JP2006309439A - Flexible cluster system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a flexible cluster system, capable of freely changing the role of a server device which belongs to each cluster. <P>SOLUTION: The flexible cluster system comprises a shared disk device including a server resource storage means storing server resource information for providing individual functions and a system information storage means storing system cluster information for managing each server device, and a plurality of server devices, each including a start means starting a part of a cluster corresponding to an assigned function according to the system cluster information, a failure report means reporting information related to a failure detected for a function under providing to other server devices, a collection means collecting the failure information, a determination means determining whether the function to be performed by the own device should be changed or not based on the collected failure information and the system cluster information, and a restart means acquiring, according to the determination result that the function is to be changed, server resource information corresponding to the function of a changing destination from the server resource storage means of the shared disk device and restarting it. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention is composed of a plurality of clusters each providing various functions such as an operation system for networks such as ATM and wide area Ethernet (registered trademark) and an online system of a financial institution, and maintains operation without interruption. In particular, the present invention relates to a technique for reliably maintaining operability in the case where a multiple failure relating to a plurality of server apparatuses constituting the system occurs.

Conventionally, an operation standby configuration or a scalable configuration is generally used in order to realize high availability required in a network operation system or the like.
The standby configuration is a configuration in which multiple systems that perform the same function are prepared, one of which is set as a standby system, while the other system is operated as an active system. The scalable configuration is a degraded configuration when a failure occurs. In this configuration, a sufficient number of servers are prepared in advance so as to provide the minimum performance necessary for operation.

FIG. 8 shows a configuration example of a conventional network operation system.
The network operation system shown in FIG. 8 includes a WEB cluster 410, a database cluster 420, an application cluster 430, and device clusters 440, 450, and 460. A LAN 403 is connected to both the WAN 401 to which the terminal device belongs and the WAN 402 to which the transmission device belongs. Connected through the network.

  The WEB cluster 410 shown in FIG. 8 is composed of three WEB servers 411a, 411b, and 411c, and a service request input from a terminal device via the WEB with a scalable configuration that assumes a stop of up to two of the three servers. Plays a reception processing function. In addition, the database cluster 420 shown in FIG. 8 includes two database (DB) servers 421a and 421b, and a scalable configuration assuming that one of these DB servers 421a and 421b is stopped enables system operation. It performs the maintenance management function of the database concerned.

  On the other hand, the application cluster 430 illustrated in FIG. 8 includes three application (AP) servers 431a, 431b, and 431c. For example, the AP servers 431a and 431b are the active system and the AP server 431c is the standby system 2 The application processing function such as the route selection process for providing the service requested by the service request is fulfilled by the 1: standby configuration. Further, each of the device clusters 440, 450, and 460 shown in FIG. 8 includes two device servers 441a (451a and 461a) and 441b (451b and 461b), respectively, and one of them becomes an operation standby. According to the configuration, each of the device setting processing functions for setting the selected route using the function of the transmission device is achieved.

  In the network operation system shown in FIG. 8, for example, each WEB server 411a, 411b, 411c provided in the WEB cluster 410 has a scalable configuration together with a general-purpose operating system such as UNIX (registered trademark) or LINUX. Cluster software for realizing, application software for realizing service request reception processing, and middleware necessary for executing the application software are incorporated.

Similarly, each of the device servers 441a and 441b provided in the device cluster 440 shown in FIG. 8 includes, in addition to the operating system described above, cluster software for realizing an operation standby configuration, application software for controlling the device, and middleware. Is incorporated.
In such a compound computer system composed of a plurality of server devices, a shared file that can be commonly accessed by all the servers belonging to the compound computer system is a method for improving the efficiency of software installation and activation processing in each server device. A method has been proposed in which a system configuration definition table and a system control program are arranged in the server, and each server device operates a necessary system control program based on the system configuration definition table (see Patent Document 1).
JP 7-325708 A

  In the cluster system composed of a plurality of clusters as shown in FIG. 8, the roles during operation of the plurality of server devices constituting each cluster are fixed both within the cluster and as a whole cluster system. That is, in the network operation system shown in FIG. 8, the server device activated as the WEB server is fixed to the role as the WEB server.

For this reason, in order to build a system that requires high availability, an active / standby configuration or a scalable configuration is adopted for each cluster, and as a whole, the number exceeds a sufficient number to maintain the normal operation of the system. It was necessary to prepare an excessive number of servers.
Note that even the technique described in Patent Document 1 only allows the role of each server device in the compound computer system to be changed at each activation by changing the system configuration definition table, for example. Therefore, the role of each server device in operation is fixed in the same way as in the cluster system shown in FIG. 8, so that an excessive number of server devices are prepared as a whole to build a system that requires high availability. It is similar in that it needs to be done.

  By the way, due to advances in mounting technology related to individual server devices, the probability that a failure will occur in each server constituting each cluster is considerably low if not negligible. On the other hand, in consumer systems such as network operation systems, although high availability is required, the cost required to build the entire system and the cost required to operate, maintain and manage the built system There is a strong demand to reduce the total cost of ownership including so-called TCO (Total Cost of Ownership).

