The driver prepends the packet with a TLV header. He then arranges address information and information identifying the Ethernet protocol type in the header. The logic then selects the correct VI (or RVI) on which to send the encapsulated packet. If this VI (or RVI) works satisfactorily, it will be used to move the package; if it does not work satisfactorily, the packet is sent to the control node switching logic (see below) so that it can send the switching logic to the correct node. Although the ARP table may contain information that actually specifies the RVI to be used, many other techniques can be used. For example, the information in the table can provide this information directly, e.g. For example, point to the information of interest or otherwise identify the information although it does not contain it.

For the driver sends any multicast or broadcast messages the message on a defined VI to the control node. The The control node then clones the packet and sends it accordingly to all nodes (with the exception of the sending node) and to the uplink.

If there was no ARP mapping, they would have upper layers never sent the packet to the driver. If no Data link mapping available the packet is dropped until ARP resolution is complete. If the ARP layer has completed the ARP processing, the Data Link Headers of Packets built, the ARP is kept pending and the packets then sent to the driver.

If the ARP logic has no assignment for an IP address of an IP packet from the IP stack and the driver 310 consequently, cannot determine the assigned addressing information (ie, the MAC address or the RVI-related information), the driver obtains this information by following the ARP protocol. As in the 4B - C is shown, builds 425 the driver issues an ARP request packet that contains the relevant IP address for which there is no MAC mapping in the local ARP table. It then precedes the ARP packet with a TLV header 430 , The ARP request is then sent to the control node-side network logic via a dedicated RVI - more precisely, to the virtual LAN server 335 - Posted.

As discussed in more detail below, the ARP request packet is processed by the control node 435 and broadcast to the relevant nodes 440 , For example, the control node indicates whether the requesting node is part of an IP service cluster.

The ethernet driver logic 310 receives in the relevant node 445 the ARP response and determined 450 by comparing the destination IP address to a list of locally configured IP addresses by making calls to the node's IP stack to see if it is the destination of the ARP request. If it is not the destination, it passes the package up without any change. If it is the goal, generated 460 the driver from the TLV header updates a local MAC header and updates it 465 the local ARP table and it generates an ARP response. The driver changes the information in the ARP request (mainly the source MAC) and then passes the ARP request up, usually for handling by the upper layers. It is the upper layers that form the ARP response if necessary. The response includes, among other things, the responding node's MAC address, and has a bit set in the TLV header indicating that the response is from a local node. In this regard, the node (in contrast to the ATM-ARP protocols in which ARP responses are handled centrally) responds in accordance with the IETF-ARP semantics. The response is then sent 470 ,

As explained in more detail below 473 the control node logic 335 the answer and changes it. For example, the control node can replace the MAC address of a responding internal node with information identifying the source cabinet, the processing node number, the RVI connection number, the channel, the virtual interface number and the name of the virtual LAN. If the ARP response is changed, send 475 the control node logic sends the ARP response to a correct node, ie the node that sent the ARP request, or in specific cases to the load balancer in an IP service cluster discussed below.

Finally, an encapsulated ARP response is received 480 , If the responding node is an external node, the ARP response contains the MAC address of the responding node. If the responding node is an internal node, the ARP response instead contains information that identifies the relevant RVI for communicating with the node. In any case, the local table is updated 485 ,

The pending datagram is removed from the queue 487 and selected the right RVI 493 , As discussed above, the correct RVI is selected based on whether the target node is internal or external. The packet is prefixed with a TLV header and sent 495 ,

For communications in a virtual LAN, the maximum transmission unit (MTU) is configured as 16896 bytes. Although the configured MTU 16896 Bytes is recognized by the Ethernet driver 310 when a packet is sent to an external network. By using the discovery of the path MTU, ICMP and IP batch changes, the path MTU becomes in the source node 105 changed. This mechanism is also used to trigger checksum formation on the package.

Certain embodiments of the invention support by combining the logic in the virtual LAN server 335 and in the virtual LAN drivers 310 the promiscuous mode. If a virtual LAN driver 310 from the virtual LAN server 335 receives a promiscuous mode message, the message contains information about the identity of the recipient who wants to enter the promiscuous mode. This information includes the location of the recipient (cabinet, node, etc.), the interface number of the virtual promiscuous interface 310 in the receiver (which is required for demultiplexing packets) and the name of the virtual LAN to which the receiver belongs. This information is then provided by the driver 310 used to determine how promiscuous packets are to be sent to the recipient (which RVI or other mechanism to use to send the packets). The virtual interface 310 maintains a list of promiscuous listeners on the same virtual LAN. When a sending node receives a promiscuous mode message, it updates its promiscuous list accordingly.

• – Falls das abgehende Paket rund- oder sammelgesendet wird, wird keine Promiscuous-Kopie gesendet. Die normale Rundsendeoperation sendet das Paket an den bzw. an die Promiscuous-Hörer.
• – Falls das Paket ein Punkt-zu-Punkt-Sendepaket mit einem anderen Ziel als dem Promiscuous-Hörer ist, wird das Paket geklont und an die Promiscuous-Hörer gesendet.

If a package has a virtual ethernet driver 310 is sent, this list is examined. If the list is not empty, the virtual ethernet interface does 310 the following:

• - If the outgoing packet is broadcast or bulk, no promiscuous copy will be sent. The normal broadcast operation sends the packet to the promiscuous listener.
• If the packet is a point-to-point transmission packet with a destination other than the promiscuous listener, the packet is cloned and sent to the promiscuous listener.

The TLV header contains additional information, which the destination can use to demultiplex the incoming packet and check. On Part of this information is the destination interface number of the virtual Ethernet (the target device number in the receiving node). Since this is between the actual Package target and the promiscuous target may be different this initial block cannot simply be cloned. So for everyone Header for every packet clone can be allocated to every promiscuous listener memory. If the packet header for a promiscuous packet is built, the packet type is sent, to indicate that the package is a promiscuous shipment instead of a point-to-point shipment is.

The virtual ethernet driver 310 is also responsible for handling the redundant control node connections. For example, the virtual Ethernet drivers periodically test end-to-end connectivity by sending a heartbeat TLV to each connected RVI. This allows the virtual Ethernet drivers to determine whether a node has stopped responding or whether a stopped node has started to respond again. If it is determined that an RVI or a control node 120 is inoperable, the Ethernet driver sends traffic through the persistent control node. If both control nodes are functional, the driver tries 310 load balancing the traffic between the two nodes.

Certain embodiments of the invention provide performance improvements. Because all elements of the platform 100 provide error detection and guaranteed data delivery, for example when there are changes to the IP stack 305 no checksum of the packages is calculated, which is only in the platform 100 be sent.

The RVI can also be used for communications in a PAN (or even in a platform 100 ) are configured so that the packets can be larger than the maximum size permitted by the Ethernet. Thus, the maximum packet size can be violated to improve performance, although in certain embodiments the model emulates Ethernet behavior. The actual packet size is negotiated as part of the data link layer.

A failure of a control node is detected either by a notification from the RCLAN layer or by a failure of the heartbeat TLVs. If a control node fails, the Ethernet driver sends 310 traffic only to the remaining control nodes. The ethernet driver 310 recognizes the recovery of a control node via the notification from the RCLAN layer or the resumption of the heartbeat TLVs. When a control node has been restored, the Ethernet driver sets 310 load balancing continues.

If a node detects that it with another knot not over a direct RVI (as outlined above) tries to communicate the knot over communicate with the control node, which acts as an intermediary. This failure can be caused by the lower RCLAN layer, for example a failure to receive a virtual interface acknowledgment or from over the heartbeat mechanism recorded failures, be signaled. In this case, the driver marks accordingly Bits in the TLV header to indicate that the message is from To be sent point to point, taking the packet to the control node sends so that it sends the packet (e.g. if necessary based on the IP address) the wished Node can send.

RCLAN layer

The RCLAN layer 315 is responsible for handling the redundancy, failover and load balancing logic of the redundant wiring NICs 107 , This includes capturing outages, redirecting traffic over a redundant connection in the event of outages, load balancing, and reporting the inability to deliver traffic to the virtual network drivers 310 return them. The virtual ethernet drivers 310 expect them to be notified asynchronously if there is a serious error in any RVI that renders the RVI unusable, or if any RVI is inoperative for any reason.

Under normal circumstances, the virtual network driver tries 310 perform load balancing of outgoing packets between available control nodes in each processor. This can be done by simply cycling between the available control nodes or by tracking how many bytes have been sent to each, always sending to the control node that has sent the fewest bytes.

The RCLAN creates a reliable asynchronous High bandwidth point-to-point communication (224 MB / s in each Direction) and low latency between the cores. If the Data cannot be provided the sender of the data is notified, making the greatest effort be made to deliver them. The RCLAN uses two cards Giganet clan 1000 to redundant communication paths between the cores It seamlessly replaces individual failures in the clan 1000 cards or in the Giganet switchboards. It captures lost data and data errors and sends the data if necessary again. As long as one of the connections is partially working, d. H. the error rate does not exceed 5%, communication is not interrupted. Include clients of the RCLAN the RPC mechanism, the remote SCSI mechanism and the remote Ethernet. Moreover the RCLAN provides a simple form of flow control. Thereby, that enables is that multiple simultaneous requests for each device by the Processor nodes are sent to the control node so that they as quickly as possible can be forwarded to the device, or alternatively, so that them rather than completing all of the requirements in the processor node queued as close to the device as possible for the Degree can be classified can low latency and high simultaneous access are made possible.

