JP2008181387A - I/o bus system and its management method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an I/O bus system which can measure performance in a unit, which is grouped for each application or for each virtual machine, independently of an I/O device unit. <P>SOLUTION: The I/O bus system is provided with: grouped ID-setting parts 113a-113f which become grouping means grouping, in an application unit or in a virtual machine unit, the data flowing on the I/O bus system where the data are transmitted between CPUs 101a-101c and I/O devices 102-104; and a traffic monitor part 140 which monitors traffics in a group unit. The grouped ID-setting means can determine an grouped ID which is allocated according to grouping conditions stored in a group management table which holds the grouping conditions. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an I / O bus system for connecting a CPU and an I / O device and a management method thereof, and more particularly to an I / O bus system capable of managing the usage status of an I / O device and a management method thereof.

  Conventionally, as a technology related to this type of I / O bus system, for example, there is a bus system that monitors and controls a bus occupation time allocated to each functional block by an arbiter described in Patent Document 1.

  FIG. 17 is a block diagram showing the configuration of the bus system described in Patent Document 1. In FIG. In the bus system shown in FIG. 17, a CPU 920, an arbiter 910, and I / O devices 941 to 943 are connected to an I / O bus 930. The arbiter 910 includes a bus monitor unit 911, a device information management unit 912, and a bus usage right management unit 913.

  The bus monitor unit 911 monitors the I / O bus 930 and acquires the number of clocks and the data transfer length occupying the I / O device bus 941 to 943. The device information management unit 912 calculates the actual bandwidth occupied by the I / O devices 941 to 943 from the number of clocks and the data transfer length acquired by the bus monitor unit 911, and determines the bus occupation time according to the actual bandwidth. To do. The bus usage right management unit 913 gives a bus usage right to each I / O device according to the bus occupation time determined by the device information management unit 912.

  Next, the operation of the I / O bus system will be described. The bus monitor unit 911 monitors the I / O bus 930, acquires the number of clocks that each I / O device occupies the I / O bus and the data transfer length therebetween, and the number of clocks and the data transfer length. Is notified to the device information management unit 912. The device information management unit 912 calculates the actual bandwidth actually used by the I / O device from the number of clocks and the data transfer length. As a result of the calculation of the real band, for example, when the real band of the I / O device 941 is smaller than the allocated band at this time and the real band of the I / O device 942 is larger than the allocated band, the device information management unit 912 The bus occupation time of the I / O device 941 is shortened, and the bus occupation time of the I / O device 942 is expanded. The device information management unit 912 notifies the bus usage right management unit 913 of the newly determined bus occupation time. The bus usage right management unit 913 gives a bus usage right to each I / O device according to the bus occupation time determined by the device information management unit 912.

  With the above operation, the bus bandwidth actually used can be monitored, and the bus occupation time of each I / O device can be controlled according to the result.

  Patent Document 2 describes that in an ATM switch in which a plurality of access modules are connected by an optical bus, the plurality of access modules are divided into a plurality of groups so that the traffic amount for each group is equal. is there. In Patent Document 3, a LAN control device and a plurality of hub devices are connected to two buses. The LAN control device divides a plurality of hub devices into groups, monitors traffic via the bus, and responds to the traffic. There is a description of determining the number of group divisions. However, in Patent Documents 2 and 3, data is not grouped by data contents (application, virtual machine unit, etc.).

In Patent Document 4, a communication performance measuring device transmits and receives counting packets at regular time intervals to measure a round trip time, investigates a ratio that is longer than the immediately preceding round trip time, and determines a network route based on the ratio. There is a description of performing performance determination for determining whether the available bandwidth exceeds a predetermined value. Patent Document 5 describes that an evaluation server generates a set of performance evaluation amounts such as a path bandwidth, throughput, goodput, and packet loss rate at a measurement point on a network. However, there is no description of converting the monitored traffic information into a performance index corresponding to the type of group target.
JP 2003-30133 A JP 2000-069048 A (paragraph numbers [0028], [0029], etc.) JP-A-7-107099 (Claims 1 and 3, paragraph number [0016], etc.) JP 2000-106557 A (paragraph number [0018] etc.) Japanese Patent Laid-Open No. 2001-197085 (paragraph numbers [0020]-[0022] etc.)

The I / O bus system described above has the following problems.
The first problem is that performance management cannot be performed in units of virtual machines or applications that operate in combination with a plurality of I / O devices.

  The reason is that the management target is an I / O device unit.

  The second problem is that the CPU load is consumed when the bandwidth is measured with a granularity other than the I / O device. For example, when a plurality of virtual machines are operating on the CPU by the virtualization technology, it is necessary to consume the CPU load when trying to grasp the usage status of the I / O device for each virtual machine.

  The reason is that it is necessary to monitor the scheduler in the virtualization software running on the CPU in order to monitor the usage status of the I / O device for each virtual machine.

  A third problem is that measurement accuracy is lowered when a band is measured with a granularity other than that of an I / O device. For example, when a plurality of virtual machines are operating on the CPU by the virtualization technology, the measurement accuracy is lowered when trying to grasp the performance status of the I / O device for each virtual machine.

  The reason is that the virtualization software associates the virtual machine with the I / O device. In general, software processing depends on timer accuracy controllable by the OS. Since the timer accuracy that can be controlled by the OS is low, the measurement accuracy of the usage status of the I / O device for each virtual machine is also low.