From these facts, a technology for constructing a cluster system that realizes high availability by flexibly responding to various failures that can occur in the system while minimizing the redundancy of hardware resources such as server devices. Is needed.
An object of the present invention is to provide a flexible cluster system capable of freely changing the roles of server devices belonging to each cluster in a cluster system composed of a plurality of clusters.

A first flexible cluster system according to the present invention includes a shared disk device having server resource information storage means and system information storage means, startup means, failure notification means, collection means, determination means, and failure notification. It comprises a plurality of server devices having means, collection means, determination means, and restart means.
The principle of the first flexible cluster system according to the present invention is as follows.

  In a shared disk device provided in a flexible cluster system provided with a plurality of server devices and a shared disk device shared by these server devices, the server resource storage means each of the plurality of server devices provides individual functions. Server resource information including middleware, application programs, and logical IP addresses necessary for the storage is stored corresponding to each function. The system information storage means stores system cluster information for managing the status of each of a plurality of server devices in the flexible cluster system and the allocation of each server resource information to each server device. In each of the plurality of server apparatuses, the activation unit acquires server resource information corresponding to the assigned function according to the system cluster information, and activates it as a part of the cluster corresponding to the function. The failure notification means detects a failure related to the function being provided, and notifies other server devices of information related to the detected failure as part of the failure information. The collection unit collects failure information related to a failure in another server device together with information related to the detected failure. Based on the collected failure information and system cluster information, the determination means determines whether or not the function to be performed by the own device in the flexible cluster system should be changed. The restarting means obtains server resource information corresponding to the function of the change destination from the server resource storage means of the shared disk device according to the determination result indicating that the function is changed, and obtains one of the clusters corresponding to the function of the change destination. Restart as part.

The operation of the first flexible cluster system configured as described above is as follows.
Each server device belonging to the flexible cluster system uses the server resource information acquired from the server resource storage unit of the shared disk device by each activation unit according to the basic allocation included in the system cluster information, for example. Start as part of a cluster that provides the functionality indicated by the dynamic allocation. After that, when a failure occurs in any of the server devices, failure information indicating that fact is notified to the other server devices by the failure notification means provided in the server device, and is collected by the collecting means. At this time, the determination means provided in each server device determines, for example, whether or not it should play its role instead of the server device that has gone down due to a failure, and the restarting means is determined according to the determination result. The restart process is performed.

  In this way, when a failure occurs in any of the server devices that make up the flexible cluster system, each server device in the system autonomously changes its role assignment to maintain the provision of various functions. Thus, the operation of the entire system can be reliably continued. In the cluster system configured as described above, a surplus server device can be shared between these clusters while taking a system configuration including a plurality of clusters corresponding to each function.

A second flexible cluster system according to the present invention is configured by including an alternative determination unit and a server selection unit in the determination unit of the first flexible cluster system described above.
The principle of the second flexible cluster system according to the present invention is as follows.
In the determination means provided in the first flexible cluster system described above, the alternative determination means indicates that a failure has occurred in any of a plurality of server devices constituting the flexible cluster system based on the collected failure information. Based on the system cluster information, it is determined whether or not the role sharing of the server device in which the failure has occurred needs to be replaced by another server device. The server selection means selects an alternative server to replace the failed server device based on the system cluster information according to the determination result of the replacement, and changes the function when the alternative server is the own device. The judgment result of is output.

The operation of the second flexible cluster system configured as described above is as follows.
For example, when a failure occurs in a server device that belongs to a cluster with a small surplus capacity, the performance that can be achieved by an operational server device that belongs to the relevant cluster is lower than the capacity required to maintain the operational status of the entire system. there is a possibility. In such a case, for example, the alternative determination unit included in the determination unit of each server device is less than the information about the minimum configuration of the corresponding cluster indicated by the system cluster information in which the cluster configuration changed due to the failure of the server device. Based on whether or not, the necessity of the alternative server is determined. When it is determined that a replacement is necessary, the server selection unit provided in the determination unit of each server device selects the replacement server based on the system cluster information, and the selected replacement server is the own device. In such a case, restart by the restarting means is performed.

Here, the substitution determination means and server selection means provided in each server device perform the respective determination processing and selection processing with reference to the same system cluster information stored in the system information storage means of the shared disk device. The processing results autonomously performed in each server device reach the same conclusion, and an appropriate server device is restarted as an alternative server.
A third flexible cluster system according to the present invention comprises a substitute means provided in the second flexible cluster system described above, comprising comparison means, remaining power detection means, imbalance detection means, and determination means. Is done.

The principle of the third flexible cluster system according to the present invention is as follows.
In the alternative determination means provided in the second flexible cluster system described above, the comparison means includes the minimum number of servers indicated by the system cluster information for the cluster to which the failed server apparatus belongs, and an operable server apparatus belonging to the cluster. Compare the number of. When the number of operable server devices belonging to the cluster is equal to the minimum number of servers, the surplus power detection means is a server that can operate more than the minimum number of servers indicated in the system cluster information in another cluster. A cluster having more than a predetermined number of devices is detected. The imbalance detection means detects an imbalance in capacity for each cluster based on information regarding the load applied to each cluster indicated by the system cluster information and the number of operable server devices belonging to each cluster. The determining means determines whether or not it is necessary to assign an alternative server for the server device in which the failure has occurred, based on the comparison result by the comparing means and the detection results by the remaining power detecting means and the imbalance detecting means.