The RCLAN layer 330 on the control node side works analogously to the above.

Giganet driver

The Giganet driver logic 320 is the logic that is responsible for it, either in a processor 106 or in the control node 120 an interface for the Giganet NIC 107 provide. In short, the Giganet driver logic establishes VI connections to which VI IDs are assigned so that the higher layers, e.g. B. the RCLAN 315 and the ethernet driver 310 , just need to understand the semantics of the VIs.

The Giganet driver logic 320 is responsible for allocating memory for the buffers and queues for the VIs and the NIC in each node 107 to be prepared so that it knows the connection and its memory allocation. Certain embodiments use VI connections provided by the Giganet driver. The Giganet NIC driver code creates a virtual interface pair (ie VI) and assigns it a corresponding Id of the virtual interface.

Each VI is a bi-directional connection that is established between one gigantic port and another, or more precisely, between the memory buffers and queues in one node to the buffers and queues in the other. As noted above, the mapping of ports and memory is handled by the NIC drivers. The data is sent by placing it in a buffer that the NIC knows and by triggering the action by writing to a specific memory-addressed register. On the receiving side, the data appears in a buffer, while the completion status appears in a queue. If the sending and receiving programs can generate and consume messages in the connection buffers, the data never need to be copied. If the operating system maps the connection buffers and control registers to the application address space, the transmission can even take place directly from application program to application program. Each Giganet port can support 1024 simultaneous VI connections and keep them separate with hardware protection so that the operating system and various applications can securely share a single port. According to one embodiment of the invention, any port can be used to any other port 14 VI connections are established simultaneously.

In preferred embodiments, the NIC drivers set up VI connections in redundant pairs, one connection of the pair via one of the two switch fabrics 115a , b and the other goes through the other operator. In addition, in preferred embodiments, the data is alternately sent in the two legs of the pair, which balances the load on the switches. Alternatively, the redundant pairs of type ei Failover can be used.

All connection pairs established by the node exist as long as the operating system remains operational. Establishing a connection pair to simulate an Ethernet connection should be analog and just as stable be like physically plugging a cable between network cards. If the defined configuration of a node changes while its Operating system running, become applicable redundant virtual interface connection pairs at the time of the change assembled or dismantled.

The Giganet driver logic 325 on the control node side works analogously to the above.

Virtual LAN server

• 1. führt das Management der Mitgliedschaft in einem entsprechenden virtuellen LAN aus;
• 2. stellt die RVI-Zuordnung und das RVI-Management bereit;
• 3. ARP-Verarbeitung und IP-Zuordnung zu den RVI;
• 4. stellt Rund- und Sammelsendedienste bereit;
• 5. erleichtert die Überbrückung und Leitung zu anderen Domänen; und
• 6. führt das Management von Dienst-Clustern aus.

The logic 335 of the virtual LAN server facilitates the emulation of an Ethernet network via the underlying NBMA network. The logic of the virtual LAN server

• 1. performs membership management in a corresponding virtual LAN;
• 2. Provides RVI mapping and RVI management;
• 3. ARP processing and IP assignment to the RVI;
• 4. Provides broadcast and collective broadcasting services;
• 5. Facilitates bridging and routing to other domains; and
• 6. performs the management of service clusters.

1. Management membership in virtual LANs

The administrators configure the virtual LANs using the management application 135 , The assignment and configuration of IP addresses in virtual LANs can be done in the same way as in a "normal" subnet. The choice of the previously used IP addresses depends on the external visibility of the nodes in a virtual LAN. If the virtual LAN is not globally visible (either from outside the platform 100 or invisible from the Internet), private IP addresses should be used. Otherwise, the IP addresses must be configured from the range provided by the Internet service provider (ISP) that provides Internet connectivity. In general, the IP address assignment of virtual LANs must be treated in exactly the same way as the IP address assignment of normal LANs. Configuration files on the local disks of the control node 120 are stored, define the IP addresses in a virtual LAN. For a virtual network interface, an IP alias only creates another IP-RVI assignment in the logic 335 of the virtual LAN server. If required, each processor can configure several virtual interfaces. The main restrictions on the creation and configuration of virtual network interfaces are IP address assignment and configuration.

Each virtual LAN has a corresponding instance of the server logic 335 that are both in the two control nodes 120 as well as in a number of processor nodes 105 is performed. The topology is defined by the administrator.

Any virtual LAN server 335 is configured to manage exactly one broadcast domain, in the given broadcast domain of the layer 2 any number of subnets of the layer 3 (IP) can be present. The servers 335 are configured and generated in response to administrator commands to create virtual LANs.

If a processor 106 boots and configures its virtual networks, it connects to the virtual LAN server via a special management RVI 335 , The processors then receive their data link configuration information, such as the virtual MAC addresses assigned to them, the virtual LAN membership information, and the like. The virtual LAN server 335 determines and confirms that the processor that tries to connect to it is a real member of the virtual LAN, that of this server 335 serves. If the processor is not a member of the virtual LAN, the connection to the server is rejected. If he is a member, the virtual network driver registers 310 his IP address at the virtual LAN server. (The IP address is through the IP stack 305 provided if the driver 310 The virtual LAN server then binds this IP address to an RVI in which the registration has arrived. This enables the virtual LAN server to find the processor assigned to a specific IP address. In addition, the assignment of IP addresses to a processor via the virtual LAN management interface 135 be carried out. The latter procedure is required to properly configure the cluster IP addresses or the special-treatment IP addresses discussed below.

2. RVI assignment and management

Use as outlined above certain embodiments RVIs that connect nodes in the data link layer and control connections form. Some of these connections are considered part of booting and generated and assigned the initialization of the control nodes. The Data link connections are made from those described above establish used. The tax connections are used to manage, Exchange configuration and health information.

Some RVI connections are between nodes for point-to-point traffic, e.g. B. 212 _1-2 , Other RVI connections are for the logic of the virtual LAN server 335 so that the server accepts the claims, e.g. B. ARP traffic, broadcasts, etc. can handle. The virtual LAN server creates and removes the RVI 335 RVIs on calls to a Giganet attendant manager 360 (provided with the switch fabric and with the Giganet NICs). The operator can be in the control node 120 and works with the Giganet drivers that generate the RVIs.

With regard to the processor connections, the virtual LAN server generates and assigns virtual MAC addresses for the nodes as described above while the nodes are with the virtual LAN server 335 Sign in. In connection with this, the logic of the virtual LAN server maintains the data structures that reflect the topology and the MAC assignments for the various nodes. The logic of the virtual LAN server then generates the corresponding RVIs for the point-to-point transmission paths between nodes. While the nodes are booting, these RVIs are subsequently assigned and made known to the nodes.

In addition, the RVIs are assigned IP addresses while the virtual LAN server handles ARP traffic. If a node is removed from the topology the RVI connections are cleared down.

If a knot 106 is rebooted at one end of an established RVI connection, the two operating systems at the respective end of the connection and the RVI management logic re-establish the connection. The software that uses the connection in the processing node that remains operational does not notice that anything has happened to the connection itself. Whether the software notices or notices that the software has been rebooted at the other end depends on what it uses the connection for and the extent to which the newly booted end can restore its state from permanent storage. For example, any software that communicates over the Transmission Control Protocol (TCP) will notice that rebooting will close all TCP sessions. On the other hand, network file system access (NFS access) is stateless and is not affected by a reboot if it takes place within a permissible period of time.

If at any time a node cannot send a packet on a direct RVI, it can always try to send the packet through the virtual LAN server 335 to send to a destination. Because the virtual LAN server 335 via the control connections with all interfaces with virtual Ethernet driver 310 connected in the virtual LAN, the virtual LAN server 335 also serve as the last resort of the packet switching mechanism.

Regarding the connections with the virtual LAN server 335 Certain embodiments use virtual Ethernet drivers 310 who algorithmically determine the RVI they should use to connect to their assigned virtual LAN server 335 connect to. Depending on the embodiment, the algorithm may need to look at identification information such as the cabinet number to identify the RVI.

3. ARP processing and IP implementation to RVIs

As explained above, the virtual Ethernet drivers support 310 certain embodiments the ARP. In these embodiments, ARP processing is advantageously used to create, at the nodes, translations between IP addresses and RVIs that can be used to carry point-to-point traffic, including IP packets, between the nodes.

To do this, send the virtual Ethernet drivers 310 ARP packet requests and responses to the virtual LAN server via a dedicated RVI 335 , The virtual LAN server 335 and more precisely the ARP server logic 355 treats the packets by adding information to the packet header. As discussed above, this information enables the source and destination to be identified and the RVI to be used between the nodes.

The ARP server logic 355 receives the ARP requests, processes the TLV header and broadcasts the request, if appropriate, to all relevant nodes in the internet platform and in the external network. The server logic 355 determines, among others, who should receive the ARP response resulting from the request. For example, if the source is a clustered IP address, the response should be sent to the cluster load balancer and not necessarily to the source of the ARP request. The server logic 355 indicates this by including information in the TLV header of the ARP request so that the destination of the ARP responds accordingly. The server 335 processes the ARP packet by including more information in the attached header and broadcasting the packet to the nodes in the relevant domain. For example, the modified header may contain information that identifies the target cabinet, processing nodes, RVI connection number, channel, virtual interface number, and virtual LAN name (some of which only the server 335 are known).

