JP2008071156A - Load distribution system, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a load distribution system, capable of performing a large volume of processing and attaining a convergent load distribution. <P>SOLUTION: A load coefficient determination means 260 determines a load coefficient 285 based on network configuration information 250. A load information exchange means 280 sets a state flag 340 to "load accommodation state" when the ratio of the load of an own server to the load of the other server exceeds an accommodation threshold 270, and determines a load target value based on the load of the own server, the load of the other server, and the load coefficient 285. The server name of the other server is stored in a transfer destination server name 335. A load accommodation means 220 determines whether the load of the own server exceeds the load target value or not when the state flag 340 is in the "load accommodation state", and transmits, when it exceeds the target value, a redirect response with the server of the transfer destination server name 335 as a redirect destination to a client. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a load distribution system, method, and program, and more particularly, to a load distribution system, method, and program for distributing server load in a network system having a plurality of servers.

  In a client-server system, there is a load distribution system that distributes processing requests from clients and distributes server load. An example of a conventional load distribution system is described in Non-Patent Document 1. FIG. 13 shows the configuration of a conventional load distribution system. This conventional load distribution system 500 includes a server group 501 composed of a plurality of servers, a load distribution means (load balancer) 510, and a client group 520 composed of a plurality of client terminals.

  Client terminals (520-a to 520-c) in the client group 520 issue a processing request to the server. The load balancer 510 receives a processing request from the client group 520 via the network 530, and distributes the received processing request to any of the servers (501-a to 501-f) in the server group 501. For example, the load balancer 510 randomly selects one server from the servers in the server group 501 when distributing processing requests. The server that receives the processing request from the client executes processing according to the processing request. In Patent Document 1, with such a configuration, the load balancer 510 is used to control processing so as not to concentrate on a specific server, and the server load is distributed within the server group 501.

  Another example of the load distribution system is described in Non-Patent Document 2. FIG. 14 shows a configuration of a load distribution system (DNS round robin) described in Non-Patent Document 2. The load distribution system 600 uses a DNS mechanism to distribute the server load. The client terminals (620-a to 620-c) in the client group 620 inquire the local DNS server 670 about the IP address corresponding to the server's FQDN (Fully Qualified Domain Name) in order to receive services from the server. . The local DNS server 670 determines whether or not there is an entry (corresponding to the correspondence) between the FQDN and the IP address, and if it is not cached, the DNS load balancer 650 that is an upper DNS server. Make an inquiry.

  The DNS load balancer 650 periodically measures the load of each server (601-a, 601-b). When receiving an IP address inquiry from the local DNS server 670, the DNS load balancer 650 returns an IP address of a server with a light load. When the local DNS server 670 receives an IP address response from the DNS load balancer 650, the local DNS server 670 caches the response and stores it internally, and returns the IP address to the client terminal that has made the IP address inquiry. When the correspondence relationship between the FQDN and the IP address is cached, the local DNS server 670 returns the IP address to the client terminal without making an inquiry to the DNS load balancer 650.

  For example, the client terminal 620-a asks the local DNS server 670 about the IP address for the FQDN in order to receive a service from the server. If the correspondence relationship between the FQDN and the IP address is not cached, the local DNS server 670 makes an inquiry to the upper DNS load balancer 650. The DNS load balancer 650 returns the IP address of the server 601-a when the load of the server 601-a is lighter than the load of the server 601-b. The local DNS server 670 caches the response result and returns the IP address of the server 601-a to the client terminal 620-a. In this way, the client terminal 620-a can receive a service from the server 601-a with a light load.

  Another example of the load balancing system is described in Patent Document 1. FIG. 15 shows a configuration of a load distribution (decentralized load distribution) system described in Patent Document 1. The load distribution system 700 includes a plurality of servers 701, and each server 701 includes a processing request search unit 707, a position information analysis unit 709, and a processing request transfer unit 710. The processing request search unit 707 searches for an object corresponding to the request from the client application 703 in the client terminal 702. At that time, the process request search unit 707 refers to the load of the own apparatus (own server) and determines whether or not the process on the own server is possible. When the process request search unit 707 determines that the server load is heavy and the process is impossible, the process request search unit 707 refers to the object position information management table and searches for another server in which an object that can execute the same process exists.