The operation of the third flexible cluster system configured as described above is as follows.
In the alternative determination means, the determination means provides a function corresponding to the cluster when, for example, the comparison means causes the number of server apparatuses operating in the cluster to which the failed server apparatus belongs to be less than the minimum number of servers. It judges that it stops, and outputs the judgment result to the effect that an alternative server is necessary. In addition, for example, the judgment means detects a load imbalance between the clusters by the imbalance detection means, and the cluster other than the cluster where the load is concentrated has a lot of remaining power by the remaining power detection means. When it is detected, a determination result indicating that an alternative server is necessary is output.

In this way, not only when the function provision by the cluster to which the failed server device belongs is exposed to a crisis of stoppage, but also when a load imbalance or extreme performance imbalance occurs due to the occurrence of the failure In addition, by allocating alternative servers, it is possible to continue the smooth operation of the entire system.
A fourth flexible cluster system according to the present invention comprises a server selection means provided in the second flexible cluster system described above, comprising standby detection means, cluster discrimination means, candidate extraction means, and determination means. Is done.

The principle of the fourth flexible cluster system according to the present invention is as follows.
In the server selection means provided in the second flexible cluster system described above, the standby detection means detects a server device whose operation state is standby based on the system cluster information. The cluster discriminating unit discriminates a cluster having a large surplus capacity based on the number of surplus servers obtained by subtracting the minimum number of servers indicated by the system cluster information from the number of server devices operating in each cluster. The candidate extraction means selects one of the clusters having a large surplus capacity according to the priority order indicated by the system cluster information, and extracts server devices belonging to the selected cluster as alternative server candidates. The determining means determines a server device selected from the detected standby server or alternative server candidate as the alternative server.

The operation of the fourth flexible cluster system configured as described above is as follows.
In the server selection means, the determination means first sets the server device in standby detected by the standby detection means as an alternative server. On the other hand, when there is no waiting server device, the determination unit selects one of the alternative server candidates belonging to the cluster selected by the candidate extraction unit from the clusters having a large remaining capacity determined by the cluster determination unit. Use an alternative server.

A fifth flexible cluster system according to the present invention is configured by including a load calculation unit and a selection determination unit in the determination unit provided in the above-described fourth flexible cluster system.
The principle of the fifth flexible cluster system according to the present invention is as follows.
In the determining means provided in the fourth flexible cluster system described above, the load calculating means, when allocating each of the alternative server candidates as an alternative server, indicates an index indicating at least the load in the allocation destination cluster and the allocation source cluster. For each. The selection determining means preferentially selects the detected standby server as an alternative server, and when no standby server is detected, selects one of the alternative server candidates as an alternative server based on the index obtained by the load calculation means To do.

The operation of the fifth flexible cluster system configured as described above is as follows.
If no standby server is detected, one of the server devices belonging to the cluster with a large surplus capacity is assigned as a cluster, so naturally the load on the other server devices left in the cluster from which the alternative server is sent increases. To do. Along with this, the load applied to the cluster other than the cluster from which the alternative server is sent and the cluster to which this server is assigned may be affected.

  An index indicating the degree of the influence on the load is obtained for each alternative server candidate by the load calculating means, and the selection determining means selects the alternative server based on these indices, so that the entire flexible cluster system can be smoothly operated. It is possible to select an optimal server device as an alternative server device in maintaining operation.

In the flexible cluster system according to the present invention, the role of each server device belonging to the system can be flexibly changed in response to a situation such as the occurrence of a failure. It is possible to achieve both the realization of availability and the excessive reduction of hardware resources constituting the cluster system.
As a result, a consumer system such as a network operation system can be realized with a significantly smaller number of servers than when a redundant configuration is adopted for each cluster corresponding to each function to be provided. Sufficient high availability can be achieved to provide various services without interruption at a reasonable cost.

  In particular, the idea of centrally managing the entire system is abolished, and each server device constituting the flexible cluster system autonomously collects and analyzes information on failures, and selects whether or not to allocate an alternative server and its candidates. By adopting the configuration, the flexibility of the entire system can be remarkably improved and a very high level of high availability can be realized.

  In addition, in the process of selecting alternative servers and determining alternative servers, we analyze the priorities of functions provided by individual clusters, the magnitude of load on each cluster, and the load imbalance among clusters. By determining an alternative server based on the analysis result, smooth operation of the flexible cluster system can be reliably maintained.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 shows an embodiment of a flexible cluster system according to the present invention.
1 that are the same as those shown in FIG. 8 are denoted by the same reference numerals as those shown in FIG. 8 and description thereof is omitted.
The flexible cluster system shown in FIG. 1 is connected to both a wide area network (WAN) 401 to which a user terminal device belongs and a WAN 402 to which a transmission device belongs, and transmission paths set by these transmission devices are Thus, an operation system for a network that distributes video content to each user terminal device is realized.