The ARP responses are from the server logic 355 received, which then converts the MAC information in the response into corresponding RVI-related information. The RVI-related information is placed in the destination MAC entry of the response and sent to the appropriate source node (which may be the sender of the request, for example, in some cases such as at a clustered IP address but can be another node).

4. Round and Multicast services

As explained above, broadcasts are handled by receiving the packet on a dedicated RVI. Thereupon the packet from the server 335 cloned and to all virtual interfaces 310 broadcast from point to point in the relevant broadcast domain.

The same access can be used for the bulk shipment be used. All bulk mailing packages are sent to the virtual LAN server mirrored. According to some alternative embodiments discussed the virtual LAN server the collective broadcast as well as the broadcast, being used to filter out unwanted packets at each node relies on IP filtering.

If an application wants to send or receive multicast addresses, it must first join a multicast group. When a process in a processor makes a multicast join, the virtual network driver sends 310 the processor sends an access request to the virtual LAN server via an RVI provided for this purpose 335 , The virtual LAN server then configures a specific multicast MAC address in the interface and if necessary informs the LAN proxy as discussed below 340 , The proxy 340 must track usage counts in specific multicast groups so that a multicast address is removed only when no processor belongs to that multicast group.

5. Bridging and Line to other domains

From a system perspective 100 can the external network 125 work in one of two modes: filtered or unfiltered. In the filtered mode, a single MAC address is used for the entire system for all outgoing packets. This hides the virtual MAC addresses of a processing node 107 behind the virtual LAN proxy 340 and leaves the system on the network 125 appear as a single node (or as multiple nodes behind a bridge or proxy). Because this is not the unique data link information for each internal node 107 exposed, some more unique identifiers are required to properly deliver incoming packets. Since the MAC address only identifies the system, when operating in the filter mode, the destination IP address of each incoming packet is used to uniquely identify the intended recipient. The virtual MACs of a node are in the unfiltered mode 107 Visible outside the system so that they can be used to route incoming traffic. This means that the filtered operating mode orders the switching of the shift 3 , while the unfiltered mode of operation is the mediation of the shift 2 allows. The filtered operating mode requires that a component (in this case the virtual LAN proxy 340 ) in all outgoing packets the replacement of the virtual MAC address of the node by the MAC address of the external network 125 performs.

Some embodiments support the ability that a virtual LAN can be connected to external networks. Accordingly, the virtual LAN must handle IP addresses that are not configured locally. To deal with this, shoot one embodiment a limitation that each virtual LAN connected in this way is on a external broadcast domain limited is. IP addresses and subnet assignments for The internal nodes of the virtual LAN must match the external ones domain stand.

The virtual LAN server 335 serves the external connection by effectively acting as a data link bridge by placing packets between the external ethernet driver 345 and shifts the internal processors and does no IP processing. However, unlike a data link layer bridge, the server cannot always rely on different layer two addresses from the external network to internal nodes, with the connection instead of layer information 3 (IP) can be used to make the bridging decisions. To do this, the software extracts IP address information from the incoming packets from the external connection and uses this information to find the right node 106 to identify so that they can move the package to the node.

A virtual LAN server 335 With an attached external broadcast domain, packets must intercept and process packets from and to the external domain so that external nodes have a consistent view of the subnet or subnets in the broadcast domain.

If the virtual LAN server 335 with an attached external broadcast domain receiving an ARP request from an external node, it forwards the request to all internal nodes. The correct node then assembles the response and sends it via the virtual LAN server 335 back to the requester. The virtual LAN server works with the virtual LAN proxy 340 together so that the proxy can handle any required MAC address translation on the outgoing requests. All ARP responses and ARP announcements from external sources are forwarded directly to the target node.

The virtual Ethernet interfaces 310 send all point-to-point transmission packets with an external destination to the virtual LAN server via the control connection RVI 335 , (The external goals can can be recognized by the driver through the MAC address format.) The virtual LAN server then moves the packet accordingly into the external network 125 ,

If the virtual LAN server 335 receives a broadcast or multicast packet from an internal node, it forwards the packet except to all internal members of the virtual LAN to the external network. If the virtual LAN server 335 receives a broadcast or multicast packet from an external source, it forwards the packet to all connected internal nodes.

According to certain embodiments connecting virtual LANs using IP routers or firewalls using an analog mechanism such as the one used when connecting physical LANs. In A processor is configured on both LANs, with routing (and possibly the IP masquerade) in the Linux core released on this processor have to be. The normal IP subnet and line semantics itself for two Always keep nodes that are on the same platform.

A processor could act as a router between two external subnets, between an external and an internal Subnet and between two internal subnets can be configured. When an internal node sends a packet over a router, there is none due to the point-to-point topology of the internal network Problems. The transmitter sends without the intervention of the virtual LAN server directly to the router (i.e. to the processor configured in this way with the control logic) (i.e., typical processor-processor communication discussed above).

When an external node sends a packet to an internal router and the external network 125 running in the filtered mode, the destination MAC address of the incoming packet is that of the platform 100 , The MAC address can therefore not be used to uniquely identify the destination node of the packet. For a packet whose destination is an internal node in the virtual LAN, the destination IP address in the IP header is used to route the packet to the correct destination node. Since the routers are not the final destinations, the destination IP address in the IP header is that of the final destination instead of the next hop (which is the internal router). Thus, there is nothing in the incoming packet that can be used to route it to the correct internal node. To deal with this situation, one embodiment imposes a limit of no more than one router exposed to an external network in a virtual LAN. This router is the virtual LAN server 335 registered as a default destination so that incoming packets without a valid destination are routed to that default node.

When an external node sends a packet to an internal router and the external network 125 running in the unfiltered mode, the destination MAC address of the incoming packet is the virtual MAC address of the internal destination node. The LAN server 335 then uses this virtual MAC to send the packet directly to the internal destination node. In this case, since the MAC address of the incoming packet uniquely identifies the destination node, any number of internal nodes can act as a router.

If one configuration has several A router in a subnet may require a router as the exposed one Router can be picked out. This router could in turn, if necessary route to the other routers.

According to certain embodiments thereby creating router redundancy that a router becomes one clustered service is done and a load balancing or one Failover on a stateless basis (i.e. each IP packet instead of per TCP connection) becomes.

Certain embodiments of the invention support the functionality the promiscuous mode by providing mediation semantics deploy in the given port as a promiscuous port can be determined so that the entire going through the mediation Traffic to the promiscuous port is repeated. The knots that hear in promiscuous mode may be organizational assigned to the virtual LAN server.

If a virtual ethernet interface 310 enters the promiscuous receive mode, it sends a message to the virtual LAN server through the management RVI 335 , This message contains all information about the virtual Ethernet interface 310 entering the promiscuous mode. When the virtual LAN server receives a promiscuous mode message from a node, it checks its configuration information to determine whether the node is allowed to hear promisk. If not, the virtual LAN server drops the promiscuous mode message without further processing. If the node is allowed to enter promiscuous mode, the virtual LAN server broadcasts the promiscuous mode message to the other nodes in the virtual LAN. The virtual LAN server also marks the node as promisk so that it can forward copies of the incoming external packets to it. When a promiscuous listener detects any change in its RVI configuration, it sends a promiscuous mode message to the virtual LAN that updates the status of all other nodes in the relevant broadcast domain. This updates any nodes that enter or leave a virtual LAN. If a virtual ethernet interface 310 exits promiscuous mode, it sends a message to the virtual LAN server informing it that the Interface leaves the promiscuous mode. The virtual LAN server then sends this message to all other nodes in the virtual LAN. The promiscuous settings allow an external connection to be put into promiscuous mode if any internal virtual interface is a promiscuous handset. This makes the traffic external to the platform (but in the same virtual LAN) available to the promiscuous listener.

6. Management the service cluster

A service cluster is a set of services that are available at one or more IP addresses (or host names). Examples of these services are HTTP, FTP, telnet, NFS, etc. An IP address and a pair of port numbers represent a specific type of service (although not a service instance) that serves the clients, including the clients in the external network 125 is offered by the cluster.

5 shows how certain embodiments use a virtual cluster 405 of services to the Internet or to another external network 125 Represent as a single virtual host via an IP address of the cluster. All services of the cluster 505 are addressed via a single IP address, but via different ports at this IP address. In the example 5 Service B is a load-balanced service.

Based on 3B become virtual clusters by incorporating virtual cluster proxy logic (VCP logic) 360 supported with the virtual LAN server 335 interacts. In short, is the VCP 360 responsible for handling the distribution of incoming connections, port filters and real server connections for each configured virtual IP address. There is one VPC for each configured clustered IP address.