  A fourth problem is that when there are a plurality of I / O devices and the usage status of the I / O devices is biased, the usage efficiency of the I / O devices cannot be improved as a whole system.

The reason for this is that the controllable item is the traffic band, and no means for changing the I / O device assigned to the application or virtual machine is provided.
(Object of the present invention)
An object of the present invention is to provide an I / O bus system capable of accurately measuring the usage status of an I / O device.

  Another object of the present invention is to provide an I / O bus system capable of measuring the usage status of an I / O device while suppressing the CPU load.

  Furthermore, another object of the present invention is to provide an I / O bus system capable of measuring the performance status for each of various objects such as each virtual machine or each application.

  Furthermore, another object of the present invention is to provide an I / O bus system capable of improving the usage efficiency of an I / O device as a whole system even when the usage status of the I / O device is biased. is there.

The I / O bus system of the present invention is an I / O bus system for transferring data between a CPU and an I / O device,
The data to be transferred is monitored, grouped according to the contents of the monitored data, and a group ID setting means for assigning a group ID of the group to the data, and traffic for each group ID assigned by the group ID setting means And a traffic monitor means for monitoring the traffic.

An I / O bus system management method of the present invention is an I / O bus system management method for an I / O bus system that transfers data between a CPU and an I / O device.
The contents of the data to be transferred are monitored, grouped according to the contents of the monitored data, and traffic is monitored in units of set groups.

  According to the present invention, the performance in units grouped for each application or virtual machine can be measured. The reason is that a group is assigned to each application or virtual machine, and traffic for each group is monitored.

  Next, the best mode for carrying out the invention will be described with reference to the drawings.

[First Embodiment]
Referring to FIG. 1, the I / O bus system 1 of the present embodiment includes a group ID setting unit 113a-113f serving as a group ID setting unit and a traffic monitor unit 140 serving as a traffic monitoring unit. The I / O bus system 1 is connected to the CPUs 101a to 101c and the I / O devices 102 to 104.

  The group ID setting means 113a-113f monitors the content of transfer data transferred between the CPUs 101a-101c and the I / O devices 102-104, and assigns the monitored data content based on preset grouping conditions. A group ID is determined, and a group ID (group identifier) is assigned to the data. The traffic monitor unit 140 monitors traffic for each group based on the assigned group ID. Although not described in detail here, based on the monitored traffic, traffic can be controlled for each group using a known method. As an implementation method for monitoring traffic, for example, there is a method of using Ethernet VLAN (Virtual LAN).

  Next, the configuration of the group ID setting unit will be described.

  FIG. 2 is a block diagram illustrating a configuration example of the group ID setting unit. 2 shows the configuration of the group ID setting unit 113a among the group ID setting units 113a to 113f, the group ID setting unit 113b to 113f has the same configuration.

  In FIG. 2, the group ID setting unit 113 a includes a data monitor unit 151, a group ID determination unit 152, and a group ID provision unit 153. The data monitor unit 151 monitors the contents of the transfer data. The group ID determining unit 152 determines a group ID to which the data content monitored by the data monitoring unit 151 is allocated based on a preset grouping condition. The group ID assigning unit 153 assigns the determined group ID to the data.

  Next, the operation of the group ID setting unit in FIG. 1 will be described with reference to the flowchart in FIG.

  FIG. 3 is a flowchart showing an operation in which the group ID setting unit assigns a group ID.

Each of the group ID setting units 113a to 113f monitors data transferred to the I / O bus (step S111), determines whether the data content matches the group condition (step S112), and if the data matches the group condition. The group ID of the matched group is assigned to the data (step S113). If it does not match the group condition in step S112, the group ID is not assigned and the data is transferred as it is (step S114).
Then, the traffic monitor unit 140 monitors the traffic for each group based on the assigned group ID.

  Next, a configuration example in the case where the CPUs 101a to 101c and the I / O devices 102 to 104 are made compatible with PCI Express will be described. The CPUs 101a to 101c and the I / O devices 102 to 104 have PCI Express addresses (bus numbers and device numbers), and the group ID setting units 113a to 113f have Ethernet (registered trademark) MAC addresses. The correspondence between these two addresses can be acquired by encapsulating the PCI Express address in an Ethernet frame and transmitting it at the time of initial setting or the like.

  The traffic monitor unit 140 may be software processing, but can perform high-speed processing by performing hardware processing.

  The traffic monitor targets include, for example, a bandwidth for each switch port and VLAN, a packet loss rate, and a packet size.

Referring to FIG. 4, the configuration of a group ID setting unit 113a that sets a VLAN ID as a group ID is shown. The group ID setting units 113b to 113f other than the group ID setting unit 113a have the same configuration.
The group ID setting unit 113a includes a PCI Express processing unit 251, a data monitoring unit 252, a protocol conversion unit 256, a group ID processing unit 253, an Ethernet processing unit 254 (Ethernet is a registered trademark), and a transmission / reception unit 255. Is done.

  When transmitting to the network, the PCI Express processing unit 251 monitors the bus number and the device number as an identifier for specifying the destination and the transmission source from the header contents of the PCI Express. Conversely, when receiving from the network, the PCI Express bus number and device number corresponding to the destination and the source MAC address are assigned.