  The process request transfer unit 710 requests a process to another server searched by the process request search unit 707 in order to search for a server that actually performs the process. The other server that has received this request determines whether or not processing by the server is possible by the processing request search unit 707 and returns the result to the requesting server. The processing request transfer unit 710 introduces another server as a transfer destination to the client application 703 if the processing can be performed by another server. If it cannot be processed, the search for a processable server is continued.

  Furthermore, as another example of the load distribution system, there is one described in Patent Document 2. FIG. 16 shows the configuration of the load distribution system described in Patent Document 2. The load distribution system 800 includes a network service server load adjustment device 803. The network service server load adjustment device 803 includes a distribution rate adjustment unit 807, a distribution relay unit 804, and a conversion table vip. Upon receiving an access request from the client group 801, the distribution relay unit 804 hashes the header portion of the received packet, searches the conversion table vip based on the hash value, and transfers the received packet to the searched server.

  The conversion table vip is composed of a plurality of throttles, which is larger than the number of server groups 805. The network service server load adjustment device 803 measures the response time of each server in the server group 805, and the server with a short response time is estimated as a server with a light load, and many throttles are allocated. In the load distribution system 800, many received packets are transferred to a server to which many throttles are assigned, and the load is distributed. In this case, by assigning a throttle from a server having a heavy load to a server having a light load, the load of the assigned server becomes heavy, and there is a risk of causing “vibration” in which the throttle is reassigned. Therefore, in Patent Document 2, the allocation is converged by gradually decreasing the allocation throttle number.

JP 2003-256390 A Japanese Patent Laid-Open No. 11-096128 "Server load balancing technology" by Tony Bruk, O'Reilly Japan, December 2001, pp. 27-34 "Server load balancing technology" by Tony Bruk, O'Reilly Japan, December 2001, pp. 4-8 #DNS Round Robin

  The first problem of the prior art is that there is a limit to processing a large capacity. The reason is that some devices become bottlenecks and the performance limit of the device becomes the overall performance limit. For example, in the conventional load distribution system 500 (FIG. 13) described in Non-Patent Document 1, all requests from clients pass through the load balancer 510, and the performance of the load balancer 510 is limited to the system performance limit. It becomes. In the conventional load distribution system 600 described in Non-Patent Document 2 (FIG. 14), the local DNS server 670 becomes the system performance limit, and the conventional load distribution system 800 described in Patent Document 2 (FIG. 16). Then, the performance of the network service server load adjustment device 803 becomes the system performance limit.

  The second problem is that transmission of load distribution is slow. For example, in the conventional load distribution system 600 (FIG. 14) described in Non-Patent Document 2, load distribution information is managed by a part of DNS servers, and is propagated to the local DNS server 670 by being cached. Since propagation of the cache takes time, it is not possible to respond quickly to fluctuations in server load.

  The third problem is that the load balancing algorithm may not converge. For example, in the conventional load distribution system 700 (FIG. 15) described in Patent Document 1, if a server with a low load exists among a plurality of servers with a high load, the server with a high load around the load By transferring a client request to a low server, a server having a low load becomes rather high because of a load inherited from a plurality of servers. The server then forwards the client's request to another server. By repeating this, the request from the client continues to be transferred and the processing is not completed. In addition, since the transfer work itself is costly, there is a problem that the load is pushed up by the transfer process, not the original process work.

  The fourth problem is that the load distribution system 800 (FIG. 16) of Patent Document 2 performs load distribution considering the vibration of the load, but even in that case, the load between servers is not necessarily equalized. It is. The reason is that, in the load distribution system 800, the load allocation change rate is gradually reduced, but when the change rate is reduced and approaches 0, it is assigned to a plurality of servers on an equal basis. This is because there is no guarantee.

  An object of the present invention is to provide a load distribution system, method, and program capable of solving the above-described problems of the prior art and capable of processing a large volume without bottlenecks.

  Another object of the present invention is to provide a load distribution system, method, and program capable of load distribution without delay of load information.

  Another object of the present invention is to provide a load distribution system, method, and program with guaranteed convergence.

  Another object of the present invention is to provide a load distribution system, method, and program that guarantees that the load between servers is leveled when the load distribution is converged.

  In order to solve the above-described problems of the prior art, a load distribution system according to the present invention includes load coefficient determination means for determining a load coefficient with reference to network configuration information storing a network configuration of a network system having a plurality of servers. Load information exchange means for obtaining load information from another server, setting a load target value based on the load of the own server, the load of the other server, and the load coefficient; and the load of the own server; Compared with the load target value, when the load of the own server is less than or equal to the load target value, the processing for the processing request from the client is executed in the own server, and when the load of the own server exceeds the load target value, Load accommodation means for causing the other server to execute processing corresponding to the processing request.

