CN1568467B - Just-once cache architecture - Google Patents

Abstract

A system for managing objects located in a clustered network includes a file system that contains at least one copy (214) of a data object. The system may include a plurality of cluster servers in communication with a file system (212). A boot server is selected that contains a distributed consensus algorithm for selecting a master server (206) and that utilizes multicasting in executing the algorithm loop. The selected primary server (206) may contain a copy (208) of the data object (214) in, for example, a local cache, so that access to the local copy (208) is provided to any other server in the cluster. Changes to items hosted by the primary server (206) may be updated in the file system (212). If the primary server (206) becomes unable to accommodate the object, a new primary server may be selected using a distributed consensus algorithm and the other servers (216, 218) notified of the new primary server using multicast.

Description

exactly once cache structure

This application claims priority from the following applications incorporated herein:

U.S. Provisional Patent Application No. 60/317,718 filed on September 6, 2001 by Dean Bernard Jacobs and Eric Halpern and titled "EXACTLY ONCE CACHE FRAMEWORK". The invention titled "EXACTLY ONCE CACHE FRAMEWORK" US patent application filed by Dean Bernard Jacobs and Eric Halpern on September 4, 2002.

U.S. Provisional Patent Application No. 60/317,566, titled "EXACTLY ONCE JMS COMMUNICATION," filed September 6, 2001 by Dean Bernard Jacobs and Eric Halpern.

US provisional patent application entitled "EXACTLY ONCE JMS COMMUNICATION" filed September 4, 2002 by Dean Bernard Jacobs and Eric Halpern.

Copyright Notice

The disclosure of this patent document includes material that is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document disclosed in the Patent and Trademark Office as it appears in the Patent and Trademark Office patent files or records, and otherwise, all copyright rights are reserved whatsoever.

Examples of cross-references:

The following U.S. patent applications are examples of cross-references and are hereby incorporated by reference:

U.S. Patent Application No. 60/305,986 filed on July 16, 2001 by Dean Bernard Jacobs, Reto Kramer, and Ananthan Bala Srinivasan, entitled "DATAREPLICATION PROTOCOL".

technical field

The present invention relates to techniques for distributing objects among servers in a network cluster.

Background technique

In distributed computer systems it is often the case that several servers and/or network nodes must work together. These servers and nodes must be coordinated when there is typical network information that needs to be shared among the multiple devices in order to allow them to function as a single entity. The methods by which devices can be coordinated are generally very expensive in terms of resources and efficiency.

Usually, since there are several information transfers between multiple nodes, some kind of synchronization is required in order for these nodes to be consistent. However, this synchronization requirement is undesirable in a clustered network environment. Many cluster environments simply avoid taking advantage of any such synchronization requirements. In some applications, however, such consistency is required.

In some cases where consistency is required, there is a device to which the cluster is trying to exclude access. One such device is a transaction registry file system. As long as transaction processing is in progress, certain objects are stored that must be persisted in such a way that if a failure occurs, the persisted objects can be restored.

For objects which need to be kept in one location, there is typically a transaction monitor running on each server in the cluster or domain which thereafter uses the local file system to access the object. Each server can have its own transaction manager so that there are few problems with persistence. Since each server has its own transaction manager, no coordination is required.

For example, there may be a cluster of three servers, each with a transaction manager. One of these servers may experience a failure or other problem due to the server being unavailable for the cluster. Since the failed server is the only server with access to a particular transaction record, it is not possible to recover any transactions on a particular transaction until that server is again available to the cluster. Restoring this record is difficult, or at least inefficient, since the server must spend a considerable amount of time resolving these issues. Important server problems might include events such as a shorted motherboard on the server or a burned out power supply.

Contents of the invention

The present invention includes a system for managing objects such as stored in servers on a network or cluster. The system includes a data source, application, or service such as a file system or a Java message service component located within or outside the cluster. The system includes several servers that communicate with the file system or applications, such as through a high-speed network connection.

The system includes a lead server such as agreed upon by another server. The boot server is contained in a hardware or software cluster. The system includes an algorithm for selecting a bootstrap server from among the servers, which may be, for example, an algorithm built into a hardware cluster device. The bootstrap server in turn includes a distributed consensus algorithm, such as the Paxos algorithm, for selecting the master server. The algorithm used to select the boot server may be different or the same as the algorithm used to select the master server.

The master server includes copies of items or objects, such as stored in a local cache. The master server provides a local copy that is accessible to any server on the network or in the cluster. The master server may also provide a unique access point to objects stored in the file system, or may provide a unique access point to applications or services. Any changes made to items cached, admitted, or owned by the master server may also be updated in the file system, application, or service.

If the primary server becomes unacceptable for the object, a new host is selected using a distributed consensus algorithm, after which the data required for the object is fetched from the file system or service. Another server in the cluster is notified that a new server is accepting the object. The server is notified by suitable means such as a point-to-point connection or by multicasting.