  The data monitor unit 252 monitors the contents of data and identifies the application and command type that process this data. As an implementation method for identifying an application, for example, there is a method of monitoring a TCP port number in data. As a method for identifying the command type, for example, there is a method for monitoring the command type in the header.

  FIG. 16 is a diagram illustrating a configuration of a PCI Express header format. In the header format information of FIG. 16, “R” indicates reserved or unused, “Fmt” indicates the type of packet format, “Type” indicates the instruction type of the packet, and “TC” indicates the traffic class. , “TD” is a bit indicating whether or not a TLP digest is used, “EP” is a bit indicating data abnormality, “Length” indicates a packet length, “Bus Number” indicates a bus number, “Device” “Number” indicates a device number, and “Function Number” indicates a function number. The command type (command type) can be monitored by “Type” of the header information, and the device type can be monitored by “Bus Number”, “Device Number”, and “Function Number”.

As other granularities for identifying data contents, there are a data transfer method using PIO (Programmed IO) and DMA (Direct Memory Access), a virtual machine, and the like.
The protocol conversion unit 256 converts a protocol on the CPU or I / O device side and a protocol on the bus side. In this conversion, an address conversion table as shown in FIG. 18 is referred to. The address conversion table shown in FIG. 18 is an example when PCI Express is used as the protocol on the CPU or I / O device side and Ethernet is used as the address on the bus side network. Referring to FIG. 18, it can be seen that, for example, the address of 00: 11: 22: 33: 44: 55 of Ethernet corresponds to the address of bus ID = 1 and device ID = 1 in PCI Express.

  The VLAN ID processing unit 253 determines a VLAN ID to be given to the data based on the monitored data contents and the group condition of the group management table.

  The Ethernet processing unit 254 assigns an Ethernet destination and monitors a transmission source address, and assigns or removes a VLAN ID.

  The transmission / reception unit 255 transmits / receives data in accordance with network protocol standards.

  The contents of the group management table are shown in FIG. In the group ID management table shown in FIG. 5, a VLAN ID as a group ID, a group condition for belonging to the group, and a band allocated to the group are registered. Referring to FIG. 5, it can be seen that VLAN ID = 11 is assigned to the data of TCP port number = 54321 in the first line. Further, a condition may be defined by combining a plurality of conditions such as VLAN ID = 21. Further, the condition definition may be any condition such as AND, OR, NOT, etc., such as VLAN ID = 31. Also, by setting the priority for each VLAN condition, it is possible to determine which VLAN ID is to be assigned even when one piece of data satisfies a plurality of VLAN conditions.

  Next, an operation of a configuration example when the CPUs 101a to 101c and the I / O devices 102 to 104 are made compatible with PCI Express will be described.

  As data to be transferred on the I / O bus system 1, data transferred from the database application (using the TCP port number 54321) operating on the CPU 101a to the I / O device 103 (device ID = 3) is considered.

  First, the data from the CPU 101a is assigned a VLAN ID in the group ID setting unit 113a according to the procedure of FIG. Specifically, the TCP port number of the data is monitored (step S111), and as a result, 54321 is acquired as the TCP port number. It is searched whether there is a VLAN ID whose TCP port number matches the group condition in the group management table (step S112), and as a result, VLAN ID = 11 is determined. The Ethernet processing unit 254 assigns VLAN ID = 11 to the data (step S113), encapsulates it in an Ethernet frame, and transfers the data. At this time, the MAC address corresponding to the device ID is searched and assigned as the destination address. If it does not match the group condition, the group ID is not assigned and the data is transferred as it is (step S114).

  Then, the traffic monitor unit 140 monitors traffic for each VLAN. From the monitored traffic information, traffic can be controlled for each group using a known method.

  The traffic monitor unit 140 measures the band of VLAN ID = 11. As a result, it is assumed that the bandwidth is measured to be 40 Mbps. Next, the measured bandwidth is compared with a reference value to determine whether access is concentrated. If the reference value is satisfied, it is determined that “no bottleneck”. When the reference value is not satisfied, the group condition of VLAN ID = 11 is referred to, “database software is determined to be a bottleneck”, and the bandwidth is controlled.

  In the above description, traffic is constantly monitored regardless of whether there is performance degradation. However, if the performance of a specific application or virtual machine deteriorates while the application or virtual machine is in use, the VLAN associated with that application or virtual machine is used. Is set as the measurement target in the traffic monitor unit 140, and the detailed cause of the performance deterioration can be specified. For example, when the scientific calculation software uses two storage devices with device IDs = 2 and 3, VLAN ID = 51 and 52 are set for each device, and the traffic of each VLAN is monitored. Thereby, it can be specified that the performance of a device belonging to a VLAN having an abnormality in traffic is deteriorated.

  Note that it is possible to decide whether to select to increase or reduce the bandwidth according to the load status or the management policy.

  In the above description, the bandwidth is measured as the traffic information. However, in the present invention, the traffic information is not limited to the bandwidth, and the packet loss rate or the data size can be measured and used as a performance index.

  For example, the traffic monitor unit 140 measures the packet loss rate and increases the allocated bandwidth for the VLAN exceeding the reference value, thereby improving the performance of the application allocated to the VLAN.