  The load distribution method of the present invention is a method of distributing a load of a server in a network system having a plurality of servers, wherein the server refers to network configuration information storing a network configuration of the network system, and a load coefficient A load coefficient determination step for determining the load target, and the server exchanges load information with another server, and the load target is determined based on the load of the own server, the load of the other server, and the load coefficient. A load target value setting step for setting a value; and the server compares the load of the server with the load target value, and executes a process for a processing request from a client on the server, or the other A processing server determining step for determining whether to execute the server, and the server executes the server at the processing server determining step. When determining, the processing executed in the local server, upon determining that is executed by the other server, the process is characterized in that is executed by the other server.

  The program according to the present invention is a network system having a plurality of servers, and causes the server to execute a process of distributing the load on the server. The server refers to the network configuration information that stores the network configuration of the network system. A load factor determination step for determining a load factor, load information is exchanged with another server, a load target based on the load of the own server, the load of the other server, and the load factor The load target value setting step for setting the value, the load of the own server and the load target value are compared, and the processing for the processing request from the client is executed by the own server or is executed by the other server A processing server determination step for determining whether or not to execute on the own server in the processing server determination step. When, the processing executed in the local server, upon determining that is executed by the other server, the process is characterized in that is executed by the other server.

  In the load distribution system, method, and program of the present invention, the load coefficient for determining the load transfer amount is determined based on the network configuration, and based on the determined load coefficient and the loads of the own server and other servers. Determine the load target value. If the load on the local server is higher than the load target value, the processing for the processing request from the client is executed on the other server and the load is transferred to the other server. Execute the process. In the present invention, the load coefficient is determined based on the network configuration, and when the load transfer is repeated, the load target value is set so that the load of each server converges to a similar load. By setting, it is possible to achieve load distribution in which the load converges and the load between servers is leveled. In this case, since processing requests are not concentrated on a specific device, a large-capacity processing can be realized without a bottleneck.

  In the load distribution system of the present invention, the load information exchanging means subtracts a value obtained by multiplying the load of the own server by a value obtained by multiplying the difference between the load of the own server and the load of another server by the load coefficient. A configuration in which the load target value is set can be adopted. Specifically, the load target value may be set by such calculation.

  In the load distribution system of the present invention, the load information exchanging means indicates whether or not to perform load accommodation when it is determined that the ratio of the load of the local server to the load of another server exceeds a predetermined threshold value. The load target value is set by setting a state flag to “load accommodation state”, and when the state flag is “load accommodation state”, the load accommodation unit sets the load of the server itself and the load target value. It is possible to adopt a configuration for comparing the above. In this case, the load information exchanging means sets the load target value when the load of the own server is higher than the load of the other server and the ratio exceeds a predetermined threshold, and sets the status flag to “load It is set to “accommodating state”, and the server state is shifted to a state in which the load is accommodated between servers. The load accommodation means, when receiving a processing request from a client, compares the load of the local server with the load target value and responds to the processing request from the client when the status flag is “load accommodation state”. It is determined whether the processing to be performed is performed by the own server or another server. With such processing, load distribution can be realized.

  In the load distribution system of the present invention, the load accommodation means indicates that the state flag indicates a state in which load accommodation is not performed when the load of the local server is equal to or less than the load target value as a result of the comparison. Can be adopted. As a result of the load accommodation means comparing the load of the own server and the load target value, if the load of the own server is equal to or less than the load target value, it is determined that the processing for the processing request from the client is performed by the own server. There is no load accommodation. In such a case, the state flag may be set to “normal state” to cancel the “load accommodation state”.

  In the load distribution system of the present invention, the load information exchanging means may adopt a configuration for acquiring load information from the other server when the state flag is not “load accommodation state”. When the server state is not the “load accommodation state”, the load information exchanging unit periodically acquires load information from another server, and compares the acquired load of the other server with the load of the own server. As a result of the comparison, when the load ratio exceeds the threshold value, the load target value is set based on the load of the local server, the load of the other server, and the load coefficient. By doing in this way, the load of other servers can be acquired without delay of load information, and a load target value can be set based on the load of other servers.

  In the load distribution system of the present invention, the load information exchanging means stores the information indicating the other server in the transfer destination server storage unit when the status flag is set to “load accommodating state”, and the load accommodating means When the load of the own server exceeds the load target value, it is possible to refer to the transfer destination server storage unit and execute a process for the processing request on the server stored in the transfer destination server storage unit. .