Description of drawings

Fig. 1 has provided the schematic diagram of the distributed object system according to an embodiment of the present invention;

Fig. 2 has provided the schematic diagram of another distributed object system according to an embodiment of the present invention;

Fig. 3 has provided the flowchart of a method for selecting master server according to the present invention;

Fig. 4 has provided the flowchart of a method for selecting new main server according to the present invention;

Figure 5 presents a flowchart of a method for using a bootstrap server according to the present invention;

Fig. 6 has provided the schematic diagram of the JMS message storage system according to an embodiment of the present invention;

Figure 7 presents a block diagram of components of a computer system used in accordance with the present invention.

Detailed ways

The system according to the invention provides solutions such as publishing availability when the server owning the data object becomes unavailable to the server cluster. Such a solution would allow another server in the cluster to take ownership of the data object. However, a problem arises in that data objects can be accessed on both servers without duplication of the data objects on the two servers.

If the cluster uses a file system, data store, or database (collectively referred to hereinafter as a "file system") to persistently store data, and more than one server has access to the file system, then if the first server that owns that object encounters problem, the second server can automatically take over the task of accessing the data object. Additionally, the cluster or an algorithm used by the servers in the cluster may be utilized to instruct the server to take over ownership of the item. However, another fundamental problem involves getting the cluster to agree on which server currently owns the resource or object, or achieving "unanimity" among the servers.

FIG. 1 shows an example of a cluster system 100 in which objects such as a transaction registry 114 are stored in a file system 112 in accordance with the present invention. All servers 106, 116, 118 in the cluster 110 have access to the file system 112, but only one of these servers has access to the registry 114 at a time. The master server 106 among the servers in the cluster 110 will "own" or "host" the registry 114 , such as by storing a copy 108 of the registry 114 or by making the registry 114 accessible in a file system 112 . Any other server 116 , 118 in the cluster 110 can access the copy 108 of the record, and/or can access the registry 114 through the master server 106 . For example, a client or browser 102 may make a request to the network 104 referring to a server 116 in the cluster 110 . The server can access a copy 108 of the transaction record on the primary server 106 via the network 104 . If the transaction record must be updated, the copy 108 is updated along with the original registry 114 on the file system 112 .

For example when a server acts as the object's store, such as by storing a copy of the data object in a local cache and making that copy available to other servers in the cluster, or by making the unique server randomly accessible in a file system The server then "owns" or "accommodates" the data objects such that all other servers in the cluster must go through the master server to access those objects. This guarantees that the object exists "exactly once" in the server cluster.

Figure 3 illustrates a process 300 for establishing admission to a subject. A master server is selected using a distributed consensus algorithm 302 such as the Paxos algorithm. Because the servers in a cluster typically must come to general agreement or consensus on how to distribute objects among the cluster servers, this algorithm is called a "distributed consensus" algorithm.

If the admitted object is cached in the host server, for example, then a copy of the data object fetched from the file system is transferred to the host server and stored as an object in the local cache 304 . Other servers in the network or appropriate cluster are thereafter notified, such as by the master server, that a local copy of the object exists in the master server and that the local copy will be used in processing 306 for future network requests.

In the Paxos algorithm, which is an example of a distributed consensus algorithm, a server is selected as a master or lead server by a network server that leads a series of "consensus rounds". In each consensus cycle, a new master or bootstrap server is proposed. The cycle continues until a majority or quorum of servers accepts the proposed server. Although the system is configured so that a round-robin is always initiated by the boot server to select a master server, any server can propose a master or boot server by starting a round-robin. Cycles for different selections can be performed at the same time. Thus, a round selection is identified by a round number or a pair of values, one of which is associated with said round and the other is associated with the server directing said round.

The steps for such a cycle are as follows, although other steps and/or methods may be suitable for certain situations or applications. First, a cycle is started by instructing the server to send a "collection" message to the other servers in the cluster. Aggregated messages aggregate information from servers in the cluster about previous rounds in which those servers participated. The centralized message also informs the server not to submit selections from previous cycles if there was a previous consistent cycle for a particular selection process. For example, once the bootstrap server has aggregated responses from at least half of the cluster servers, the bootstrap server can decide on the value to suggest the next round and transmit that suggestion to the cluster servers as a "start" message. In order for the guidance server to select a value to provide a suggestion in this method, initial value information must be received from the server.

Once the server receives the start message from the boot server, it responds by transmitting an "accept" message, which indicates that the server accepts the proposed master/boot server. If the boot server receives an accept message from a majority or quorum of servers, the boot server sets its output value to the value suggested in the round robin. If the bootstrap server does not receive a majority or quorum of acceptances ("consensus") within a specified period of time, the bootstrap server may start a new round. If the bootstrap server receives a consensus, the bootstrap server may notify the cluster or web server that the server will be designated as the selected server. The notification may be broadcast to the web server via any suitable broadcast technique, such as via a point-to-point connection or multicast.

The consensus condition of the consensus method can be guaranteed by suggesting options utilizing information about previous cycles. This information is required to come from at least a majority of web servers so that for any two rounds there is at least one server participating in both rounds.