  In the flexible cluster system shown in FIG. 1, the server apparatuses S <b> 1 to S <b> 12 are connected to each other via an interconnect (not shown) and are connected to the above-described WANs 401 and 402 via a LAN 403. In the flexible cluster system shown in FIG. 1, the shared disk device is formed by, for example, a storage area network (SAN) 201, and each of the server devices S1 to S12 described above via a high-speed LAN such as the fiber channel 202. It is connected to the.

  The storage area network 201 shown in FIG. 1 includes WEB necessary for operating each of the server devices described above as the WEB server, application server, database management server, and device server shown in FIG. Server resource information (WEB1, WEB2, WEB3), application (AP) server resource information (AP1, AP2), database (DB) server resource information (DB1, DB2) and device server resource information (device 1, device 2, device 3) ) Is stored.

  These server resource information includes, in addition to application programs and middleware for providing corresponding functions, monitoring means for monitoring these operations and hardware operations of the server device, application programs and middleware Information such as a start / stop command required for start / stop and a logical IP address to be assigned is included.

  The system cluster information stored in the storage area network 201 indicates a server status table indicating the status of each of the server devices S1 to S12 in the flexible cluster system and a basic function assignment to each of the server devices S1 to S12. It includes a basic allocation table, a server correspondence table indicating the correspondence between each function type, each server resource information, and each server device S1 to S12, and reference information that is used as a judgment material when changing the function assignment. .

On the other hand, in each of the server apparatuses S1 to S12 shown in FIG. 1, a general-purpose operating system such as UNIX (registered trademark) or LINUX and these server apparatuses S1 to S12 are operated as a part of the flexible cluster system. Built-in cluster software. The cluster software acquires and incorporates appropriate server resource information based on the above-described system cluster information, so that each server device Si can handle the incorporated server resource information as shown in FIG. A function providing processing unit 211 that provides the function to be performed, and function monitoring units 212 _{1 to} 212 _n that respectively monitor the monitoring target process and hardware belonging to the function providing processing unit 211, and server management is performed by the cluster software. A processing unit 213 is formed.

  For example, as shown in FIG. 3B, according to the assignment shown in the basic assignment table of the system cluster information, the server devices S1 to S3 shown in FIG. 1 are WEB server resource information, and the server devices S4 and S5 are DB servers. The server apparatuses S6 and S7 acquire AP server resource information, and the server apparatuses S8 to S10 acquire apparatus server resource information. By incorporating the server resource information acquired by each of these server apparatuses S8 to S10, a function provision processing unit 211 that performs a function corresponding to the type of the incorporated server resource information is formed (see FIG. 2). In this manner, the server apparatuses S1 to S10 illustrated in FIG. 1 are activated as part of the cluster corresponding to each function.

  At this time, the server management processing unit 213 of these server apparatuses S1 to S10 sets the relative performance level of each server apparatus S1 to S10 together with information indicating that each is operating in the server status table of the system cluster information. The figure of merit shown is written (see FIG. 3 (a)). On the other hand, since the server resource information to be allocated is not shown for the server devices S11 and S12, these server devices become standby servers, and information indicating that is written in the server status table (FIG. 3A). reference).

  The above-mentioned performance index is measured based on a method unified in the flexible cluster system by the server management processing unit 213 formed in each of the server devices S1 to S12 (see FIG. 2), for example, and according to the measurement result. Can be determined. FIG. 3A shows a value obtained by normalizing the performance of other server devices, with the performance of the server device having the lowest performance as a numerical value “1” by the above-described measurement.

Further, as shown in FIG. 3C, each of the server apparatuses S1 to S10 that have acquired the server resource information as described above assigns a server name indicating itself to the server corresponding to the acquired server resource information. It is written in the correspondence table, and shows the correspondence between each server device S1 to S10 and the cluster to which each belongs.
In this way, the flexible cluster system including the server apparatuses S1 to S12 is started as a cluster system including the WEB cluster, the database cluster, the application cluster, and the apparatus cluster, and the two server apparatuses S11 and S12 are It can be secured as a standby server that can be assigned to other clusters.

Next, a configuration related to the operation of the server apparatus activated as part of the cluster system as described above will be described.
In the server management processing unit 213 illustrated in FIG. 2, the monitoring information collection unit 214 collects the monitoring information obtained by each of the function monitoring units 212 described above, and the operation status of the function provision processing unit 211 based on the collected monitoring information. When a failure is detected, information relating to the failure detected via the failure notification unit 215 is reflected in the system cluster information stored in the storage area network 201. Further, the server monitoring unit 216 shown in FIG. 2 detects a failure of the message exchange partner server device by stopping the message exchange with other server devices performed via the interconnect, and a failure notification unit. Information about the detected failure is reflected in the system cluster information via H.215. At this time, as shown in FIG. 4A, the failure notification unit 215 writes status information indicating that a failure has been detected in the server status table included in the system cluster information, corresponding to the corresponding server device. In the server correspondence table shown in FIG. 4B, the information about the detected failure is reflected by clearing the information indicating the association of the server resource information regarding the corresponding server device. FIG. 4 shows a case where a failure is detected in the server device S4.