When a packet arrives at the IP address of the virtual cluster, the virtual LAN proxy logic sends 340 the package for processing to the VCP 360 , The VCP then decides where to send the packet based on the packet content, its internal connection cache, any load balancing algorithms that apply to incoming traffic, and the availability of the configured services. The VCP forwards incoming packets based on the destination IP address as well as the TCP or UDP port number. It also only distributes packets that are intended for port numbers known to the VCP (or for existing TCP connections). It is the configuration of these ports and the conversion of the port number to one or more processors that creates the virtual cluster and makes specific service instances available in the cluster. If multiple instances of the same service are configured by multiple application processors, the VCP can load balance the service instances.

The VCP 360 maintains a cache of all active connections that exist at the cluster's IP address. Any load balancing decisions that are made are made only when a new connection is established between a client and a service. When the connection is established, the VCP uses the source and destination information in the header of the incoming packet to ensure that all packets in a TCP stream go to the same processor 106 which is configured to provide the service. In the absence of the ability to determine a client session (e.g., HTTP sessions), the actual connection / load balancing translation cache routes packets based on the client address so that subsequent connections from the same client to the same Processor go (which makes a client session permanent or "sticky"). Since only certain types of services require session persistence, session persistence should be selectable based on the session port number.

Responses to ARP requests and routing ARP requests are handled by the VCP. When a processor sends any ARP packet, it sends it through the virtual Ethernet driver 310 out. The packet is then sent to the virtual LAN server for normal ARP processing 335 Posted. The virtual LAN server broadcasts the packet as usual, but ensures that it is not broadcast to every member of the cluster (not just the sender). In addition, it arranges in the packet header TLV information which indicates to the ARP target that the ARP source can only be reached via the virtual LAN server and more precisely via the load balancer. The ARP target, whether internal or external, processes the ARP request normally and sends back a response via the virtual LAN server. Since the source of the ARP was an IP address of the cluster, the virtual LAN server cannot determine which processor sent the original request. Thus, the virtual LAN server sends the response to each cluster member so that they can handle it correctly. When an ARP packet is sent from a source with an IP address of the cluster as the destination, the virtual LAN server sends the request to each cluster member. Each cluster member receives the ARP request and processes it normally. They then compile an ARP response and send it back to the source via the virtual LAN server. If the virtual LAN server receives any ARP response from a cluster member, it drops that response while the virtual LAN server assembles an ARP response and sends it to the ARP source. The virtual LAN server thus replies to all ARPs of the cluster's IP address. The ARP-Ant word contains the information required for the ARP address to send all packets for the cluster's IP address to the VCP. For external ARP sources, this is simply an ARP response with the external MAC address as the source hardware address. For internal ARP sources, this is the information required to tell the source to send packets for the cluster's IP address along the virtual LAN management RVI rather than through a directly connected RVI. Any redundant ARP packets that are received are forwarded to all cluster members. Any redundant ARP packets sent by a cluster member are sent normally.

Virtual LAN proxy

The virtual LAN proxy 340 performs the basic coordination of the physical network resources among all processors, the virtual interfaces to the external physical network 125 have. It bridges the virtual LAN server 335 to the external network 125 , If the external network 125 runs in the filtered mode, the virtual LAN proxy sets 340 the internal MAC addresses of each node into the single external MAC assigned to the system 100 is assigned. If the external network 125 In the unfiltered mode, no such MAC translation is required. The virtual LAN proxy also works 340 inserting and removing IEEE 802.1Q virtual LAN ID tagging information and demultiplexing packets based on their VLAN IDs. It also serializes access to the physical Ethernet interface 129 and coordinates the assignment and removal of the MAC addresses such as the multicast addresses in the physical network.

If the external network 125 in the filtered mode and the virtual LAN proxy is running 340 outgoing packets (ARP or otherwise) from a virtual LAN server 335 receives, it replaces the MAC address in the internal format with the MAC address of the physical Ethernet device 129 as the source MAC address. If the external network 125 No such replacement is required in the unfiltered mode.

If the virtual LAN proxy 340 receives the incoming ARP packets, it moves the packet to the virtual LAN server 335 who handles the package and forwards it to the right goal or goals. If the ARP packet is a broadcast packet, the packet is forwarded to all internal nodes in the virtual LAN. If the packet is a point-to-point send packet, it is only sent to the destination node. The destination node is by the IP address in the ARP packet when the external network 125 runs in the filtered mode, or determined by the MAC address in the Ethernet header of the ARP packet (not the MAC address is the ARP packet).

physical LAN driver

According to certain embodiments, the connection to the external network is made 125 via Gigabit or 100 / 10baseT Ethernet connections that are connected to the control node. The physical LAN drivers 345 are responsible for forming an interface with such connections. Packets sent into the interface are also normally queued to the device, which includes placing the packets in socket buffers. The queue used to queue the packets is that used by the protocol stack to queue packets for the sending routine of the device. For incoming packets, the socket buffers containing the packets are passed around, and the packet data is never copied (although it may be cloned for multicast operations if necessary). According to these embodiments, general Linux network device drivers can be used in the control node without modification. This allows new devices to be added to the platform without requiring additional device drivers to work.

The physical network interface 345 is only available with the virtual LAN proxy 340 in communication. This prevents the control node from using the external connection in any way that would interfere with the operation of the virtual LANs and improves the security and isolation of the user data, ie an administrator cannot "sniff" in any user packages.

load balancing and failover

According to some embodiments, the redundant connections to the external network 125 alternatively used to load balance the packet transfers between two redundant interfaces to the external network 125 perform. Other embodiments perform load balancing by configuring each virtual network interface in alternate control nodes so that the virtual interfaces are evenly distributed between the two control nodes. Another embodiment sends over one control node and receives over another.

In the filtered mode, there is an externally visible MAC address to which external nodes send the packets for a number of virtual network interfaces. If this adapter becomes inoperable, not only must the virtual network interfaces establish a failover to the other control node, but the MAC address must also establish a failover that the external nodes can continue to send packets to the MAC address that are already in the ARP caches. In accordance with one embodiment of the invention, a single MAC address is manipulated, the MAC address not having to be converted again during the recovery if a failed control node is recovered.

According to a further embodiment the invention a load balancing is carried out in that the shipment is permitted at both control nodes during the Reception only over one is admitted. The failover case is that both are over the send and receive the same control node. The recovery case is sending over the recovered control node as this is no MAC manipulation requires.

The control node that performs the reception has IP information for filtering as well as multicast address information for the multicast MAC configuration. This information is required for processing incoming Packets and should be fail-safe if the recipient Control node fails. If the sending control node fails, the virtual ones need Network drivers only with sending outgoing packets only to begin at the receiving control node. Apart from that Detection that the sending control node has failed is none special failover processing required. If the failed control nodes can be recovered without the virtual network drivers additional special recovery processing for outgoing packets send the recovered control nodes. If the receiving one Control node fails, the sending control node must assume the role of the receiving interface accept. For this, he must have all MAC addresses on his physical Configure the interface so that it enables packet reception. Alternatively, you could both control nodes have the same MAC address at their interfaces configured while receptions by the device driver on the ethernet device physically could be locked until a control node is ready to receive the packets. thereupon would Failover simple receipts enable on the device.

Because the interfaces with multicast MAC addresses need to be configured if any processor has joined a multicast group, the Collective transmission information shared among the control nodes be so that the failover is transparent to the processor he follows. Because the virtual network drivers the multicast membership have to track anyway is this information for the LAN proxy if necessary via the virtual LAN server is always available. So one leads Failover to a group membership is queried by the virtual network drivers to the local multicast group membership tables rebuild. These operations have a low level of organization and require except during failover and recovery no special processing and no special repetition of the data between the control nodes. If the reception has been secured against failure and the failed control node is recovered, only shipments to the recovered control node postponed. Thus, the recovery algorithm includes virtual network interfaces, the programs always on the recovered Move control nodes and receive processing there too leave where it is.

Virtual service clusters can also load balance and use failover.

More cabinet platforms

Some embodiments allow cabinets to be connected together to form larger platforms. Each cabinet has at least one control node that is used for connections between cabinets. Each control node contains a virtual LAN server 335 to treat local connections and local traffic. One of the servers, such as the one located at the control node with the external connection for the virtual LAN, is configured as a master. The other virtual LAN server acts as the proxy server or slaves, so that the local processors of these cabinets can participate. The master maintains the overall state and control of the virtual LAN, while the proxies forward packets between the processors and masters.

Any virtual LAN server proxy maintains one RVI to every virtual master LAN server. Each local processor connects to the virtual LAN proxy server as if he would be a master. If a processor connects an IP and a MAC address and registers, reports the proxy sends this IP and MAC address to the master. This causes the master binds the addresses to the RVI from the proxy. Thus the master contains RVI bindings for all internal nodes while the proxies bind only to nodes in the same closet contain.

If a processor sends any packet to its virtual LAN server anywhere in a multi-cabinet virtual LAN, the packet is forwarded to the master for processing. The master then performs normal processing on the packet. If necessary, the master forwards packets for collective and broadcast transmission to the proxies. In addition, the master routes point-to-point transmission packets to the proxies based on the destination IP address of the point-to-point transmission packet and the registered IP addresses. It it is noted that a proxy connection in the master looks very much like a node with many configured IP addresses.