The bootstrap server may select a value by asking each server the latest cycle number that the server accepted a value, and possibly the accepted value. Once the bootstrap server has obtained this information from a majority or quorum of servers, it can choose a value for the new round equal to the value of the latest round in the response. The bootstrap server can also choose an initial value if none of the servers were involved in the previous cycle. For example, if the bootstrap server receives a response that the last accepted cycle was x, and the current cycle is y, then the server says not to accept any cycles between x and y in order to maintain consistency.

A sample interaction between the loop boot server and the web server includes the following messages:

(1) "Collect" - sends a message to the server which is starting a new cycle "r". The message may take the form m=("Collect", r).

(2) "Last" - sends a message to the bootstrap server from a web server providing the "a" accepted for the last round and the value "v" for this round. The message may take the form m=("Last", r, a, v).

(3) "Begin" - send a message to the server publishing the value associated with cycle r. This message may take the form m=("Begin", r, v).

(4) "Accept" - transmits a message from the server accepting the value associated with cycle r to the bootstrap server. The message may take the form m=("Accept", r).

(5) "Success" - transmits a message to the server for publishing the selection of said value v in relation to the cycle r. This information may take the form m=("Success", r, v).

(6) "Ack" - transmits a message from a server to the bootstrap server acknowledging that a decision related to round r has been received. This information may take the form m=("Ack", r).

There is a file system separate from its servers inside or outside the hardware cluster or software cluster. The file system persistently stores transaction records, such as by storing the records on a first disk and copying the records to a second disk located within the file system. If the first disk gets scratched, the file system can make the cluster and/or server invisible to the scratch and can obtain record information from the second disk. The file system also has the option to copy the record to a third disk which it uses as a backup of the second disk.

From the perspective of the servers in the cluster, the file system can be a single resource. In one embodiment, the server may only be concerned that a single server owns the file system at any one time.

Another example in accordance with the invention includes a cache located in a server cluster. For example, for network performance reasons, it is therefore desirable in a cluster environment to have a single cache server in the cluster show data objects. It may be advantageous to keep multiple entries in a single cache, because servers in the cluster can access the cache without having to continually go back to persistent storage. Fetching multiple items that are already in memory can greatly increase the utilization of the system because of the relatively intensive time to hit the database or file system.

However, one problem with a single cache is that it must be guaranteed that the objects stored in memory are the same as those stored in the file system-disk. The reason for this consistency is to ensure that any operation or calculation performed on the cached object produces the correct result. Another reason is that a file system's cache must be repaired due to cache scratches or otherwise becoming infected or unavailable.

There are at least two basic ways to handle this type of caching in a cluster, although others work for at least some applications. One way is to duplicate the cache in multiple places. This process is problematic because any change to the item being cached requires that all servers copying the cache agree on the change, or at least know of the change. This has proven to be very expensive in terms of resources as well as performance.

An alternative method according to the invention distributes specific servers so that they are the owners of the caches in the cluster and the caches are all accessible by these specific servers. Any server in the cluster hosts this cache. Each server can host one, several, or no caches. Host the cache on a single server, or spread it across some or all servers in the cluster. The cluster itself may be any suitable cluster, such as a hardware cluster or a group of servers specified by a software application residing in a given "software" cluster.

Also consider a transaction log and/or cache as an example of an object located somewhere on the system. It can be counted on to guarantee that any such object exists only once in a cluster, and that the object is always available. It can also be counted on to guarantee that if the server hosting the object fails, the object can be restored to another server and made available to the cluster.

Figure 4 shows one method 400 for recovery. In this method, it is determined whether the host server can continue to accommodate objects 402, eg, whether the server is still available for the network. If not, a new host is selected using a distributed consensus algorithm. This selection may be performed according to the method used to select the original host 404 . A copy of the data object fetched from the file system is provided to the new host and stored in local cache 406 . Other servers on the network or in the appropriate cluster are notified that the new master server contains a local copy of the object, and that the local copy will be used in processing any future network requests 408 .

Systems and methods according to the present invention can define objects that exist in exactly one location in a cluster, and can guarantee that these objects always exist. From the server's point of view, it does not matter whether objects such as transaction records are mirrored or copied using, for example, a file system. From the server's point of view, there is always a persistent store accessible by any server in the cluster, and the system periodically checks for the existence of objects, or assigns object ownership to short periods of time for frequent reassignment object to ensure existence on the network or on a device in the cluster.

A hardware cluster consists of a collection of devices, each of which can run multiple servers. There is also a file system behind each device. Servers in a hardware cluster are usually made of hardware, so they can make faster decisions and handle server failures within a hardware cluster. The size of the hardware cluster is limited to the actual hardware of the devices comprising the servers. Servers in a hardware cluster can be used as servers in a software cluster, and since individual servers on the device can be used for the network, network servers can also be included.