On the other hand, the autonomous management unit 217 shown in FIG. 2 performs processing for autonomously managing the roles of the server devices belonging to the flexible cluster system based on the system cluster information in which the failure information is reflected as described above. Do.
In the autonomous management unit 217 illustrated in FIG. 2, the change condition determination unit 221 periodically refers to the server status information included in the system cluster information, and each time information about a new failure is indicated, the change condition determination unit 221 indicates the server correspondence table. On the basis of the information indicating the current state of the cluster system and the determination condition indicated by the reference information. When the change condition determining unit 221 obtains a determination result indicating that an alternative server needs to be allocated, the alternative candidate extracting unit 222 provides information on the current state of the cluster system indicated by the server state table and the server correspondence table. And alternative server candidates are extracted from the twelve server devices S1 to S12 belonging to the flexible cluster system based on the selection conditions indicated by the reference information. For each alternative server candidate extracted in this way, the load index calculation unit 223 calculates a load index indicating the load applied to each cluster when each alternative server candidate is an alternative server. Based on the load index, The alternative server determination unit 224 determines the final alternative server. When the determined alternative server is the own device, the alternative server determining unit 224 causes the restart processing unit 225 to restart the own device as a part of the cluster to which the failed server device belongs. Instruct. In response to this, the restart processing unit 225 reflects the switch from the failed server device to the alternative server in the server correspondence table and instructs the function providing processing unit 211 to release the server resource information being acquired. Thereafter, server resource information corresponding to the designated cluster is newly acquired, and a function providing processing unit 211 corresponding to the server resource information is newly formed.

Hereinafter, the operation of the above-described autonomous management unit 217 will be described in detail based on a specific example.
For example, when the server status table indicates that a failure has occurred in the server device S4 that has been operating as an application server by acquiring the AP server resource information (AP1), the change condition determination unit 221 displays It is determined whether or not the allocation of the alternative server should be executed according to the procedure as shown.

  The change condition determination unit 221 first obtains the number of server devices belonging to each cluster by referring to the server correspondence table, and compares this with the number of servers indicating the minimum configuration of the cluster system indicated by the reference information (see FIG. Step 301 of 5). For example, as the minimum configuration of the cluster system, in order for the cluster system to continue operation, the number of server devices that should inevitably belong to each cluster is obtained in advance, and as shown in FIG. The minimum value of the server device corresponding to the cluster can be stored. In the example of FIG. 3D, it is indicated that the minimum number of servers in the WEB cluster, the application cluster, and the database cluster is 1, and the minimum number of servers in the device cluster is 2.

  When a failure occurs only in the server device S4 described above, as can be seen from FIG. 4B, the number of server devices assigned to each cluster is the server device indicated by the minimum configuration described above. Therefore, the determination in step 302 shown in FIG. 5 is negative. In this case, the change condition determination unit 221 determines whether or not there is a cluster having the minimum configuration based on the comparison result described above (step 304).

  When a failure has occurred only in the server device S4 described above, the number of server devices belonging to the application cluster matches the number of servers indicated in the minimum configuration. As a determination, the process proceeds to step 304, and it is determined whether or not there is a cluster to which a predetermined number (for example, 2) more server devices than the minimum configuration are allocated (step 304).

  In the example described above, since a sufficiently larger number of server devices than the number of servers indicated in the minimum configuration are allocated to the WEB cluster, the change condition determination unit 221 determines that there is a surplus server (in step 304). An affirmative determination) outputs a determination result indicating that an alternative server needs to be allocated in order to maintain stable operation of the cluster system (step 307), and ends the process.

  On the other hand, at least one of the servers belonging to each cluster indicated in the server correspondence table is less than the number of servers indicated in the minimum configuration, as in the case where a failure occurs simultaneously in the server devices S8 and S9 belonging to the device cluster. If it is determined (Yes in Step 302), the change condition determination unit 221 determines that it is difficult to continue the operation of the cluster system without assigning an alternative server, and a determination result indicating that an alternative server needs to be assigned. Is output (step 307), and the determination process is terminated.

On the other hand, the number of server devices belonging to the device cluster coincides with the minimum number of servers, as in the case where failures occur simultaneously in the server devices S1, S4 and S8 respectively belonging to the WEB cluster, application cluster, and device cluster, In addition, there may be a case where no surplus server exists in any cluster.
In such a case, the change condition determination unit 221 evaluates the load applied to each cluster based on, for example, the performance index of each server device indicated in the server state table (step 305), and the evaluation result On the basis of this, it is determined whether or not there is a cluster that becomes a bottleneck in which processing is delayed compared to the processing flow in other clusters due to load imbalance (step 306). When there is a cluster that becomes a bottleneck, the change condition determination unit 221 determines that there is a load imbalance that cannot be ignored (affirmative determination in step 306), and proceeds to step 307, where an alternative server needs to be allocated. A determination result to the effect is output and the process is terminated.

On the other hand, when there is no bottleneck cluster, the change condition determination unit 221 determines that the load imbalance is negligible (negative determination in step 306), and the allocation of the alternative server is unnecessary. A determination result to that effect is output and the process is terminated.
Note that when there is a margin with respect to the minimum number of servers except for the server device in which the failure has occurred, such as when a failure has occurred in the server device S1 belonging to the WEB cluster, the change condition determination unit 221 performs step 304. Is skipped and the process proceeds to step 305 to determine the necessity of the alternative server allocation according to the evaluation result regarding the load imbalance.