Further, by assigning a VLAN to the NACK command and monitoring the VLAN bandwidth by a traffic monitor, it is possible to identify a device that has issued many NACKs, thereby identifying a device in an error state. In addition, when the bandwidth allocated to NACK exceeds a certain value, the data contents are monitored to obtain more detailed error information, so that detailed information can be monitored in the event of an abnormality, and the monitor load during normal operation Can be reduced.
In the above description, PCI Express data is encapsulated in Ethernet, and Ethernet VLAN is used as a group ID. However, encapsulation in Ethernet is omitted, and the group ID is used as the PCI Express header or data in the PCI Express data. It may be assigned to a specific location, and the traffic monitoring means may monitor that location and monitor the traffic for each group.
In the above description, PCI Express is described as an example of a protocol supported by the CPU and the I / O device, and Ethernet is described as an example of a protocol supported by the switch. However, the feature of the present invention is not limited to a specific protocol. Even if other protocols are used, the effects of the present invention can be obtained.

  In addition, the data monitoring unit has been described with respect to the case where the PCI Express header information is monitored as the data content. However, since the feature of the present invention is not limited to a specific protocol, the header information of other protocols is monitored. In addition, the present invention is applicable.

  In the above description, storage has been described as an example of an I / O device. However, the I / O device is not limited to storage, and other I / O devices include a network card, a display, a keyboard, and a mouse. Even if a CD drive, a speaker, or the like is used, the effects of the present invention can be obtained.

  Further, in the above description, an example in which three CPUs and three I / O devices are provided has been described. However, the number of I / O devices is not limited to this number. For example, a plurality of I / O devices are provided for one CPU. The present invention can also be applied to a case where one I / O device is provided for a plurality of CPUs.

  Next, effects of the first exemplary embodiment of the present invention will be described. In the first embodiment of the present invention, a VLAN is assigned to each application or virtual machine, and the bandwidth for each VLAN is measured. Therefore, the performance in units grouped for each application or virtual machine can be measured.

In addition, since the performance measurement for each application and virtual machine is also performed by the VLAN traffic monitor, the performance can be measured without increasing the load on the CPU.
In addition, since the VLAN traffic monitor is implemented by hardware processing on the switch, the measurement accuracy can be improved.

[Second Embodiment]
Next, a second embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 6 is a block diagram showing the configuration of the second exemplary embodiment of the present invention.
Referring to FIG. 6, the I / O bus system 2 according to the second embodiment of the present invention has a configuration including a resource allocation unit 210 in addition to the configuration of the first embodiment shown in FIG. is there. Since the configuration other than the resource allocation unit 210 shown in FIG. 6 is the same as that of FIG. 1, the description thereof is omitted here.

  The resource allocation unit 210 allocates storage or a network interface card (NIC) to an application or a virtual machine, and performs resource allocation settings. Specific implementation methods of resource allocation setting include, for example, switching by a switch and setting change of a virtual machine monitor (VMM: Virtual Machine Monitor). The virtual machine monitor is software for providing a virtual machine, such as Vmware or Xen.

  Next, the operation will be described by taking as an example a case where the allocation status of virtual machines and storage and the monitoring result of the bandwidth for each virtual machine are as shown in FIG. As shown in FIG. 8A, the virtual machine 1 (virtual machine ID = 1) and the virtual machine 2 (virtual machine ID = 2) that consume a large amount of bandwidth use the same storage device (device ID = 2). I understand that. Assuming that the storage performance of device ID = 2 and device ID = 3 is the same, in order to improve the performance and reliability of the entire system, it is better to make the access frequency of each storage uniform.

Therefore, the resource allocation unit 210 calculates resource allocation so as to equalize the bandwidth, and changes the resource allocation. As a result, an allocation state as shown in FIG. As can be seen from this result, the access bandwidth to the two storages is made uniform, and the performance and reliability can be improved. As a method of changing the allocation of storage used by a virtual machine, there is a method of moving an image file in which a virtual machine is activated.
As shown in FIG. 7, the resource allocation unit 210 includes a resource allocation calculation unit 211 and a resource setting unit 212. Based on the traffic information provided from the traffic monitor unit, the source allocation calculation unit 211 distributes the resource amount so that the performance of each virtual machine is equalized. For example, when virtual machine image files that use a large amount of storage bandwidth as resources exist on the same storage, storage access becomes a bottleneck. In order to solve this problem, the arrangement of the virtual machine image file is calculated so that the storage access bandwidth is made uniform. The resource setting unit 212 sets the virtual machine monitor in order to allocate the resource calculated by the resource allocation calculation unit 211 to the virtual machine. In the example of FIG. 8, the image file of the virtual machine 2 is moved from device ID = 2 to device ID = 3, and the image file of the virtual machine 3 is moved from device ID = 3 to device ID = 2.

  In this embodiment, the resource amount is allocated based on the traffic information provided from the traffic monitor unit, but based on the performance index created by the performance index conversion unit of the third embodiment to be described later, The resource amount may be allocated.

  In the second embodiment of the present invention, since the control target is expanded from traffic to resource allocation status, in addition to the effects of the first embodiment, it is possible to realize performance improvement in the entire system.