Networking management logic

During times when an operating system is not running in a processing node, such as during booting or during core debugging, the serial console traffic and boot image requests of the node are handled by the switch driver code located in the core debugging software or located in the BIOS of the processing node, routed to the management software that runs in a control node (not shown). From there you can access the console traffic either from the fast external network 125 or be accessed again via the control node's management ports. The boot image requests can either come from the local disks of the control node or from partitions that are outside in the external SAN 130 are fulfilled. The control node is preferably 120 it boots and runs normally before anything can happen in a processing node. The control node itself is booted or tested from its management ports.

Some customers may want to limit controller booting and debugging to local access only while plugging their management ports into a local computer when needed. Others may choose to allow remote booting and debugging by creating a management-safe, secure, isolated network segment into which to plug their management ports. When a controller is booted and running normally, all other management functions for it and for the rest of the platform can be accessed from the fast external network 125 and, if permitted by the administrator, can be accessed from the management ports.

Serial console traffic to and from each processing node 105 is by an operating system core driver over the switch fabric 115 sent to management software in a control node 120 running. From there, either from the normal, fast external network 125 or the console traffic of any node is accessed through one of the control node's management ports.

storage architecture

Certain embodiments follow a SCSI filing model. Each virtual PAN has its own virtualized I / O space and issues SCSI commands and SCSI status within this space. If necessary, the logic in the control node translates or converts the addresses and commands from a PAN and sends them to the SAN accordingly 130 that operates the commands. From the SAN perspective, the client is the platform 100 while the actual PANs that issued the commands are hidden and anonymous. Since the SAN address space is virtualized, the device numbering of a PAN that is in the platform 100 works with a device number 1 start while a second PAN also has a device number 1 can own. However, each of the device numbers corresponds 1 another, distinct section of the SAN repository.

According to certain embodiments an administrator set up a virtual repository. Each of the PANs has its own independent View of the mass storage. As explained below, a First PAN assigns a given device / LUN address own a first location in the SAN while a second PAN the same given device / LUN mapping to a second, different location in the SAN. Any processor sets a device / LUN address into a main and a device number um, for example to identify a disk and a partition. Although the major and minor numbers of the PAN and by the processors in a PAN as a physical address are actually perceived by the platform as a virtual address to the bulk storage provided by the SAN treated. This means, the major and minor numbers of each processor are implemented in corresponding SAN locations.

6 shows the software components used to implement the filing architecture of certain embodiments. A configuration component 605 that are typically in a control node 120 is in communication with the external SAN 130 , A management interface component 610 provides an interface to the configuration component 605 ready and in communication with the IP network 125 and thus with the remote management logic 135 (please refer 1 ). Any processor 106 in the system 100 contains an instance of the processor-side storage logic 620 , Any such instance 620 communicates via 2 RVI connections 625 with a corresponding instance of the storage logic on the control node side 615 ,

In short, are the configuration component 605 and the interface 610 responsible for discovering those sections of the SAN repository that those of the platform 100 are assigned, and that enables an administrator to section specific PANs or processors 106 assigns sub. In addition, the filing configuration logic 605 responsible for the transfer of the SAN storage assignments to the logic on the control node side 615 , The processor-side filing logic 620 is responsible for the transmission of the filing request processor changes via internal wiring 110 and the filing fabric 115 via RVIs provided for this purpose 625 to the logic on the control node side 615 , According to certain implementations, the requirements include virtual filing addresses and SCSI commands. The control node logic is responsible for receiving and handling these commands by identifying the appropriate actual address for the SAN and limiting the commands and protocol to that for the SAN, which includes, for example, an optical fiber channel (i.e. Gigabit Ethernet with iSCSI is another exemplary connectivity), suitable form implements.

configuration component

The configuration component 605 determines what elements in the SAN 130 for every single processor 106 are visible. It provides a mapping function that translates the device numbers (e.g., SCSI target and LUN) that the processor uses into the device numbers that are used for the control nodes via their attached SCSI and fiber channel I / O interfaces 128 are visible. It also provides an access control feature that prevents processors from accessing external storage devices attached to the control nodes but not included in the processor configuration. The model presented to the processor (and the system administrator and applications / users on this processor) makes it appear as if each processor had its own mass storage devices connected to the interfaces on the processor.

Among other things, this functionality enables the software in a processor 106 easily moved to another processor. For example, in certain embodiments, the control node can change software PAN configurations (without any physical rewiring) to allow a new processor to access the required devices. Thus, a new processor can be made to inherit another's filing personality.

According to certain embodiments the control nodes appear as hosts in the SANs, while alternate ones embodiments enable, that the processors act as such.

As outlined above, the configuration logic discovers that of the platform 100 (for example during the booting of the platform) assigned SAN storage, this pool being subsequently assigned by an administrator. If the discovery is activated later, the control node that performs the discovery operation compares the new view with the previous one. A new available storage is added to the storage pool, which can be assigned by an administrator. Disappearing partitions that were not assigned are removed from the available storage pool that can be assigned to the PANs. Disappearing partitions that were assigned trigger error messages.

Management interface component

The configuration component 605 Allows management software to access and update information that the device implementation between for the control nodes 120 visible devices and for the individual processors 106 visible virtual devices. It also enables you to access tax information. The assignments can be made by the processing node in conjunction with an identifier of the simulated SCSI disks e.g. B. according to the name of the simulated controller, the simulated cable, the simulated unit or the simulated logic unit number (LUN).

• – Gesamtzahl der ausgeführten E/A-Operationen
• – Gesamtzahl der übertragenen Bytes
• – Gesamtzahl der ausgeführten Leseoperationen
• – Gesamtzahl der ausgeführten Schreiboperationen
• – Gesamtmenge der Zeit, in der die E/A im Gang war.

According to certain embodiments, the interface component operates 610 with the configuration component to collect and monitor information and statistics, such as:

• - Total number of I / O operations performed
• - Total number of bytes transferred
• - Total number of read operations performed
• - Total number of write operations performed
• - Total amount of time the I / O was in progress.

processor side Holder logic

The processor-side logic 620 of the protocol is implemented as a host adapter module that emulates a SCSI subsystem by being in the processors 106 provides a hardware-related virtual interface to the operating system. The processors 106 use this virtual interface to send SCSI I / O commands for processing to the control nodes 120 to send.

According to embodiments, the redundant control node 120 each processing node contains 105 an instance of logic 620 per control node 120 , According to certain embodiments, the processors refer to storage using physical device numbering instead of logical. That is, the address is specified as a device name that includes the LUN, the SCSI target, the channel, the host adapter, and the control node 120 (e.g. the knot 120a or 120b ) identified. As in 8th an embodiment converts the target (T) and the LUN (L) into a host adapter (H), into a channel (C), into a converted target (mT) and into a converted LUN (mL).

7 shows an exemplary architecture for the processor-side logic 720 , The logic 720 contains a device type specific driver (e.g. a disk driver) 705 , a SCSI I / O driver 710 middle tier and wrapper and wiring logic 715 ,

The device type specific driver 705 is a traditional driver that comes with the operating system and is associated with specific device types.

The SCSI I / O driver 710 The middle level is a conventional middle level driver by the device type specific driver 705 is called when the driver 705 determines that the device is a SCSI device.

The wrapper and wiring logic 715 is through the SCSI I / O driver 710 called the middle level. This logic provides the interface of the SCSI subsystem and thus emulates the SCSI subsystem. In certain embodiments that use the Giganet fabric, the logic is 715 responsible for wrapping the SCSI commands when needed, and for interacting with the Giganet and RCLAN interfaces, which causes the NIC to send the packets to the control nodes via the dedicated RVIs described above to the control nodes. The header information for the Giganet package is changed to indicate that this is a filing package that contains other information that is described below in context. Although this is in 7 not shown, the wrapper logic can 715 use the RCLAN layer to make redundant wiring 110 and fabrics 115 to support and use.

For embodiments that use the Giganet fabric 115 use the RVIs of the connection 725 assigned virtual interface numbers (VI numbers) from the range of 1024 available VIs. The mediation is for the two endpoints to communicate 115 with a bidirectional path between the pair (switching node of the control node, VI number of the control node), (switching port of the processor node 105 , VI number of the processor node).

For any message type that is sent in one direction used a separate RVI. So in every RVI is for a message always has a receive buffer pending from the other side of the Protocol can be sent. Because in every direction in every RVI only one message type is sent, the ones posted on each of the RVI channels Receive buffer as well suitable for the maximum message length measured using the protocol for this message type. According to others embodiments be all sorts Message types multiplexed to a single RVI instead of 2 VIs to use. The protocol and message format require not exactly using 2 RVIs while the news itself in their early blocks Have message type information so that it is demultiplexed could.

One of the two channels will for exchanging SCSI commands (CMD) and status messages (STAT messages) used. The other channel is used to buffer (BUF messages) and exchange transmission messages (TRAN messages). Besides, will this channel for handling data payloads of the SCSI commands used.

CMD messages contain control information, the SCSI command to be executed, and the virtual addresses and sizes of the I / O buffers in the node 105 , STAT messages contain control information and a completion status code that reflects any errors that may have occurred while processing the SCSI command. BUF messages contain control information and the virtual addresses and sizes of the I / O buffers in the control node 120 , TRAN messages contain control information and are used to successfully transfer data from the node 105 to the control node 120 to confirm.

The processor-side wrapper logic 715 examines the SCSI command to be sent to determine whether the command requires data to be transferred, and if so, in which direction. Depending on the analysis, the wrapper logic sets 715 accordingly flag information in the message header. The section that describes the control node logic describes how the flag information is used.