A shared file system for one of these devices is available to all servers in the cluster, such as over a high-speed network. File systems can also be redundant. In one embodiment, this redundancy is achieved by using multiple data disks of the file system. In such a redundant implementation, objects can be copied across multiple disks anytime they are written to the file system. When viewed as a "black box," the file system can withstand a failure in any one of the disks and still provide access to data items by any server in the cluster.

Assuming that these objects kept in memory are always restored by a reliable, persistent storage mechanism, it is possible to build what is referred to as an "exactly-once" structure according to the present invention. For example, there exists an object representing transaction records. As long as a call to the object is generated, the corresponding event record is updated. This includes a call to read from or write to the database. The object representing the transaction record resides on a server in a cluster such as the master server. As long as at least one server in the cluster is up and running, the exactly-once structure ensures that if another server fails, that server can take over ownership of the record.

There may be an object representing a cache. Whenever the cache is updated, the update can also be written back to persistent storage. When a server among multiple servers needs to use a data item, the server is required to go through the object. If the server hosting the object representing the cache fails, the object is restored on another server. Restored objects have all necessary information retrieved from persistent storage.

Just-once structures are available as memory buffers used by the cluster. This architecture provides a single cache which can provide data in the system which is retrieved from a reliable, persistent memory. Whenever data is read from the cache, the read can be performed without accessing permanent memory. However, when the update is written into the cache, it must be written back through persistent storage in order to allow the system to recover if there is a failure.

An important aspect of the exactly-once architecture includes a way in which it can abstract a method that varies by application and/or implementation. New types of distributed objects such as those called "exactly-once objects" were created. An exactly-once object is, for example, a locally cached copy of a data item in a file system, or the only point of access to that data item for a server in a cluster. The underlying technology that implements that abstraction is also important.

The system of the present invention can utilize any of a number of methods it can use for distributed consensus, such as the one utilizing the Paxos algorithm described above. Choose an algorithm that provides an efficient way to get multiple nodes and/or distributed nodes to agree on a value for an object. This algorithm can be chosen to work even if a node is down and/or returns during the negotiation process.

The general approach of network clusters is to use reliable broadcast, in which every message is guaranteed to be delivered to its intended recipient, or at least to every intended server. This approach makes it difficult to parallelize the system because reliable broadcasting requires the receiver to acknowledge receipt of a message before moving on to the next message or receiver. A distributed algorithm using multicast reduces the number of guarantees, because multicast cannot guarantee that all servers receive a message. However, multicasting simplifies the approach so that the system can engage in parallel processing, since a single message is multicasted concurrently to all cluster servers without waiting for a response from each server. A server that does not receive a multicast message can eventually fetch the information from the bootstrap server or another cluster server or web server. As used herein, a web server may refer to any server on the network, whether the web server is in a hardware cluster, in a software cluster, or outside of any cluster.

An important aspect of the exactly-once structure is that it reduces the difficulty of consistency. According to the present invention, the performance of a distributed consensus implementation can be improved by utilizing a distributed consensus algorithm to multicast messages. The method may minimize the message exchange and/or network traffic required for all servers to agree with each other.

When multicasting, one of several methods may be employed. In the first method, which is called "single-phase distribution", the directed server multicasts a message to all other servers located on the network, for example for use in the loop of the Paxos algorithm or for such a state , a new host has been selected for an object. In this method, the master server only needs to send a message, which is sent to any available server in the network. If a server is temporarily disconnected from the network, the server is required to identify new hosts after returning to the network.

Using another multicast method called "bi-phase distribution", the lead server can pre-select a master server using an appropriate algorithm. However, before distributing an object to the host, the bootstrap server can connect with all other servers in the cluster to determine if the servers agree with the selection of the new master. The bootstrap server can connect to each server through a point-to-point connection, or can transmit a multicast request and thereafter wait for each server to reply. If the server disagrees with the chosen host, the bootstrap server utilizes this algorithm to pre-select a new host. The bootstrap server thereafter transmits another multicast request and the identity of the re-preselected host in another cycle.

If each server agrees on the pre-selected host, the bootstrap server distributes the object to the master. The server is then directed to multicast a commit message to inform the server that the new changes have taken effect and that the server will therefore update its information.

Exactly one-time structures can also make use of "rental" agencies. In utilizing this mechanism, an algorithm is used to bring the cluster server into agreement with the bootstrap server, for example by utilizing a distributed consensus algorithm. Once selected, the bootstrap server is responsible for distributing the object to the various servers in the cluster exactly once. The system is set up so that if the existing master server fails, the cluster server is always identical to the new boot server.

At the same time as the bootstrap server is activated, the bootstrap server may know all exactly-once objects that need to exist in the system. The bootstrap server can decide which server will host each object, and can therefore "lease" the object to the selected server. When an object is leased to a server, the server may own or host the object for a certain period of time, such as the lease period. Configure the bootstrap server to update these leases periodically. This method can provide a way to ensure that if a server fails or is disconnected in any way or is otherwise unable to operate normally within the cluster, then the server will not have its lease renewed.