  In the load distribution system of the present invention, when the load information exchanging means determines that the ratio of the load of the local server to the load of the other server exceeds a predetermined threshold, the load can be accommodated to the other server. When transmitting a load accommodation request for inquiring whether or not, and receiving a response for granting the request, the status flag is set to “load accommodation state” and information indicating the other server is transferred A configuration of storing in the destination server storage unit can be adopted. Further, when the load information exchanging means receives the load accommodation request from another server, the load information exchanging means determines whether or not to grant the permission by referring to the load information of the own server. A configuration for returning a rejection response can be employed. The load transfer destination server has a lower load than the load transfer server, but it may be inappropriate to accept the load, for example, when a load increase is expected thereafter. Therefore, a load accommodation request is transmitted in advance, and an inquiry is made to other servers as to whether or not the load can be accepted. The other server that has received the load accommodation request determines whether or not to grant the load accommodation request based on the load information or the like. When the server that issued the load accommodation request obtains a response for granting the license to the load accommodation request, the other server that has transmitted the authorization response stores it as a load transfer destination server. By doing in this way, load distribution according to the situation of the other party becomes possible.

  In the load distribution system of the present invention, when the load of the own server exceeds the load target value, the load accommodation unit transmits a redirect response with another server as a redirect destination to the client in response to the processing request, A configuration in which processing for the processing request is executed by another server can be employed. In this case, after receiving the redirect response, the client can obtain a processing response from the redirect server by transmitting a processing request to the redirect destination server.

  In the load distribution system of the present invention, when the load of the own server is equal to or less than the load target value, the load accommodation unit activates an application corresponding to the processing request by a dispatch unit, and displays the processing result of the application as the processing result. It is possible to adopt a configuration for transmitting as a response to the client.

であることから、各サーバの負荷を平準化することができることがわかる。 In the load distribution system of the present invention, the load coefficient determination unit associates rows and columns with each server based on the network configuration information, and sets elements corresponding to servers in a connection relationship between servers to -1. A connection matrix A is created in which the element corresponding to the server that is not connected is 0, and the diagonal element is the number of servers connected to the server corresponding to the row. Using E as a unit matrix, eigenvalues k (l) of G = (− lA + E) are obtained, and the maximum of l of 0 <l <1 where all eigenvalues k (l) are greater than −1 and less than or equal to 1 A configuration can be adopted in which a value is obtained and the maximum value of l is determined as a load coefficient. By creating the connection matrix A in this way and selecting l for which all eigenvalues of the matrix G are larger than −1 and smaller than 1 for G = (− 1A + E), load accommodation is repeated. The load on each server can be converged. Also, one of the eigenvalues of the matrix G is known to be “1”, and its eigenvector is

Therefore, it can be seen that the load on each server can be leveled.

  In the load distribution method and program of the present invention, in the processing server determination step, the load of the own server is compared with the load target value, and if the load of the own server is equal to or less than the load target value, the process is performed. It is possible to adopt a configuration in which it is determined that the other server is to execute when the load of the own server exceeds the target value.

  In the load distribution method and program according to the present invention, in the load target value setting step, a value obtained by multiplying a difference between the load of the local server and the load of another server by the load coefficient is subtracted from the load of the local server. A configuration in which a value is set as the load target value can be employed.

  In the load distribution method and program of the present invention, in the load target value setting step, if it is determined that the ratio of the load of the own server and the load of the other server exceeds a predetermined threshold, whether or not to perform load accommodation Is set to “load accommodation state”, the load target value is set, and when the processing request is received, the state flag is referred to and the state flag is “load accommodation state”. For example, a configuration for executing the processing server determination step can be adopted.

  In the load distribution method and program of the present invention, in the processing server determination step, when it is determined that the processing for the processing request is executed by the own server, the state flag is set to “normal state” indicating a state in which load accommodation is not performed. Can be adopted.

  The load distribution method and the program according to the present invention adopt a configuration in which, in the load target value setting step, when the state flag is set to “load accommodation state”, information indicating the other server is stored in the transfer destination server storage unit. it can.

  In the load distribution method and program according to the present invention, when the server determines that another server is to execute processing for the processing request, the server is stored in the transfer destination server storage unit with reference to the transfer destination server storage unit. A configuration in which the server executes the process can be adopted.