According to certain embodiments of the invention, the connection 725 between the processor-side filing logic 720 and the storage logic on the control node side 715 used to carry control messages that are not part of the SCSI protocol and are not sent to the SAN 130 are to be transmitted. These control messages are instead through the control node logic 715 to treat.

The protocol control messages are always generated by the processor side of the protocol and via one of the two virtual interfaces (VIs) that define the processor-side logic 720 with the storage logic on the control node side 715 connect, sent to the control node side of the protocol. Except that other flag bits are used to distinguish the message from a protocol control message, the message header used for protocol control operations is the same as a command message header. The control node 120 performs the requested operation and responds via the RVI with a message header that is the same as that used by a status message. In this way, a separate RVI is not required for the rarely used protocol control operations.

According to certain embodiments that use redundant control nodes, the processor-side logic captures 720 certain errors from issued commands and, in response, reissues the command to the other control node. This repetition can be done in a driver 710 the middle level.

Control node-side Holder logic

According to certain embodiments, the storage node-side storage logic 715 implemented as a device driver module. The logic 715 puts in the control node 120 a device level interface to the operating system. This device level interface is also used on the configuration component 705 access. When this device driver module is initialized, it responds to protocol messages from all processors 106 and the platform 100 , The device level interface introduces all configuration activity. All I / O activity is introduced through messages that are wired 110 and over the switch fabric 115 be sent and received. In the control node 120 there is an instance of logic 715 per processor node 105 (although in 7 is only shown as a box). According to certain embodiments, the node logic communicates 715 through the FCP or FCP-2 protocol or through the iSCSI or other protocols using the SCSI-2 or SCSI-3 instruction set, through various media with the SAN 130 ,

As described above, the processor logic in the RVI message headers sets flags that indicate whether the command is associated with a data flow and, if so, in which direction. The storage logic on the control node side 715 receives messages from the processor logic and then analyzes the header information to determine how to act, e.g. B. allocate buffers or the like. The logic also translates the address information contained in the messages from the processor into the corresponding translated SAN address and gives the commands (e.g. via FCP or FCP-2) to the SAN 130 out.

A SCSI command, such as a TEST UNIT READY command, that does not require a SCSI data transfer phase is provided by the processor logic 720 that sends a single command in the RVI used for command messages and handled by the node logic that sends a single status message back over the same RVI. More specifically, the processor side of the protocol constructs the message with a standard message header, a new sequence number for this command, the desired SCSI target and LUN, the SCSI command to be executed, and a list size of zero. The control node side of the logic receives the message, takes the information from the SCSI command and transports it via the interface 128 into the SAN 130 , After receiving the command completion recall, the control node constructs a status message for the processor using a standard message header, the sequence number for this command, the status of the completed command, and optionally the request scan data if the command was completed with a test status.

A SCSI command, such as a READ command, that requires a SCSI data transfer phase to transfer data from the SCSI device to host memory is handled by the processor-side logic that sends a command message to the node logic 715 sends, the control node having one or more RDMA WRITE operations in the memory in the processor node 105 and respond with a single status message from the control node logic. The processor-side logic constructs more precisely 720 a command message with a standard message header, a new sequence number for this command, the desired SCSI target and LUN, the SCSI command to be executed and a list of storage areas in which the data of the command are to be stored. The logic on the control node side 715 allocates temporary memory buffers that store the data from the SCSI operation while the SCSI command is being executed in the control node. After the control node logic 715 the SCSI command for processing to the SAN 130 and the command is complete, it sends the data to the processor's memory with a sequence of one or more RDMA-WRITE operations 105 back. It then constructs a status message with a standard message header, the sequence number for this command, the status of the completed command and optionally the REQUEST-SENSE data if the command was completed with a SCSI CHECK CONDITION status.

A SCSI command, such as a WRITE command, that requires a SCSI data transfer phase to transfer data from host memory to the SCSI device is handled by the processor-side logic 720 a single command message to the control node logic 715 sends one or more BUF messages from the control node logic 715 sends to the processor side logic, sends one or more RDMA WRITE operations from the processor side filing logic to the memory in the control node, sends one or more TRAN messages from the processor side logic to the control node side logic, and returns a single status message from the control node side logic to the processor-side logic. The use of the BUF messages, to transmit the location of the temporary buffer memory in the control node to the processor-side filing logic and the use of the TRAN messages to indicate the completion of the RDMA-WRITE data transfer is due to the lack of RDMA-READ capability in the underlying Giganet fabric. If the underlying fabric supports RDMA-READ operations, another sequence of appropriate actions can be used. The processor-side logic constructs more precisely 720 a CMD message with a standard message header, a new sequence number for this command, a desired SCSI target and a desired LUN as well as the SCSI command to be executed. While the SCSI command is being executed in the control node, the control node logic orders 715 temporary storage buffers that store the data from the SCSI operation. The control node side of the protocol then constructs a BUF message with a standard message header, the sequence number for this command, and a list of areas of virtual memory to be used for the temporary memory buffers in the control node. The processor-side logic then sends 720 the data to the control node memory with a sequence of one or more RDMA-WRITE operations. It then constructs a TRAN message with a standard message header and a sequence number for this command. After the control node logic sends the SCSI command to the SAN for processing 130 has sent and received the command completion, it constructs a STAT message with a standard message header, the sequence number for this command, the status of the completed command and optionally the REQUEST-SENSE data if the command was completed with a status CHECK CONDITION ,

According to some embodiments contains the CMD message a list of areas of virtual memory, from which the data for the command can be saved. In addition, the BUF and The TRAN message is an index field that enables the control node side of the log a separate BUF message for each entry in the area list sends in the CMD message. The processor side of the protocol responds on such a message that it is the RDMA-WRITE operations for the executes what amount of data described in the BUF message a TRAN message follows that concludes a single segment of the data transfer indicates.

The protocol between the processor-side logic 720 and the logic of the control node 715 enables scatter-collect I / O operations. This functionality enables the data affected by an I / O request to be read from or written to several different areas of the virtual and / or physical memory. This enables multiple non-contiguous buffers to be used for the request to the control node.

As stated above, the configuration logic is 705 for the discovery of the SAN storage assigned to the platform and for the interaction with the interface logic 710 responsible so that an administrator can sub-assign the storage to specific PANs. As part of this assignment, the configuration component creates and maintains 705 a filing data structure 915 that contains information that identifies the correspondence between processor addresses and actual SAN addresses. 7 shows such a structure. As described above, the correspondence between the processing node and the identifier of the simulated SCSI disks can e.g. B. by the name of the simulated controller, by the cable, by the unit or by the logic unit number (LUN).

Management logic

The management logic 135 is used to create an interface to supply the PANs. Among other things, logic enables 135 that an administrator knows the virtual network topology of a PAN, its visibility to the external network (e.g. as a service cluster) and the types of devices in the PAN, e.g. B. bridges and line.

It also represents logic 135 an interface with the filing management interface logic 710 so that an administrator can define the storage for a PAN during the initial assignment or afterwards. The configuration definition contains the storage correspondence (SCSI to SAN) discussed above as well as access control authorizations.

As described above each of the PANs and each of the processors one by its virtual Networking (including a virtual MAC address) and defined by its virtual filing Personality. As described below, the structures that record this personality can be characterized by the management logic is accessed to the processor clustering to implement. Moreover can be accessed by an administrator or as described above be accessed with an agent administrator. An agent can for example used to be a PAN in response to certain Events such as the time of day or season or in response to reconfigure to certain loads in the system.

The operating system software in a processor contains code for the serial console driver that handles the console I / O traffic for the node through the Giganet switch 115 to the management software that runs in a control node. From there, the management software can run the Konso len I / O stream of any node through the control node's management ports (its slow Ethernet port and its emergency management port) or through the fast external network 125 as permitted by the administrator. Console traffic can be logged for audit and history log purposes.

Cluster management logic

9 illustrates the cluster management logic of certain embodiments. The cluster management logic 905 accesses the data structures 910 that record the network information described above, such as the network topologies of the PANs, the MAC address assignments in a PAN, etc. The cluster management logic also applies 905 on the data structures 915 to which the filing correspondence of the different processors 106 record. The cluster management logic also applies 905 on a data structure 920 too, the free resources such as unallocated processors in the platform 100 records.

The cluster management logic 905 may change data structures in response to processor failure events or administrator commands to cause the filing and networking personalities of a given processor to be "switched" to a new processor. In this way, the new processor "inherits" the personality of the previous processor. It can be arranged for the cluster management logic 905 does this to replace a new processor in a PAN to replace a failed one.

The new processor inherits a MAC address earlier Processor and acts like the previous one. The Control node transmitted the connectivity information, when the new processor boots and updates the if necessary Connectivity information for the not failed processors. For example, in certain embodiments the RVI connections for the other processors updated transparently; d. that is, the software in the other processors needs to connect to that newly substituted processor not to be involved. Also inherits the new processor the filing correspondence of the previous and hence the permanent state of the previous processor.

Among other advantages, this enables a free pool of resources including processors through the entire platform rather than given Share PANs. In this way, the free resources (which as such can be considered to be reliability and to increase fault tolerance of the system) used more efficiently become.