As described above, the change condition determination unit 221 executes the above-described procedure based on the information included in the system cluster information, so that it can flexibly cope with various cases and replace the failed server device. It can be determined whether a server should be allocated.
Then, according to the determination result that the alternative server should be assigned, the alternative candidate extraction unit 222 shown in FIG. 2 extracts the alternative server candidate from the server devices belonging to the flexible cluster system according to the procedure described below. To do.

FIG. 6 is a flowchart showing the operation of extracting alternative server candidates.
First, the alternative candidate extraction unit 222 refers to the server state table (step 311) and determines whether or not there is a standby server device (step 312).
In a state where only the server device S4 has a failure, as shown in FIG. 4A, the server devices S11 and S12 are in a standby state (affirmative determination in step 312). These standby servers (server devices S11 and S12) are selected as alternative server candidates (step 313), and the process is terminated.

  On the other hand, when the standby server has already been assigned as an alternative server of any cluster, a negative determination is made in step 312 and the alternative server extraction unit 222 performs the server belonging to each cluster in the same manner as in step 301 described above. The number of devices is compared with the number of servers indicating the minimum configuration of the cluster system (step 314), and it is determined whether there is a cluster to which an excess server device is allocated in the same manner as in step 304 described above. (Step 315).

  For example, as in the example shown in FIG. 4B, if the number of server devices sufficiently larger than the number of servers shown in the minimum configuration is assigned to the WEB cluster, the alternative candidate extracting unit 222 It is determined that there is a surplus server in the cluster (affirmative determination in step 315). In this case, the alternative candidate extraction unit 222 selects all server devices belonging to a cluster including the surplus server (for example, a WEB cluster) as alternative server candidates (step 316), and ends the process.

  On the other hand, in the case of a negative determination in step 315, the alternative candidate extraction unit 222 detects a cluster to which more server devices than the number of servers indicated in the minimum configuration belong based on the comparison result described above (step 315). 317), the server apparatus belonging to the cluster with the lowest priority is selected as an alternative server candidate according to the priority for maintaining the function of the cluster system shown in the priority table included as part of the reference information (step 318), and the processing Exit.

As described above, the alternative candidate extraction unit 222 executes the above-described procedure based on the information included in the system cluster information, so that the alternative candidate extraction unit 222 can be appropriately adapted to the state of the cluster system when allocation of the alternative server is required. Alternative server candidates can be extracted.
When a server device other than the standby server is extracted as an alternative server candidate in this way, the load index calculation unit 223 performs the alternative server candidate for the extracted N alternative server candidates (C _{1 to} C _N ). Load index X _i (i = 1 to N) related to the cluster from which the data is sent, load index Y _i (i = 1 to N) related to the cluster assigned as an alternative server, and load index Z _ij (related to other Q clusters) i = 1 to N and j = 1 to Q) are calculated according to the equations (1) to (3), respectively.

In the expressions (1) to (3), the load index of the cluster K _{j to be} calculated is the performances P _{j1 to} P _{jM of the} server apparatuses S _{j1 to} S _jM belonging to the cluster, and these server apparatuses S _{j1 to} S _{j. It is expressed} using the loads L _{j1 to} L _jM applied to _jM , the performance P _r of the alternative server candidate, and the performance P _{d of the} server device in which the failure has occurred. Note that the sum of Expression (1) includes the performance and load of the alternative server candidate, and the sum of Expression (2) includes the performance and load of the server apparatus in which the failure has occurred.

  On the other hand, when the standby server is extracted as an alternative server candidate, the load index calculation unit 223 loads the load index Y related to the cluster (for example, application cluster) that receives the allocation of the alternative server according to the equations (2) and (3). And load indexes Z relating to other clusters (for example, WEB, DB, and each cluster of the apparatus) are calculated, and these load indexes are provided to the processing of the alternative server determination unit 224.

Based on the load indexes X _i , Y _i , and Z _i obtained for each alternative server candidate C _i in this way, the alternative server determination unit 224 determines the load between clusters when assigning each alternative server candidate. The degree to which the balance is realized is evaluated, and an alternative server is determined based on the evaluation result. At this time, for example, the alternative server determination unit 224 obtains the load index of each cluster realized in the system configuration shown in the basic allocation table shown in FIG. It is possible to compare the obtained load index and evaluate the degree of load balance based on the magnitude of the deviation.

By executing the above-described processing by each unit of the autonomous management unit 217 provided in all the server devices except the server device in which the failure has occurred, the alternative server determination unit 224 of each server device causes the same server device to It is determined as the most suitable alternative server in the operating state of the cluster system at the time.
For example, in the example shown in FIG. 4A, since a failure has occurred in the server device S4 having a performance index of 2.0, the server device S12 is replaced among the two standby server devices S11 and S12. The server server 4 determines that it is a server by the alternative server determination unit 224 of each server device except the server device S4 (see FIG. 2). In response to this determination, the restart processing unit 225 provided in the autonomous management unit 217 of the server device S12 (see FIG. 2) displays the server status table included in the system cluster information as shown in FIG. In addition, information indicating that the server device S12 is in operation is written, and information indicating the correspondence between the server device S12 and the AP server resource information (AP1) is written in the server correspondence table as shown in FIG. Then, change the cluster configuration to the system class information. Next, the restart processing unit 225 acquires the AP server resource information (AP1) stored in the storage area network 201 illustrated in FIG. 1, and provides a function providing processing unit 211 that provides a function as an application server and corresponding functions. A monitoring unit 212 is formed, and the server apparatus S12 is restarted as a part of the application cluster.