[Third Embodiment]
Referring to FIG. 9, the I / O bus system 3 of the present embodiment includes a group ID setting unit 113a-113c connected to each of the CPUs 101a-101c, and a group ID setting unit connected to each of the I / O devices 102-104. 113d-113f and a switch 280 connected to the group ID setting unit 113a-113f. Further, the I / O bus system 3 includes a group traffic management unit 211 connected to the group ID setting units 113a to 113f and the switch 280, a group management table connected to the group ID setting units 113a to 113f and the group traffic management unit 211. 212, a group management table 212, and a performance management unit 220 connected to the CPUs 101a to 101c.
In FIG. 9, the group traffic management unit 211 only shows connection with the group ID setting units 113c and 113f, but the group traffic management unit 211 is also connected with other group ID setting units 113a, 113b, 113d, and 113e. Similarly, the group management table 212 only shows connection with the group ID setting units 113c and 113f, but the group management table 212 is also connected with other group ID setting units 113a, 113b, 113d, and 113e.

9 shows only the connection between the performance management unit 220 and the CPU 101 with the CPU 101c, the performance management unit 220 is also connected to the other CPUs 101a and 101b.
The I / O bus system 3 is connected to the CPUs 101a to 101c and the I / O devices 102 to 104. The group traffic management unit 211 allocates a band to each group according to the priority of the group registered in the group management table 212. The group management table 212 holds conditions for grouping. The group ID setting units 113a to 113f monitor the contents of the transfer data, determine a group ID to be assigned based on the grouping conditions registered in the group management table 212, and assign the group ID to the data.

  In the first embodiment, the case has been described in which the group ID setting unit holds the conditions for grouping. However, in the present embodiment, the group management table 212 registers the conditions for grouping, and the group ID is set. The setting unit assigns a group ID under the group condition registered in the group management table.

  The configuration example of the group ID setting units 113a to 113f is the same as the configuration of FIG.

  The performance management unit 220 converts the traffic information monitored by the traffic monitoring unit 240 into a performance index corresponding to the type of group target, determines whether the performance index satisfies a predetermined reference value, and if not satisfied Notifies the location as a bottleneck. Examples of types of groups include virtual machines provided by applications, devices, data transfer methods such as PIO (Programmed IO) and DMA (Direct Memory Access), and virtualization technology.

  The traffic control unit 230 controls traffic for each group. The traffic monitor unit 240 monitors traffic for each group. As an implementation method for controlling and monitoring the traffic for each group, for example, there is a method using Ethernet VLAN (Virtual LAN).

  Next, a detailed configuration of the performance management unit 220 will be described.

  FIG. 10 is a block diagram illustrating a configuration example of the performance management unit 220. 10, the performance management unit stage 220 includes a performance management distribution unit 221, an application performance management unit 222, a device performance management unit 223, a data transfer method performance management unit 224, a command performance management unit 225, and a virtual machine performance management unit 226. Consists of

  The performance management distribution unit 221 selects a performance management method as an application performance management unit 222, a device performance management unit 223, a data transfer method performance management unit 224, a command performance management unit 225, or a virtual machine according to the performance monitoring target. Select from the performance management unit 226.

The application performance management unit 222 generates a performance index from traffic when an application is targeted, and determines whether the generated performance index is within a specified range.
FIG. 11 is a block diagram showing the configuration of the application performance management unit. As shown in FIG. 11, the application performance management unit 222 includes a performance index conversion unit 261 that creates a performance index from monitored traffic, and a performance index determination unit that determines whether the created performance index is within a specified range. 262.

  The device performance management unit 223, the data transfer method performance management unit 224, the command performance management unit 225, and the virtual machine performance management unit 226 each have a performance management target of an I / O device, a data transfer method, a command, and a virtual machine. Then, a performance index is generated from the band according to each target, and it is determined whether the generated performance index is within a specified range.

  The above performance management target is an example shown for explanation, and a new management unit can be added according to the target, or a management unit that is not a management target can be omitted.

  Next, the operation of the I / O bus system 3 of FIG. 9 will be described with reference to the flowcharts of FIGS.

  FIG. 12 is a flowchart showing an operation in which the I / O bus system 3 assigns a group ID.

  Each of the group ID setting units 113a to 113f monitors data transferred to the I / O bus (step S101) and refers to the group management table 112 to determine whether the data content matches the group condition (step S102). ), If it matches the group condition, the group ID of the matched group is assigned to the data (step S103). If it does not match the group condition in step S102, the group ID is not assigned and the data is transferred as it is (step S104).

  FIG. 13 is a flowchart showing an operation in which the I / O bus system 3 finds a performance bottleneck.

  The traffic monitor unit 240 monitors traffic for each group (step S201). The performance index conversion unit 261 creates a performance index from the monitored traffic (step S202). In this conversion, for example, the number of processes can be obtained as a performance index by dividing the traffic by the average data length for one process. The performance index determination unit 262 determines whether the created performance index is within a specified range (step S203). If the performance index is within the specified range, the performance index determination unit 262 notifies the administrator that there is no bottleneck (step S204). If the performance index is out of the specified range, the performance index determining unit 262 refers to the group management table 212 to identify the bottleneck location from the group condition (step S205), and identify the bottleneck location. The administrator is notified (step S206).

  FIG. 14 is a flowchart showing an operation in which the I / O bus system 3 controls the performance index.

The performance index conversion unit 261 included in the performance management unit converts the requested performance index into a band (step S301). The group traffic management unit 211 included in the group management unit determines a band to be given to each group based on the priority for each group registered in the group management table 212 and the band requested in step S301 (step S302). The traffic control unit 230 sets the bandwidth determined in step S302 for each group (step S303). The group management unit includes a group ID setting unit 113a-113f, a group traffic management unit 211, and a group management table 212.
Next, a configuration example in the case where the CPUs 101a to 101c and the I / O devices 102 to 104 are made compatible with PCI Express will be described.