  In the load distribution method and program according to the present invention, in the load target value setting step, if it is determined that the ratio of the load of the local server to the load of the other server exceeds a predetermined threshold value, the load is distributed to the other server. When a load accommodation request for inquiring whether or not accommodation is possible is transmitted and a response indicating that the request is granted is received, the status flag is set to “load accommodation state” and information indicating the other server is set. A configuration in which the load target value is set by storing in the transfer destination server storage unit can be adopted.

  In the load distribution method and program according to the present invention, when the server receives the load accommodation request from another server, the server determines whether or not to grant the license by referring to the load information of the own server. When rejecting, a configuration in which a rejection response is returned can be adopted.

  In the load distribution method and program of the present invention, when it is determined in the processing server determination step that another server is to execute, a redirect response with the other server as a redirect destination is transmitted to the client in response to the processing request, A configuration in which processing for the processing request is executed by another server can be employed.

  In the load distribution method and program according to the present invention, when it is determined in the processing server determination step that the server itself is executed, the application corresponding to the processing request is started by dispatch means, and the processing result of the application is returned to the client. The structure which transmits as can be employ | adopted.

  In the load distribution method and program according to the present invention, in the load coefficient determination step, the server corresponds to each server with a row and a column based on the network configuration information, and corresponds to a server having a connection relationship between servers. The connection matrix A is created with the element to be −1, the element corresponding to the server that is not connected to 0 as the number of servers connected to the server corresponding to the row as the diagonal element, and the created connection matrix A On the other hand, with e as a variable and E as a unit matrix, eigenvalues k (l) of G = (− 1A + E) are obtained, and all eigenvalues k (l) are greater than −1 and less than or equal to 1 <0 <l <1 The maximum value of l can be obtained, and the maximum value of l can be determined as a load coefficient.

  In the load distribution system, method, and program of the present invention, the load coefficient for determining the load transfer amount is determined based on the network configuration, and based on the determined load coefficient and the loads of the own server and other servers. Determine the load target value. If the load of the own server is higher than the load target value, the process for the processing request from the client is executed by another server, and if the load is equal to or less than the load target value, the process is executed in the own server. In the present invention, the load coefficient is determined based on the network configuration, and when the load transfer is repeated, the load target value is set so that the load of each server converges to a similar load. By setting, it is possible to achieve load distribution in which the load converges and the load between servers is leveled. In this case, since processing requests are not concentrated on a specific device, there is no bottleneck device, and a large-capacity processing can be realized.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 shows the configuration of the load distribution system according to the first embodiment of the present invention. The load distribution system 100 includes a plurality of servers 200 connected to each other via a network 310. The server 200 includes a central processing unit (CPU) 240 that operates by program control, a transport stack 230, a load accommodation unit 220, a dispatch unit 210, an application group 300, a load measurement unit 290, a load information exchange unit 280, and a load coefficient. A determination unit 260 is included. In addition, it has storage means such as a load target value 330, a transfer destination server name 335, a status flag 340, an accommodation threshold value 270, a load coefficient 285, and a register for storing the network configuration information 250.

  The server 200 receives a processing request from a client (not shown), performs a service according to the request, and transmits a processing response to the client. The transport stack 230 is connected to the network 310 and accepts and responds to various requests from clients and other servers. The load measuring means 290 observes the status of the central processing unit 240 and the transport stack 230 and measures the load on the server 200. The status flag 340 stores a flag for managing the device status. The flags stored in the status flag 340 include a “load accommodation state” indicating a state in which load transfer (load accommodation) is performed between servers, and a “normal state” indicating a state in which load transfer is not performed.

  If the status flag 340 is “normal state”, the load information exchanging unit 280 periodically communicates with other servers through the network 310 to exchange the load information obtained by the load measuring unit 290. If the status flag 340 is “load accommodation state”, the load information is not exchanged. The load information exchanging means 280 exchanges load information, for example, at the timing when the timer 320 that periodically times out generates a timeout. When the load information exchanging means 280 obtains the load of the other server, it calculates the ratio of the load of the own server and the load of the other server, and the value becomes a decision threshold when starting the load transfer. It is determined whether or not the value is larger than 270. When the load information exchange unit 280 determines that the load ratio is larger than the accommodation threshold value 270, the load information exchange unit 280 transmits a load accommodation request to another server. When the load information exchanging means 280 receives a response to permit the load accommodation request, it stores information indicating another server in the transfer destination server name 335 indicating the load transfer destination, and sets the status flag 340 to “load Set to "Flexible state".