The biggest problem with distributed systems in the event of a failure is that it is difficult to tell the difference between a server with a failure and just one that is not responding. Any server that is disconnected from the network for any reason no longer accepts objects. Even though the server is not available for the cluster, it is still known that it will finish hosting all objects after the lease time. When the server is no longer available for the cluster, its lease will not be renewed.

The bootstrap server also knows that if the bootstrap server cannot reach the master server within a certain amount of time, the master server will relinquish ownership of the object. The rental period may be any suitable time, such as a few seconds. The lease period is the same for all objects in the cluster, or can vary between objects.

The system can also be more compact using exactly one structure. Operating systems often provide specific mechanisms that are closer to the hardware and provide more control. However, a problem with this approach is that it is limited by the available hardware. For example, a hardware cluster of servers has 16 servers in sequence. Because these systems require some kind of tight hardware coupling, there is a limit to the number of servers that can be included in a cluster.

On the other hand, exactly one fabric can handle many more clusters than a dedicated hardware cluster can handle. This structure allows some leverage of the quality of service obtained from a dedicated cluster, allowing for larger clusters. Different qualities of service may include, for example, whether messages are delivered over a reliable protocol such as a point-to-point connection, or a less reliable but more resource-friendly protocol such as multicast. An advantage of using an exactly-once structure is the ability to balance scalability and fault tolerance so that the user can adapt the system to the needs of a particular application.

Existing systems such as hardware cluster mechanisms attempt a high-availability solution by having a single mechanism (to be provided to the cluster) supported by a second mechanism. If the first mechanism fails, there is a "buddy" that performs takeover, and any software originally running on the first mechanism is transferred to the second mechanism.

The exactly-once architecture according to the invention can distribute the boot servers to servers located in one of these hardware clusters so that a boot server failure becomes faster to handle than a failure in a software cluster. However, the bootstrap server issues a small number of leases to servers regardless of whether those servers are located in a hardware cluster or a software cluster. This configuration allows for faster recovery of the boot server, and although software clusters are allowed to be larger than hardware clusters, software clusters still include hardware clusters.

One such system 200 is shown in FIG. 2 . A hardware cluster 218 includes a single facility that includes multiple servers 220 , 222 , 224 . The hardware cluster may be used to select a boot server 220 from among the servers at the facility, for example, for efficiency. Once the bootstrap server 220 is selected, the bootstrap server may select the host 206 for the object 214 in the file system 212 , which may be located inside or outside the software cluster 210 . The file system 214 itself may copy the object 214 as a second object 216 located on another disk in the file system, thereby providing persistence. The new host 206 fetches the object 214 from the file system 212 using the cached copy 208 of the object on the host 206 . When a request such as services 206, 216, and 220 is received from a browser or client 202 over network 204, the server knows to connect to master server 206 if it needs to access a cached copy of object 208.

One method 500 of utilizing such a system is shown in FIG. 5 . A bootstrap server is selected using an algorithm of the hardware cluster 502 . The algorithm may be, for example, a dedicated algorithm of the hardware cluster device, or may be a distributed consensus algorithm which only requires the hardware cluster servers to be consistent. Thereafter, a master server is pre-selected using a distributed consensus algorithm such as the Paxos algorithm possessed by the bootstrap server 504 . Thereafter the identity of the preselected host is multicast to another server located in the software cluster which includes the hardware cluster. The bootstrap server receives approval or disapproval from each server that will soon be operational and connected to the cluster 508 . If the server agrees with the pre-selected master, then a commit message is multicast to the cluster server to inform the server that the pre-selected host now accepts the object; otherwise, if the server does not agree that the new host is pre-selected and start the process again.

Exactly-once structures are used, for example, to process or cache transaction records. This structure can also be used, for example, to define a management server as an exactly one-time object and to rent out a management server so that it never stops working.

FIG. 6 shows another example of a cluster system 600 according to the present invention, wherein an object 608 is used as a message store in relation to a Java Message Service 612 (JMS). All servers 606 , 614 , 616 in cluster 610 use JMS, but they must transfer messages to message store 608 and get any messages from message store 608 over network 604 . The master server 606 of the servers in the cluster 610 will "own" or "host" the message store 608 . Client or browser 602 may make a request to network 604 directed to server 616 in cluster 610 . Server 616 accesses JMS simply by sending a message over network 604 to message store 608 located on master server 606 .

FIG. 7 shows a block diagram 700 of a computer system that can be used as a component of the present invention or can implement the method of the present invention. The computer system of FIG. 7 includes a processing unit 704 and main memory 702 . Processing unit 704 includes a single microprocessor, or may include multiple microprocessors to configure the computer system as a multi-processor system. Portion of main memory 702 stores instructions and data that are executed by processor unit 704 . If the present invention is implemented in whole or in part in software, main memory 702 may store executable program code when operating. Main memory 702 includes dynamic random access memory (DRAM), high-speed cache memory, and other types of memory well known in the art.