  In this way, by sharing the standby server among a plurality of clusters, the shared standby server can be used as an alternative server regardless of the cluster to which the failed server device belongs. Thereby, while maintaining high availability as a cluster system, the number of standby servers provided in the flexible cluster system can be reduced, and the cost required for constructing the cluster system can be greatly reduced.

On the other hand, when an evaluation result indicating that the load balance is largely lost is obtained as an evaluation result regarding the degree of load balance, the substitute server determination unit 224 determines the final result according to the assignment purpose of the substitute server. It is also possible to make a decision not to determine an alternative server.
For example, information indicating the reason why it is necessary to assign an alternative server is received from the change condition determining unit 221 illustrated in FIG. 2, and the alternative server determining unit 224 is based on the information indicating the reason and the evaluation result described above. Determines whether to determine a final alternative server. In other words, when it is indicated that the purpose of allocation of the alternative server is “maintenance of service provision” or “maintenance of high availability”, the alternative server determination unit 224 has a significant load even with the best evaluation result described above. Even if it shows an imbalance, the final alternative server is determined. On the other hand, if the allocation purpose is other than the above-mentioned purpose, the final evaluation result will be the final one if the best evaluation result indicates a significant load imbalance. The process ends without determining a specific alternative server.

Furthermore, in the above-described flexible cluster system, even when a failure occurs in the number of server devices exceeding the number of standby servers, it is possible to flexibly cope with the maintenance of function provision by each cluster.
Hereinafter, as shown by the server status table and the server correspondence table shown in FIGS. 7A and 7B, the functions of the server device S4 and the server device S6 are replaced by the standby servers S11 and S12. Furthermore, an operation for autonomously changing the role assignment across the boundaries of the cluster will be described by taking as an example the case of a quadruple failure in which a failure has occurred in the server devices S8 and S10.

In this example, the device cluster has the minimum configuration, whereas the WEB cluster has more surplus servers allocated than the number of servers “1” indicated in the minimum configuration.
From this, in the autonomous management unit 217 provided in each operating server device, the change condition determination unit 221 determines that an alternative server needs to be allocated (step 304 in FIG. 5). Accordingly, the alternative candidate extraction unit 222 extracts server devices S1 to S3 belonging to the WEB cluster to which the surplus server is assigned as alternative server candidates. And about these server apparatuses S1-S3, based on the load parameter | index obtained by the load parameter | index calculation part 223 shown in FIG. 2, for example, server apparatus S3 is determined as an alternative server by the alternative server determination part 224, In response to this, the restart processing unit 225 provided in the autonomous management unit 217 of the server device S3 performs a process of changing the role of the server device S3 from the WEB server to the device server.

  At this time, the restart processing unit 225 first terminates the function providing processing unit 211 and the function monitoring unit 212 for providing a function as a WEB server, and then releases the WEB server resource information (WEB3) and the server. The correspondence relationship between the WEB server resource information (WEB3) and the server device S3 in the correspondence table is cleared. Thereafter, the restart processing unit 225 newly acquires device server resource information (device 1) (or device server resource information (device 3)) from the storage area network 201, and stores this device server resource information (device 1). Based on this, a function providing processing unit 211 and a function monitoring unit 212 for performing the function as the device server are formed, and the server device S3 is restarted as a device server belonging to the device cluster (see FIGS. 1 and 2).

In this way, even when multiple failures as described above occur, extremely high availability can be realized by interchanging server devices between clusters.
Note that, when the degree of load balance obtained by the alternative server determination unit 224 based on the load index obtained by the load index calculation unit 223 described above is equivalent among a plurality of alternative server candidates, the alternative server determination unit 224 Can select any one based on, for example, the priorities given in advance to each of the server apparatuses S1 to S12. In such a case, it is also possible to assign the one that has been restarted first among the plurality of corresponding server devices as an alternative server.

Further, when the flexible cluster system is normally operated, the storage shown in FIG. 1 is obtained as a part of the reference information constituting the system cluster information by obtaining statistical information indicating the load applied to the server device belonging to each cluster. may be stored in area network 201, the value of the load L _i in accordance with the server apparatus S1~S12 used for processing in the load index calculation unit 223 may be determined based on the above statistical information.

In this way, the load distribution in the flexible cluster system is determined by determining the load index used for the alternative server determination process based on the statistically determined load in the operation state of the flexible cluster system. It is possible to assign an alternative server that reflects the processing and is most suitable for the current state of the flexible cluster system.
In addition, the statistical information accumulated as described above is the index used when optimizing the basic allocation and the minimum configuration of the cluster system shown in FIG. 3B, and when the server device is added to the flexible cluster system. It can also be used as a guide.