When a new processor is "substituted", it has to be renewed for it an ARP processing accomplished to learn how to assign IP addresses to MAC addresses.

alternatives

Because each Giganet port of the switch fabric 115 Supporting 1024 simultaneous virtual interface connections and keeping them separate with hardware protection, the operating system can safely share a node's Giganet ports with application programs. This enables direct connection between application programs without the need to run over the full stack of driver code. For this purpose, an operating system call creates a virtual interface channel, it carries out a memory conversion of its buffers, and it queues its application address space. In addition, a library that encapsulates the hardware-related details of the interface to the channel facilitates the use of such virtual interface connections. In addition, the library could automatically establish redundant virtual interface channel pairs and manage or fail over them without any effort or knowledge from the application being called.

Emulate the above-described embodiments the ethernet internally an ATM-like Fabric. The draft could changed to use an internal ethernet fabric whatever the architecture would significantly simplify and Z. B. Eliminate the need to emulate features would. If the external network according to the ATM would communicate a further modification of the ATM internally without emulation of the Ethernet use while the ATM could be communicated externally to the external network if it is addressed in this way. Another variation would be one ATM internal to the platform (i.e. without emulation of the Ethernet) enable, while only external communications are converted to the ethernet. This would rationalize internal communications, but emulation logic in the controller.

Certain embodiments use PANs based on software configuration commands. It is clear that the application can be based on program control. For example can while the peak operating time for this PAN according to software control more processors can be used or can be controlled according to a software algorithm accordingly more or less memory can be used for a PAN.

It is clear that the scope of the present invention is not limited to the above-described embodiments, but is defined by the appended claims; these claims changes and improvements to the Be include written.

Summary

Virtual networking system and method in a processing system System and method for virtual networking ( 100 ). A semantics of a mediated local Ethernet network is provided via an underlying point-to-point mesh network. The computer processor nodes can communicate directly via virtual interfaces via a switch fabric or can use an Ethernet switching emulator ( 115a . 115b ) communicate. To assign IP addresses to virtual interfaces, ARP logic (address resolution protocol logic) ( 135 ) is used, allowing computer processors with virtual MAC addresses to respond to ARP requests.
( 1 )

Claims (58)