The system shown in FIG. 7 further includes mass storage 706 , a peripheral device 708 , a user input device 712 , a portable storage media drive 714 , a graphics subsystem 718 and an output display 716 . For simplicity, the components shown in FIG. 7 are connected by a single bus 720 . However, it will be apparent to one of ordinary skill in the art that the components may be connected by one or more data transfer means. For example, processor unit 704 and main memory 702 are connected by a local microprocessor bus, and mass storage 706, peripherals 708, portable storage media drive 714, and graphics subsystem 718 are connected by one or more input/output (I/O) /O) connected to the bus. Mass storage 706 , implemented by a magnetic disk drive, optical disk drive, and other drives known in the art, is a non-volatile storage device for storing data and instructions used by processor unit 704 . In one embodiment, mass storage 706 stores software for implementing the present invention for loading into main memory 702 .

A portable storage medium drive 714 is connected to a portable nonvolatile storage medium such as a floppy disk to input/output data and program codes to/from the computer system shown in FIG. 7 . In one embodiment, system software for implementing the present invention is stored on such a portable medium and imported to the computer system via the portable storage medium drive 714 . Peripherals 708 may include any kind of computer support device that adds auxiliary functionality to the computer, such as an input/output (I/O) interface. For example, peripherals 708 may include a network interface that enables the computer system to connect to a network, as well as other network hardware such as a modem, router, or other hardware known in the art.

User input device 712 provides part of the user interface. User input devices 712 may include an alphanumeric keypad or pointing devices such as a mouse, track pen, stylus, or cursor direction keys for entering alphanumeric and other information. To display textual and graphical information, the computer system of FIG. 7 includes a graphics subsystem 718 and an output display 716 . Output display 716 includes a cathode ray tube (CRT) display, liquid crystal display (LCD), or other suitable display device. Graphics subsystem 718 receives textual and graphical information and processes the information for output to display 716 . Additionally, the system of FIG. 7 includes an output device 710 . Examples of suitable output devices include microphones, printers, network interfaces, monitors, and other output devices known in the art.

Components included in the computer system of FIG. 7 are typically found in computer systems suitable for use in certain particular embodiments of the invention, and represent a broad class of such computer elements known in the art. Thus, the computer system of FIG. 7 may be a personal computer, workstation, server, minicomputer, mainframe, or any other computer. Computer system 700 may also employ different bus structures, network platforms, multi-processor platforms, and the like. Any operating system can be used including Unix, Linux, Windows, Macintosh OS, Palm OS, and other suitable operating systems.

The preferred embodiments of the invention have been presented for purposes of illustration and description. It is not intended to detail or to limit the precise structures disclosed. Obviously, various modifications and changes will be apparent to those skilled in the art. The embodiment was chosen and described in order to better illustrate the principles of the invention and its practical application, so that those of ordinary skill in the art can understand that the invention can be utilized in various embodiments and can make its suitable use specific purpose. Revise. The scope of the invention is defined by the following claims and their equivalents.

Claims (51)

1. system that is used for object on the managerial grid comprises:

A plurality of webservers, each webserver are used for and the network data sources traffic; And

Be arranged in a Boot Server of described a plurality of webservers, this Boot Server comprises a distributed consistency algorithm that is used for selecting from described a plurality of webservers a master server, this master server comprises an object relevant with a data item in the network data source, thereby makes in the described a plurality of webservers that need the visit data item any one can visit described object on the described master server.

2. according to the system of claim 1, wherein, the described webserver is to choose from a group of being made up of hardware cluster server and software bundling storage server.

3. according to the system of claim 1, wherein, described distributed consistency algorithm comprises the message circulation between described Boot Server and the described a plurality of server, and this circulation continues always, till the great majority in described a plurality of webservers are agreed described master server.

4. according to the system of claim 1, wherein, described master server comprises a data object, and this data object comprises the copy from the data in described network data source.

5. according to the system of claim 1, wherein, described master server comprises a data object, unique access point that this data object conducts interviews to the data item in the network data source with work.

6. according to the system of claim 1, wherein, described data item is a transaction record.

7. according to the system of claim 1, wherein, described distributed consistency algorithm is the Paxos algorithm.

8. system that is used for object on the managerial grid comprises:

A plurality of webservers, each webserver are used for and the network data sources traffic; And

Be arranged in a Boot Server of described a plurality of webservers, this Boot Server comprises a distributed consistency algorithm that is used for selecting from described a plurality of webservers a master server, this master server comprises a copy of a data item that is arranged in the network data source, thereby makes in the described a plurality of webservers that need the described data item of visit any one can visit described copy on the described master server.

9. system that is used for object on the managerial grid comprises:

A plurality of webservers, each webserver are used for and the network data sources traffic; And

Boot Server in described a plurality of webserver, this Boot Server comprises a distributed consistency algorithm that is used for selecting from described a plurality of webservers a master server, this master server comprises unique access point that the data item that is arranged in the network data source is conducted interviews, thus make described a plurality of webservers of needing this data item of visit any one all must pass through this master server and visit this data item.