  A group ID setting unit 113a-113f and a switch 280 are connected by the Ethernet protocol, and a configuration of an I / O bus system 3 that uses an Ethernet VLAN (Virtual LAN) as a unit for realizing a group is illustrated.

  The CPUs 101a to 101c and the I / O devices 102 to 104 have PCI Express addresses (bus numbers and device numbers), and the group ID setting units 113a to 1213f have Ethernet MAC addresses. The correspondence between these two addresses can be acquired by encapsulating the PCI Express address in an Ethernet frame and transmitting it at the time of initial setting or the like.

  The switch 280 is an Ethernet switch that can monitor traffic for each port and VLAN.

  The traffic control unit 230 and the traffic monitor unit 240 are configured using a traffic control function and a traffic monitor for each VLAN included in the switch 280. The traffic control function and traffic monitor (traffic monitor unit 240 and traffic control unit 230) included in the switch 280 can perform high-speed processing because it performs hardware processing instead of software processing.

  Traffic control targets include, for example, bandwidth guarantee, priority control, and latency control for each switch port and VLAN. Traffic monitoring targets include, for example, bandwidth for each switch port and VLAN, packet loss rate, and the like. There is a packet size. In addition, as a traffic control, the switch transfer method can be switched between a cut-through method and a store-and-forward method.

  The configuration of the group ID setting unit 113a is the same as the configuration already described in the first embodiment with reference to FIG.

The contents of the group management table 212 are the same as those already described with reference to FIG.
The performance management unit 220 holds a management table representing a management target according to the performance monitoring target. The held management table is shown in FIG.
FIG. 15A shows an application management table held by the application performance management unit. In this table, correspondence between an identifier for identifying an application and the application is registered. Examples of the application identifier include a TCP port number and a process ID on the OS. Referring to FIG. 15A, it can be seen that the TCP port number = 54321 is the database software and the TCP port number = 56789 is the scientific calculation software as the application identifier.

  FIG. 15B is a device management table, which is held by the device performance management unit. In this table, correspondence between identifiers for specifying devices and device types is registered. For example, as an identifier for identifying a device, there are a bus ID and a device ID in PCI Express. In addition, a device name or the like can be registered. FIG. 15B shows that the device ID = 2 is connected to the group ID setting unit having the network address 01: 12: 23: 34: 45: 56, and the device type is storage.

  FIG. 15C is a data transfer method management table held by the data transfer method performance management unit. In this table, the type of data transfer method is registered. FIG. 15C shows that there are PIO and DMA as data transfer methods.

  FIG. 15D shows a command management table held by the command performance management unit. In this table, command types (command identifiers) are registered. As can be seen from FIG. 15D, examples of command types include ACK, NACK, Read, and Write.

  FIG. 15E is a virtual machine management table held by the virtual machine performance management unit. In this table, virtual machine IDs and allocated resources are registered. In FIG. 15E, a CPU (device ID = 1), a storage (device ID = 2), and a NIC (Network Interface Card) (device ID = 4) are allocated to the virtual machine ID = 1 as allocation resources. You can see that

  By referring to the table contents held by the performance management unit 220 described above, an identifier representing the data contents and a name that can be easily understood by the administrator can be converted.

  Next, the operation of the embodiment will be described.

  As data to be transferred on the I / O bus system 3, consider data transferred from a database application (using TCP port number 54321) running on the CPU 101a to the I / O device 103 (device ID = 3).

  First, data from the CPU 101a is assigned a VLAN ID in the group ID setting unit 113a according to the procedure of FIG. Specifically, the TCP port number of the data is monitored (step S101), and as a result, 54321 is acquired as the TCP port number. Whether or not there is a VLAN ID whose TCP port number matches the group condition in the group management table is searched (S102), and as a result, VLAN ID = 11 is determined. The Ethernet processing unit 254 assigns VLAN ID = 11 to the data, encapsulates it in an Ethernet frame, and transfers the data. At this time, the MAC address corresponding to the device ID is searched and assigned as the destination address.

  The data transferred to the switch 280 is switched in the switch 280 to the group ID setting unit 113e which is a destination address. At this time, the traffic monitor unit 240 in the switch 280 monitors the traffic for each VLAN. The performance index is evaluated from the monitored traffic information according to the procedure shown in FIG. The traffic monitor unit 240 measures the band of VLAN ID = 11 (S201). As a result, it is assumed that the bandwidth is measured to be 40 Mbps. Next, the performance management unit converts the measured bandwidth into a performance index (S202). For example, by dividing this bandwidth by the average data size per process, the number of processes can be estimated as a performance index. For example, if the average data size per process is 10 kbytes, it can be converted into an application performance index when 500 processes are performed per second. Next, it is determined whether the access is concentrated compared with the reference value (S203). If the reference value for the number of processing cases per second is 300 to 700, it satisfies the reference value, so that “no bottleneck” is determined (S204). When the reference value is 100 to 400, the reference value is not satisfied, so the group condition of VLAN ID = 11 is referred to and “database software is determined to be a bottleneck” (S205, S206). Further, the performance index can be converted as the processing time of the I / O device by dividing the time required for the processing by the number of processing cases obtained above.