Method for emulating a mediated local Ethernet network in a platform with a plurality of computer processors, a switch fabric and point-to-point connections to the processors, full: Providing Ethernet driver emulation logic for execution in at least two computer processors; Deploy one Switch emulation logic for execution in at least one of the Computer processors; Establish a virtual interface between the switch emulation logic and each computer processor, in which an ethernet driver emulation logic is executed, to enable software communication in between, with each virtual interface via the Switch fabric a software communication path from a computer processor defined to another computer processor; Making one virtual interface between each computer processor in which an ethernet driver emulation logic is running, and each other Computer processor in which an ethernet driver emulation logic accomplished becomes; in which, if the virtual interface between one Computer processor and another works in such a way that it fulfills predetermined criteria, the ethernet driver emulation logic of a computer processor a virtual interface that is a software communication path defined in between, point to point with the other computer processor communicating; and if the virtual interface between one computer processor and another does not work in the way that it meets specified criteria, the ethernet driver emulation logic of a computer processor a virtual interface to the switching emulation logic, which is the point-to-point communication to the other computer processor, point to point with communicates with the other computer processor. The method of claim 1, wherein each of the computer processors in which an Ethernet driver emulation logic is executed a virtual MAC address is assigned, and in which the MAC addresses according to rules be formed that the computer processor as one of the majority of Identify computer processors, and one of the MAC addresses external network are different. The method of claim 2, wherein the platform via a Interface logic for communication with an external network connected is where the external network interface logic has its own MAC address is assigned and in which the messages in the external network using the MAC address the external network interface logic are communicated. The method of claim 1, wherein the first processor a first virtual interface for point-to-point communication used with a second computer processor while the second computer processor another virtual interface for communication with the first Computer processor used. The method of claim 1, wherein each computer processor switch fabric driver logic Communication on the point-to-point connections includes as well an ability for the creation of checksums contains and where the ethernet driver emulation logic is a capability for the creation of checksums contains taking this formation of checksums but locks if the switch fabric driver logic already the checksum a message. The method of claim 5, wherein the switch fabric driver logic a reliable Communication protocol implemented to receive messages over the Ensure switch fabric. The method of claim 1, wherein the switch fabric and the point-to-point connections in are arranged in a redundant configuration. The method of claim 1, wherein the Ethernet driver emulation logic communicates a message by broadcasting the message to the switch emulation logic through a virtual interface, and wherein the switch emulation logic receives and clones a broadcast message from a virtual interface and clones the clone sends th message to other computer processors on the network. The method of claim 1, wherein the switch emulation logic defines a computer processor membership in an emulated network and entertains. The method of claim 1, wherein the Ethernet driver emulation logic is messages sends that larger than the size of the maximum transmission unit (MTU) are. System for emulating a mediated local Ethernet network, comprising: a plurality of computer processors; a Switch fabric and point-to-point connections to the processors; a virtual interface logic for creating virtual interfaces via the Switch fabric and over that Point-to-point connections, each virtual interface a software communication path from a computer processor to another computer processor defines the switch fabric; an Ethernet driver emulation logic, executed in at least two computer processors; a Switch emulation logic that is in at least one of the computer processors accomplished is containing a logic for making a virtual Interface between the switch emulation logic and everyone Computer processor in which the Ethernet driver emulation logic is executed to enable software communication in between, a logic to Receive a message from one of the virtual interfaces for one Computer processor in which an Ethernet driver emulation logic is executed and send a message to another computer processor, in which an ethernet driver emulation logic is executed, in response to addressing information associated with the message are; and logic for establishing a virtual interface between each computer processor in which an ethernet driver emulation logic accomplished and any other computer processor that contains an Ethernet driver emulation logic accomplished becomes; in which the ethernet driver emulation logic from Punkt to point over a virtual interface that is a software communication path defined in between, with another computer processor in the emulated ethernet network communicates if the virtual interface works in such a way that it meets predetermined criteria while it is from point to point over the switch emulation logic communicates with it if the virtual interface does not work in such a way that it is predetermined Criteria met. The system of claim 11, wherein each of the computer processors in which an Ethernet driver emulation logic is executed a virtual MAC address is assigned, and in which the MAC addresses are formed according to rules that the Computer processor as one of the majority of computer processors identify, and are different from the MAC addresses of an external network. The system of claim 12, further comprising an external one Network interface logic for communication with an external network, where the external network interface logic has its own MAC address is assigned and in which the switching emulation logic is a Logic for sending messages to the external network interface logic contains for communication in the external network, these messages using the MAC address of the external network interface logic be communicated in the external network. The system of claim 11, wherein the first processor a first virtual interface for point-to-point communication used with a second computer processor while the second computer processor another virtual interface for communication with the first Computer processor used. The system of claim 11, wherein the computer processor switch fabric driver logic for communication in the point-to-point connections and at which has switch fabric driver logic an ability to generate checksums and in which the Ethernet driver emulation logic has a capability for the creation of checksums owns and contains logic which is the formation of checksums locks in the ethernet driver emulation logic if the switch fabric driver logic the checksum a message. The system of claim 15, wherein the switch fabric driver logic a reliable Communication protocol implemented to receive messages over the Ensure switch fabric. The system of claim 11, wherein the switch fabric and the point-to-point connections in are arranged in a redundant configuration. The system of claim 11, wherein the Ethernet driver emulation logic includes logic for communicating a broadcast message by sending the message to the switch emulation logic via a virtual interface, and wherein the switch emulation logic includes broadcast logic for receiving and cloning a broadcast message from a virtual interface and for sending the cloned message to other computer processors on the network. The system of claim 11, wherein the switch emulation logic logic to define and maintain computer processor membership in an emulated network. The system of claim 11, wherein the Ethernet driver emulation logic contains logic for sending messages that are larger than the size of the maximum transmission unit (MTU) are. Method for implementing the address resolution protocol (ARP) in a computer platform with a plurality of processors, full: Define a topology in the IT platform Ethernet network to be emulated, the topology being processor nodes and contains a switching node; Allocate a lot of processors from the majority as processors acting as the processor nodes Act; Assign a processor that acts as the switch node acts; Assign virtual MAC addresses to each processor node the emulated Ethernet network; Assign virtual interfaces via a underlying physical network to provide a direct software communication from every processor node to everyone other processor nodes, with each virtual interface one has the corresponding identifier; in which a processor node a ARP request communicates to the switching node, the ARP request contains an IP address; the switching node sends the ARP request to all other processor nodes communicates in the emulated Ethernet network; a processor node, to which the IP address is assigned, one at the switching node ARP response that outputs the virtual MAC address associated with the IP address Processor node contains; the Switching node receives the ARP response and changes it in such a way that it contains a virtual interface identifier for a virtual interface that the processor node that issues the ARP request for the subsequent one Communication with the processor node to which the IP address is assigned is to use. A method according to claim 21, in which the basis lying physical network is a point-to-point mesh network, that connects the majority of processors. 22. The method of claim 21, wherein a subset the processors are organized as a cluster and one of the processors in the cluster is a load balancing processor node and when any processor in the cluster makes an ARP request outputs, the switching node changes the ARP response in such a way that the virtual interface identifier for the load balancing processor node contains. 22. The method of claim 21, wherein the switching node is in communication with an external IP network and in which the Act of communicating an ARP response identifying that the ARP response comes from a processor node in the platform, contains. Address resolution protocol system (ARP system), full: an IT platform with a plurality of processors, connected by an underlying physical network are; logic that is executable in one of the processors, a topology of an Ethernet network to be emulated in the IT platform to define, with the topology processor node and a switching node contains; a Logic that is executable in one of the processors to a lot of Assign processors from the plurality that are processors that act as the processor nodes; a logic that is in one of the Processors is executable, a virtual one around each processor node of the emulated Ethernet network Assign MAC address; a logic in one of the processors is executable, around virtual interfaces assign the underlying physical network so that a direct Software communication from every processor node to every other Processor node is provided, each virtual interface has a corresponding identifier; in which  every processor node has ARP request logic around an ARP request to the switch node communicate, the ARP request containing an IP address; the Switching node contains an ARP request broadcast logic to the ARP request to all other processor nodes in the emulated Ethernet network to communicate; each processor node one ARP response logic to determine if it is the processor node to which the IP address is assigned in an ARP request, and to issue an ARP response to the switch node if that is the case where the ARP response is the virtual MAC address of the Processor node contains to which the IP address is assigned; the switching node one ARP response logic contains to receive the ARP response and change it in such a way that it contains a virtual interface identifier for the ARP request node. The system of claim 25, wherein the underlying physical network is a point-to-point mesh network that the majority of Processors connects. The system of claim 25, wherein a subset the processors are organized as a cluster and one of the processors in the cluster is a load balancing processor node and in which the switching node contains logic to detect whether an ARP response from a processor node from any Processor comes in the cluster, and around the ARP response in the way to change, that they are the virtual interface identifier for the load balancing processor node contains if that's the case. The system of claim 25, wherein the switching node is in communication with an external IP network and in which the ARP response logic of the processor node contains logic to identify that the ARP response from a processor node in the platform comes. Platform for computer processing, comprising: a plurality of computer processors, connected to an internal communication network; at least a control node in communication with an external communication network and with an external filing network that addresses an external filing space has, and wherein the at least one control node with the internal Is connected and thereby in communication with the majority of the Computer processors stands; a configuration logic to a Define and manufacture a virtual processing area network with: a corresponding set of computer processors from the majority of processors, a virtual local communication network that provides communication among the crowd of computer processors, while there are those processors from the majority that are not in the defined amount are excludes and a virtual storage space with a defined correspondence to the address space of the filing network. The platform of claim 29, wherein the control node is via the internal communication network receives communication messages that of entities are addressed in the external communication network, and at the control node contains logic that Messages that correspond to the messages received in the provides external communication network. The platform of claim 29, wherein the control node is via the external communication network receives communication messages that of entities are addressed in the platform, and at which the control node is a Logic contains the messages that correspond to the received messages for the addressed ones entities provides. The platform of claim 29, wherein the computer processors and the control node includes network emulation logic to communicate over the internal communication network to emulate the ethernet functionality. The platform of claim 32, wherein the internal Communication network is a point-to-point switch fabric. The platform of claim 29, wherein the internal Communication network includes redundant wiring that the Computer processors and the at least one control node with a redundant switch Fabrics connects. The platform of claim 34 having at least one another control node connected to the wiring to form redundant control nodes. The platform of claim 29, wherein the control node is via the internal communication network filing messages from computer processors receives and the control node includes logic to extract from a received one Filing message to extract an address that corresponds to the defined one Address in the external filing address space identified and the Messages that correspond to the received filing messages and that have the appropriate address in the external filing network. The platform of claim 36, wherein the control node contains logic the data representing the write messages received by the computer processor correspond to buffer and the buffered data in the corresponding to provide message provided in the external filing network. The platform of claim 36, wherein the control node Contains filing messages from the external filing network and at the control node contains logic to match a corresponding one Identify computer processor or control node for which the received message is a reply, and a corresponding one Message for to provide the identified processor or control node. A method for deploying a virtual processing area network, comprising the following steps: providing a platform with a plurality of computer processors and with at least one control node which is connected to an internal communication onsnetz is connected, and in which the control node is in communication with an external communication network and with an external filing network, which has an external filing address space, defining a corresponding amount of computer processors for the virtual processing network, establishing a virtual local communication network that communicates under the Provides set of computer processors while excluding those processors from the plurality that are not in the defined set, defining a correspondence between a virtual storage space of the virtual processing network and the address space of the storage network. A method according to claim 39, wherein the control node via the internal communication network receives communication messages that of entities are addressed in the external communication network, and in which the control node messages which correspond to the received messages, in the external communication network. A method according to claim 39, wherein the control node via the external communication network receives communication messages that of entities are addressed in the platform and where the control node messages, which correspond to the received messages for the addressed entities. The method of claim 39, wherein the computer processors and the control node over that internal communication network emulate an ethernet functionality. The method of claim 42, wherein the internal Communication network is a point-to-point switch fabric and at the emulation of the Ethernet functionality via the internal point-to-point switch fabric provided. The method of claim 39, wherein the computer processors have a redundant wiring that communicate the computer processors and connects the at least one control node. The method of claim 44 with at least one another control node connected to the wiring to form redundant control nodes. A method according to claim 39, wherein the control node via the internal communications network receives and receives filing messages from the computer processors in which the control node from a received filing message Address takes the defined corresponding address in the external Filing address space identifies and messages that are received Correspond to filing messages and have the corresponding address, in the external filing network. The method of claim 46, wherein the control node Data that corresponds to the write messages received by a computer processor correspond, buffers and the buffered data in the corresponding Providing message for the external filing network is provided. The method of claim 46, wherein the control node Receives filing messages from the external filing network and in which the control node has a corresponding computer processor or control node for which the receiving message is identified is an answer, and a corresponding message for the identified Provides processor or control node. Platform for computer processing using an external communication network and can be connected to a filing network and includes: a Plurality of computer processors connected to an internal communication network are connected; a configuration logic to define and manufacture: (a) a virtual local communication network through the internal network where every computer processor in the virtual local communication network a corresponding virtual MAC address owns and in which the virtual local network communicates deployed under a lot of computer processors while it those processors from the majority that are not defined in the Amount are excludes and (b) a virtual storage space with a defined correspondence to the address space of the filing network; failover logic, that responds to a failure of a computer processor Assign computer processor from the majority, so that it is the failed Processor replaced, the failover logic includes logic that assigns the MAC address of the failed processor to the processor, that replaces the failed processor contains logic that the virtual storage room and the defined equivalent of the failed Processor assigns the processor that the failed processor replaced, and contains logic which restores the virtual local area network so that it is the processor contains which replaces the failed processor while it replaces the failed Processor excludes. The platform of claim 49, wherein the configuration logic creates virtual interfaces to define software communication paths among the processors of the virtual network, and wherein the failover logic includes logic that virtual interfaces from the processors in the vir current network to the processor that replaces the failed processor. The platform of claim 49, wherein the configuration logic a second virtual local area network from a second set of Computer processors and a second virtual storage room with produces a defined correspondence to the filing network address space and where the failover logic causes the processor to which replaces the failed processor, the personality of the virtual local network and the virtual filing of the failed processor inherits. Computer processing methods in a platform with a plurality of computer processors that work with an internal Communication network, comprising: Define and Establish a virtual local communication network via the internal network in which each computer processor in the virtual local Communication network has a corresponding virtual MAC address and in which the virtual local area network communicates under one Lot of computer processors while providing those processors excludes from the majority that are not in the defined amount; Define and creating a virtual storage space with a defined one Correspondence to the address space of the filing network; in reaction on a failure of a computer processor, assigning a computer processor the majority to replace the failed processor, including assignments the MAC address of the failed processor to the processor that the failed processor replaced, allocating the virtual storage space and the defined correspondence of the failed processor the processor to replace the failed processor and restore of the virtual local area network so that it contains the processor that replaced the failed processor while it replaced the failed Processor excludes. 53. The method of claim 52, wherein in manufacturing of a virtual local area network the software communication paths among the processors of the define virtual network, where if a processor has a failed Processor replaced, virtual interfaces to the processor established to replace the failed processor. The method of claim 52, wherein a second virtual local area network with a second set of computer processors and with a second virtual storage room with a defined one Correspondence to the filing network address space is established and where if a processor fails, the processor that replaces the failed processor, the personality the virtual local network and the virtual filing of the failed Processor inherits. System for providing one by an IP address addressed service, including: at least two computer processors, each containing logic to provide the service; a Cluster logic that receives a request message for the service, wherein the messages have the IP address and the request distributed to one of the at least two computer processors, the one Has logic to provide the service. The system of claim 55, wherein the logic that distributed, contains logic, in determining which processor should serve the message the source information is analyzed in an incoming message. Procedure for providing one by an IP address addressed service, including: Incorporating a logic that provides the service in each of the at least two computer processors; Receive a request message for the service, where the message has the IP address, and distribute the request to one of the at least two computer processors, which has logic to provide the service. The method of claim 57, wherein the source information an incoming message can be analyzed to determine which processor should serve the message.