As described above, the flexible cluster system according to the present invention continuously uses 24 hours and 365 days while effectively using a minimum amount of hardware, realizing extremely high availability, and reducing the cost required for system construction. A system that can be operated can be realized.
Such a feature is extremely useful in various consumer systems including, for example, an operation system for a network and an online system of a financial institution, and reduces the cost required for system construction and system design and the maintenance cost of the system. It is possible to reduce the total cost of ownership (TCO).

  This can greatly reduce the cost required to open various services that are desired to operate without interruption, such as digital TV relay services and financial services, so it can be used in the broadcasting and financial services industries. Not limited to this, it is possible to promote new entry into various businesses and to activate various business fields.

It is a figure which shows embodiment of the flexible cluster system concerning this invention. It is a figure which shows the detailed structure of the started server apparatus. It is a figure which shows the example of each table contained in system cluster information. It is a figure explaining system cluster information. It is a flowchart showing the operation | movement which determines the necessity of alternative server allocation. It is a flowchart showing the operation | movement which extracts an alternative server candidate. It is a figure explaining system cluster information. It is a figure which shows the structural example of the operation system for the conventional networks.

Explanation of symbols

S1 to S12 Server apparatus 201 Storage area network 202 Fiber channel 211 Function provision processing unit 212 Function monitoring unit 213 Server management processing unit 214 Monitoring information collection unit 215 Failure notification unit 216 Server monitoring unit 217 Autonomous management unit 221 Change condition determination unit 222 Alternative Candidate extraction unit 223 Load index calculation unit 224 Alternative server determination unit 225 Restart processing unit 401, 402 WAN
403 LAN
410 WEB cluster 411 WEB server 420 database cluster 421 database (DB) server 430 application cluster 431 application (AP) server 440, 450, 460 device cluster 441, 451, 461 device server

Claims (5)

In a flexible cluster system comprising a plurality of server devices and a shared disk device shared by these server devices,
The shared disk device is
Server resource storage means for storing server resource information including middleware, application programs and logical IP addresses necessary for each of the plurality of server devices to provide individual functions, corresponding to the individual functions;
System information storage means for storing system cluster information for managing the status of each of the plurality of server devices in the flexible cluster system and the allocation of the server resource information to the server devices;
Each of the plurality of server devices includes:
In accordance with the system cluster information, acquiring server resource information corresponding to the allocated function, and starting means for starting as part of the cluster corresponding to the function;
A failure notification means for detecting a failure related to the function being provided and notifying other server devices of information related to the detected failure as part of the failure information;
Collecting means for collecting failure information related to a failure in the other server device together with information related to the detected failure;
Based on the collected failure information and the system cluster information, determination means for determining whether or not the function to be performed by the device itself in the flexible cluster system should be changed,
In response to the determination result indicating that the function is to be changed, server resource information corresponding to the change destination function is obtained from the server resource storage unit of the shared disk device, and is re-created as a part of the cluster corresponding to the change destination function. A flexible cluster system comprising a restarting means for starting. The flexible cluster system according to claim 1,
The determination means includes
When the collected failure information indicates that a failure has occurred in any of the plurality of server devices constituting the flexible cluster system, the role sharing of the failed server device is assigned based on the system cluster information. Substitution determination means for determining whether or not substitution by another server device is necessary;
Based on the system cluster information, a replacement server that replaces the failed server device is selected based on the system cluster information, and a determination is made that the function is changed when the replacement server is its own device. A flexible cluster system comprising server selection means for outputting results. The flexible cluster system according to claim 2,
The alternative determination means includes
Comparing means for comparing the minimum number of servers indicated by the system cluster information for the cluster to which the failed server device belongs, and the number of operable server devices belonging to the cluster;
When the number of operable server devices belonging to the cluster is equal to the minimum number of servers, the number of server devices that are other clusters and operable than the minimum number of servers indicated in the system cluster information A surplus power detecting means for detecting a cluster having a predetermined number or more,
An imbalance detection means for detecting an imbalance in capacity for each cluster based on information on the load applied to each cluster indicated by the system cluster information and the number of operable server devices belonging to each cluster;
Judgment means for judging whether or not it is necessary to allocate an alternative server for the server device in which the failure has occurred based on the comparison result by the comparison means and the detection results by the remaining power detection means and the imbalance detection means. A flexible cluster system characterized by The flexible cluster system according to claim 2,
The server selection means includes
Based on the system cluster information, standby detection means for detecting a server device whose operating state is standby;
Cluster discrimination means for discriminating a cluster with a large surplus capacity based on the number of surplus servers obtained by subtracting the minimum number of servers indicated by the system cluster information from the number of server devices operating in each cluster;
Candidate extraction means for selecting any one of the clusters determined to have a large surplus power according to the priority order indicated by the system cluster information and extracting server devices belonging to the selected cluster as alternative server candidates;
A flexible cluster system, comprising: a determination unit that determines a server device selected from the detected standby server or alternative server candidate as an alternative server. The flexible cluster system according to claim 4,
The determining means includes
For each of the alternative server candidates, when assigning as an alternative server, load calculation means for obtaining an index indicating the load in at least the allocation destination cluster and the allocation source cluster, respectively;
Selection determining means for preferentially selecting the detected standby server as an alternative server, and selecting one of the alternative server candidates as an alternative server based on the index obtained by the load calculating means when no standby server is detected And a flexible cluster system.

Images