  In the above description, traffic is constantly monitored regardless of whether there is performance degradation. However, if the performance of a specific application or virtual machine deteriorates while the application or virtual machine is in use, the VLAN associated with that application or virtual machine is used. Is set as the measurement target in the traffic monitor unit 240, and the detailed cause of the performance degradation can be specified. For example, when the scientific calculation software uses two storage devices with device IDs = 2 and 3, VLAN ID = 51 and 52 are set for each device, and the traffic of each VLAN is monitored. Thereby, it can be specified that the performance of a device belonging to a VLAN having an abnormality in traffic is deteriorated.

  In addition, the traffic control unit 230 can control the performance index by controlling the traffic according to the performance index control procedure of FIG.

  For example, since the number of processing cases 500 in the above example is too large, the performance management unit 220 converts the number of processing cases 400 into a bandwidth of 32 Mbps when it is desired to suppress the processing to 400 cases per second (S301). Next, the group traffic management unit 211 adjusts the calculated bandwidth with other VLAN bandwidths to be allocated (S302). Finally, the traffic control unit 230 sets the bandwidth of VLAN ID = 11 to 32 Mbps. In the description here, the bandwidth is reduced, but it is possible to decide whether to increase or reduce the bandwidth according to the load status or the management policy.

  In the above description, the bandwidth is measured as the traffic information. However, in the present invention, the traffic information is not limited to the bandwidth, and the packet loss rate or the data size can be measured and used as a performance index.

  For example, the traffic monitor unit 230 measures the packet loss rate, and the traffic control unit 240 increases the allocated bandwidth for the VLAN exceeding the reference value, thereby improving the performance of the application allocated to the VLAN. it can.

  Further, by assigning a VLAN to the NACK command and monitoring the VLAN bandwidth by a traffic monitor, it is possible to identify a device that has issued many NACKs, thereby identifying a device in an error state. In addition, when the bandwidth allocated to NACK exceeds a certain value, the data contents are monitored to obtain more detailed error information, so that detailed information can be monitored in the event of an abnormality, and the monitor load during normal operation Can be reduced.

  In the above description, PCI Express data is encapsulated in Ethernet, and Ethernet VLAN is used as a group ID. However, encapsulation in Ethernet is omitted, and the group ID is used as the PCI Express header or data in the PCI Express data. It may be assigned to a specific location, and the traffic monitor unit may monitor that location and monitor the traffic for each group.

In the above description, PCI Express is described as an example of a protocol supported by the CPU and the I / O device, and Ethernet is described as an example of a protocol supported by the switch. However, the feature of the present invention is not limited to a specific protocol. Even if other protocols are used, the effects of the present invention can be obtained.
In addition, the data monitoring unit has been described with respect to the case where the PCI Express header information is monitored as the data content. However, since the feature of the present invention is not limited to a specific protocol, the header information of other protocols is monitored. In addition, the present invention is applicable.

  In the above description, the example of one switch has been described. However, in the embodiment of the present invention, the number of switches is not limited to one, and this is a case where a plurality of switches constitute a network. However, the effect of the present invention can be obtained.

In the above description, storage has been described as an example of an I / O device. However, the I / O device is not limited to storage, and other I / O devices include a network card, a display, a keyboard, and a mouse. Even if a CD drive, a speaker, or the like is used, the effects of the present invention can be obtained.
Further, in the above description, an example in which three CPUs and three I / O devices are provided has been described. However, the number of I / O devices is not limited to this number. For example, a plurality of I / O devices are provided for one CPU. The present invention can also be applied to a case where one I / O device is provided for a plurality of CPUs.

  Next, effects of the third exemplary embodiment of the present invention will be described. In the third embodiment of the present invention, a VLAN is assigned to each application or virtual machine, and the bandwidth for each VLAN is measured. Therefore, the performance in units grouped for each application or virtual machine can be measured.

  In addition, since the performance measurement for each application and virtual machine is also performed by the VLAN traffic monitor, the performance can be measured without increasing the load on the CPU.

  In addition, since the VLAN traffic monitor is implemented by hardware processing on the switch, the measurement accuracy can be improved.

  The present invention is used for an I / O bus system capable of managing the usage status of an I / O device and a management method thereof.

It is a block diagram of a 1st embodiment of the present invention. It is a block diagram of a structure of a group ID setting part. It is a flowchart of group ID assignment. It is a block diagram of a structure of a group setting part. It is an example of a group management table. It is a block diagram of the 2nd Embodiment of this invention. It is a block diagram of a structure of a resource allocation part. It is an example of the performance information of a virtual machine. It is a block diagram of the 3rd Embodiment of this invention. It is a block diagram of a structure of a performance management part. It is a block diagram of a structure of an application performance management part. It is a flowchart of group ID assignment. It is a flowchart of performance index evaluation. It is a flowchart of performance index control. It is an example of the information which a performance management part hold | maintains. It is a figure which shows the header format of PCI Express. It is a block diagram of a conventional I / O bus system. It is a figure which shows an address conversion table.