10. system that is used for object on the managerial grid comprises:

One file system, this document system comprises the copy of at least one data item;

The a plurality of servers that communicate with this document system; And

Be arranged in a Boot Server of described a plurality of servers, this Boot Server comprises a distributed consistency algorithm that is used for selecting from described a plurality of servers a master server; And

Be arranged in a master server of described a plurality of servers, described master server comprises the local copy of described data item, described master server can make in described a plurality of server any one visit this local copy, and upgrades the described copy of the described data item in the described file system in any time of upgrading local copy.

11. according to the system of claim 10, wherein, described master server is further used for local copy is stored in the local cache buffer memory device.

12. according to the system of claim 10, wherein, described a plurality of servers comprise a cluster.

13. according to the system of claim 10, wherein, described file system is replicated in data item on a plurality of disks.

14. a system that is used for object on the managerial grid comprises:

One file system, this document system comprises the copy of at least one data item;

The a plurality of servers that communicate with this document system;

One hardware cluster, this hardware cluster comprise the hardware cluster server that is arranged in described a plurality of servers, and described hardware cluster comprises an algorithm that is used for selecting from described a plurality of hardware clusters a Boot Server;

Be arranged in a Boot Server of described hardware cluster server, this Boot Server comprises a distributed consistency algorithm that is used for selecting from described a plurality of servers a master server; And

Be arranged in a master server of described a plurality of servers, described master server comprises the local copy of data item, if when described master server can make any one addressable this local copy in described a plurality of server and upgrade local copy with regard to the copy of the data item in the updating file system.

15. according to the system of claim 14, wherein said master server is arranged in described hardware cluster.

16. a method that is used for object on the managerial grid comprises:

Utilize distributed consistency algorithm from a plurality of webservers, to select a master server;

Send the copy of a data item to described master server from file system; And

The described master server of notifying other webserver to include a data item copy will be used in the processing of network requests.

17., further comprise when the copy on the master server is modified the step of the data item in the updating file system according to the method for claim 16.

18., further comprise and limit other webserver visits file system through described master server step according to the method for claim 16.

19., comprise that further a copy that guarantees to have only data item is present in the step of the outside of this document system according to the method for claim 16.

20., comprise that further a copy that guarantees data item always is present in the step of this document system outside according to the method for claim 16.

21. according to the method for claim 16, further comprise, then utilize distributed consistency algorithm from a plurality of webservers, to select the step of new master server if master server is no longer admitted this object.

22. method according to claim 16, further comprise if master server is no longer admitted this object, then send the copy of a data item to the step of a new master server, the master server that this new master server is to use described distributed consistency algorithm to select from described file system.

23., comprise further that the new master server of notifying one of other webserver to comprise described data item so will be used in the step in the processing that network requests is carried out according to the method for claim 16 if master server is no longer admitted this object.

24. a structural system that is used for object on the managerial grid comprises:

A plurality of servers, all cacheable data object of each server;

One file system, this document system comprise at least one copy of data item;

One distributed consistency algorithm is used for selecting a master server from described a plurality of servers, and this master server is used for the copy of cached data object; And

One compartment system is used for the copy that this principal computer of server on the informing network comprises described data object.

25. one kind is used for object is hired out method to server on the network, comprises:

Utilize distributed consistency algorithm from a plurality of webservers, to select a master server;

Data object is hired out to described master server, and wherein this master server has in the time period or admits this data object hiring out;

From file system, take out the copy of data item to deliver to master server; And

The copy that comprises data item to other webserver notice master server is handled network requests being used for.

26., further comprise step: in another hires out the time period, upgrade termly this data object is hired out to master server according to the method for claim 25.

27., further comprise step:, then this data object is hired out to one of new master server and hired out the time period in case the taxi time period on the master server has expired according to the method for claim 25.

28. one kind is used for object is hired out method to the server on the network, comprises:

Select a Boot Server in a plurality of hardware cluster servers from the hardware cluster;

Utilize the distributed consistency algorithm on the described master server from a plurality of webservers, to select a master server;

Data object is assigned to master server, and appointed master server can carry out independent visit at an official hour to data item in the cycle;

From file system, take out the copy of data item to deliver to master server; And

The copy that comprises data item to other webserver notice master server is handled network requests being used for.

29. the method that the entitlement that is used for the object on the network distributes comprises:

Select a Boot Server in a plurality of hardware cluster servers from the hardware cluster;

Utilize the distributed consistency algorithm on the described Boot Server from a plurality of webservers, to select a master server;

Data object is assigned to this master server, and appointed master server can be visited separately the data object on the network;

From file system, take out the copy of data item to deliver to master server; And

The copy that comprises data item to other webserver notice master server is handled network requests being used for.

30. a method that is used for guaranteeing to exist at cluster object comprises:

The master server that use is arranged in a plurality of servers provides the visit to the data object;

If described master server can not provide the visit to described data object, so, utilize distributed consistency algorithm from a plurality of servers, to select a new master server;

With information send to need provide to described data item provide visit new master server and

Notify other server in described a plurality of server new addressable this data object of master server.