Explanation of symbols

1-3, 930 I / O bus system 101, 920 CPU
102-104, 941-943 I / O device 110 Group management unit 111, 211 Group traffic management unit 112, 212 Group management table 113, 213 Group ID setting unit 120, 220 Performance management unit 121 Performance management distribution unit 130, 230 Traffic control unit 140, 240 Traffic monitor unit 151 Data monitor unit 152 Group ID determination unit 153 Group ID assignment unit 161 Performance index conversion unit 162 Performance index determination unit 280 Switch 910 Arbiter 911 Bus monitor unit 912 Bus band management unit 913 Bus usage right Management Department

Claims (31)

An I / O bus system for transferring data between a CPU and an I / O device,
The data to be transferred is monitored, grouped according to the contents of the monitored data, and a group ID setting means for assigning a group ID of the group to the data, and traffic for each group ID assigned by the group ID setting means An I / O bus system comprising: a traffic monitoring means for monitoring the traffic.   2. The I / O bus system according to claim 1, wherein the group ID setting means determines a group ID to be assigned in accordance with the grouping condition held in a group management table holding a grouping condition. .   The I / O bus system according to claim 1, wherein the content of the data is information included in a header of the data.   The content of the data is a TCP port number of the data, and the group ID setting unit performs grouping for each TCP port number and determines the group ID. The I / O bus system described.   The content of the data is a command identifier in the data, and the group ID setting unit performs grouping for each command identifier and determines the group ID. I / O bus system.   The content of the data is a bus number and a device number in the data, and the group ID setting unit performs grouping for each bus number and device number to determine the group ID. Item 3. The I / O bus system according to Item 1 or 2.   The group ID setting means includes a data monitoring means for monitoring the contents of the transfer data, a group ID determining means for determining a group ID to be assigned according to the grouping condition held in the group management table based on the monitored data contents, The I / O bus system according to any one of claims 1 to 6, further comprising group ID assigning means for assigning the group ID to the monitored data.   The I / O according to any one of claims 1 to 7, further comprising performance index conversion means for converting the traffic information monitored by the traffic monitoring means into a performance index corresponding to a type of group target. Bus system.   9. The I / O bus system according to claim 8, wherein the performance index corresponding to the group target is the number of processes per unit time of the application.   9. The I / O bus system according to claim 8, wherein the performance index corresponding to the group target is the number of processes per unit time of the I / O device.   9. The I / O bus system according to claim 8, wherein the performance index corresponding to the group target is a processing time of the I / O device.   The I / O bus system according to any one of claims 8 to 11, further comprising performance index determination means for determining whether the performance index converted by the performance index conversion means is a normal value. The performance index conversion means converts the requested performance into traffic,
The I / O according to any one of claims 8 to 12, further comprising group traffic management means for determining traffic to be allocated for each group based on the converted traffic and the priority of the group. Bus system.   2. The apparatus according to claim 1, further comprising a plurality of the group ID setting means, wherein the group ID setting means are connected by a switch, and the traffic monitoring means and traffic control means for controlling traffic are provided in the switch. 14. The I / O bus system according to any one of items 13.   15. The I / O bus system according to claim 14, wherein the switch is compatible with Ethernet.   The I / O bus system according to claim 15, wherein an Ethernet VLAN is used as the group.   17. The I / O bus system according to claim 1, wherein protocol conversion is performed in the group ID setting means.   18. The I / O according to claim 8, further comprising: a resource management unit that calculates resource allocation based on the performance index created by the performance index conversion unit and sets the resource allocation. O bus system.   The I / O according to any one of claims 1 to 17, further comprising resource management means for calculating resource allocation based on traffic monitored by the traffic monitoring means and setting resource allocation. Bus system. In an I / O bus system management method of an I / O bus system for transferring data between a CPU and an I / O device,
An I / O bus system management method characterized by monitoring the contents of the transferred data, grouping according to the contents of the monitored data, and monitoring the traffic in units of set groups. In an I / O bus system management method of an I / O bus system for transferring data between a CPU and an I / O device,
The contents of the data to be transferred are monitored, and the group ID of the group in which the contents of the monitored data and the group condition registered in the group management table match is given to the data, and the traffic is monitored for each group ID. An I / O bus system management method.   The I / O bus system management method according to claim 21, wherein a TCP port number of the data to be transferred is monitored, a group is assigned to each TCP port number, and the group ID is assigned.   22. The I / O bus system management method according to claim 21, wherein a command identifier in the transferred data is monitored, a group is assigned to each command identifier, and the group ID is assigned.   The I / O bus system according to claim 21, wherein a bus number and a device number in the transferred data are monitored, a group is assigned to each bus number and device number, and the group ID is assigned. Management method.   The I / O bus according to any one of claims 21 to 23, wherein after monitoring traffic for each group ID, the monitored traffic information is converted into a performance index corresponding to the type of group target. System management method.   26. The I / O bus system management method according to claim 25, wherein the performance index corresponding to the group target is the number of applications processed per unit time.   26. The I / O bus system management method according to claim 25, wherein the performance index corresponding to the group target is the number of processes per unit time of an I / O device.   26. The I / O bus system management method according to claim 25, wherein the performance index corresponding to the group target is a processing time of an I / O device.   29. One of the claims 25 to 28, wherein the monitored traffic information is converted into a performance index corresponding to the type of the group target, and then it is determined whether the performance index corresponding to the type of the group target is a normal value. The I / O bus system management method according to the item.   29. The resource allocation according to the performance index is calculated after the monitored traffic information is converted into the performance index according to the type of the group target, and the resource is set according to the resource allocation result. The I / O bus system management method according to any one of the above.   29. The I / O bus system management method according to any one of claims 25 to 28, wherein resource allocation according to the monitored traffic information is calculated, and resources are set according to the resource allocation result.

Images