31. a method that is used at the cluster distribute objects comprises:

Utilize distributed consistency algorithm to come from a plurality of webservers, to select a master server;

One data object is assigned to this master server, and appointed this master server can be visited separately a data item; And

The one described master server of notifying other server that is multi-cast in the cluster to comprise the copy of described data item with notice will be used in the processing of network requests.

32. a method that is used at the cluster distribute objects comprises:

Utilize distributed consistency algorithm to come from a plurality of webservers, to select a master server;

Whether each server that connects in the cluster can be accepted by those servers with definite selected master server; And

If it is acceptable that the Servers-all in the cluster is all agreed selected master server, notifies other server that is multi-cast in the cluster to transfer to selected new master server with one so and be responsible for notice.

33. a system that is used at the cluster distribute objects comprises:

Device is used for utilizing distributed consistency algorithm to select a master server from a plurality of webservers;

Device is used for taking out the copy of data item to deliver to master server from a file system; And

Device, this master server that is used for notifying other webserver to comprise this data item copy will be used in the processing that network requests is carried out.

34. a method that is used for object on the managerial grid comprises:

Utilize and select a master server in a plurality of webservers of Paxos algorithm from the software bundling storage; And

Data object is assigned to described master server, and wherein this master server is provided in the network the independent visit to this data object.

35., further comprise the step of from a file system, taking out these data object data according to the method for claim 34.

36., comprise that further this master server of notifying other webserver to comprise the object relevant with described data item will be used in the step in the processing that network requests is carried out according to the method for claim 34.

37., further comprise with the sign multileaving of new master server step to another webserver according to the method for claim 34.

38., wherein utilize the Paxos algorithm to come to select in a plurality of webservers from the software bundling storage the described step of a master server to comprise cyclical information be multi-cast to another webserver according to the method for claim 34.

39. the system that the object that is used for cluster distributes comprises:

Device is used for utilizing distributed consistency algorithm to select a master server from a plurality of webservers;

Device is used for giving master server with a JMS object distribution, and this master server that is distributed provides the independent visit to the JMS object; And

Device is used to notify this master server of its webserver that independent visit to JMS is provided.

40. a method that is used for object on the managerial grid comprises:

Utilize and select a master server in a plurality of webservers of Paxos algorithm from the software bundling storage; And

The JMS message memory is assigned to master server, and this JMS message memory exists only on the described master server.

41., further comprise and notify other webserver step that this master server is admitted described JMS message memory according to the method for claim 40.

42., further comprise step according to the method for claim 40: with the sign multileaving of new master server to another webserver.

43., wherein utilize the Paxos algorithm to come to select in a plurality of webservers from the software bundling storage the described step of a master server to comprise cyclical information be multi-cast to another webserver according to the method for claim 40.

44. a system that is used for JMS object on the managerial grid comprises:

A plurality of webservers; And

Boot Server in described a plurality of webserver, this Boot Server comprises a distributed consistency algorithm that is used for selecting from the described a plurality of webservers master servers, and this master server comprises a JMS object so that its any one of a plurality of webservers that need visit JMS must conduct interviews to the JMS object on the master server.

45. according to the system of claim 44, wherein, the described webserver is to choose from be made up of hardware cluster server and software bundling storage server such one group.

46. according to the system of claim 44, wherein, described distributed consistency algorithm comprises the circulation message between described master server and the described a plurality of server, this circulation continues to agree master server up to the great majority of described a plurality of webservers always.

47. one kind is used for the JMS object is hired out method to the server on the network, comprises:

Utilize distributed consistency algorithm to come from a plurality of hardware cluster servers, to select a master server;

The JMS object is hired out to this master server, and wherein this master server has in the time period or admits this JMS object hiring out; And

Notify this master server that this JMS object is admitted to other webserver.

48., further comprise step: in another hires out the time period, upgrade termly this JMS object is hired out to master server according to the method for claim 47.

49., further comprise step:, then described JMS object is hired out to one of new master server and hired out the time period in case the taxi time period on the described master server has expired according to the method for claim 48.

50. one kind is used for the JMS object is hired out method to the server on the network, comprises:

Select a Boot Server in a plurality of hardware cluster servers from the hardware cluster;

Utilize the distributed consistency algorithm on the described Boot Server to come from a plurality of webservers, to select a master server;

The JMS object is assigned to this master server, and appointed master server can provide the independent visit to described JMS object in a time cycle of regulation; And

Notify this master server of other webserver to admit described JMS object.

51. one kind is used for method that the object on the network is distributed, comprises:

Select a Boot Server in a plurality of hardware cluster servers from the hardware cluster;

Utilize the distributed consistency algorithm on the described Boot Server to come from a plurality of webservers, to select a master server;

The JMS object is assigned to this master server, and appointed master server can provide the independent visit to described JMS object; And

Notify this master server of other webserver to admit described JMS object.

Images