  The load coefficient determining unit 260 determines the load coefficient 285 with reference to the network configuration information 250 including information on the connection status between servers. The load coefficient 285 is a proportional coefficient for determining how much of the load difference is transferred to another server when the load of the former server is high and the load of the other server is compared. is there. The load information exchanging means 280 sets the load target value 330 based on the load of the own server, the difference between the load of the own server and the load of the other server, and the load coefficient 285 when shifting to the “load accommodation state”. To do. Specifically, a value obtained by subtracting a value obtained by multiplying the load between the load of the own server and the load of the other server by the load coefficient 285 from the load of the own server is set as the load target value 330. The load target value 330 indicates the target value of the server load in the “load accommodation state”.

  When the transport stack 230 receives a request from a client, the transport stack 230 passes the received request to the load accommodation unit 220. The load accommodation unit 220 refers to the state flag 340 and determines whether it is “normal state” or “load accommodation state”. If it is “normal state”, the load accommodation unit 220 decides to execute the process in its own server without transferring the load, and passes the request from the client to the dispatch unit 210. If the state flag 340 is “load accommodation state”, it is determined whether or not the server load measured by the load measuring unit 290 exceeds the load target value 330. If it exceeds, it is determined to transfer the load, and a redirect response to the server indicated by the transfer destination server name 335 is returned to the client. If the load target value 330 is not exceeded, it is determined that the process is to be executed in its own server, and the request from the client is passed to the dispatch unit 210.

  When the dispatch unit 210 receives a processing request from a client from the load accommodation unit 220, the dispatch unit 210 selects an application that provides a service required by the processing request from the application group 300 and starts it. The application group 300 executes a service based on the client request received from the dispatch unit 210 and returns the result to the dispatch unit 210 as a response to the client.

  FIG. 2 shows the overall processing flow of the load distribution system 100. Server 200 waits for an event such as reception of a communication packet or a trigger from timer 320 (step Z110). When an event occurs, it is dispatched to an appropriate process according to the event that has occurred (step Z120). When the timer 320 is periodically started, a load information exchange process is executed (step Z130). When a load information exchange request is received from another server, load information provision processing is executed (step Z140). When a load accommodation request is received from another server, load accommodation request processing is executed (step Z150). If a request from the client is received, client request processing is executed (step Z160).

  FIG. 3 shows a detailed procedure of the load information exchange process in step Z130. The load information exchanging means 280 periodically performs load information exchanging processing with other nearby servers by starting the timer. When there are a plurality of other servers connected to the server 200 in the vicinity of the server 200, the load information exchange is periodically performed with each of the servers. Specifically, when there are two servers A and B in the vicinity, the timing when the load information should be exchanged with the server A by the timer 320 and between the server B and the server B A timeout occurs at both timings at which information should be exchanged, and load information is periodically exchanged with server A, and load information is also exchanged with server B periodically.

  In the load information exchange process, the load information exchange unit 280 refers to the state flag 340 and determines whether the state flag 340 is “normal state” or “load accommodation state” (step A210). If it is “load accommodation state”, the processing is completed as it is. If the status flag 340 is “normal state”, the load information exchanging means 280 communicates with other servers in the vicinity that are subject to the load information exchange via the network 310, transmits a load information request, Load information is acquired from the server (step A220).

  When the load information exchange unit 280 acquires the load information, the load information exchange unit 280 determines whether or not the load ratio exceeds the accommodation threshold value 270 based on the acquired load information of the other server and its own load information. (Step A230). More specifically, the load of the server 200 (S) is changed to S.I. l, the load of the other server (S ') is changed to S'. l and the value of the accommodation threshold 270 as D, S.I. l / S '. It is determined whether l> D is satisfied. If not, the process is completed as it is. If it is established, the load information exchanging means 280 requests load accommodation from other servers (step A240). When receiving a response to the load accommodation request from another server (step A250), the load information exchanging means 280 determines whether or not permission for load accommodation has been obtained (step S260). If permission is not obtained from another server, the processing is completed as it is.

  When the permission is obtained from another server, the load information exchanging means 280 sets the status flag 340 to “load accommodation state” and shifts to the load accommodation state (step A270). At that time, the load information exchanging means 280 stores the server name of the other server in the transfer destination server name 335. Further, the load target value 330 is determined based on the load of the own server, the load of other servers, and the load coefficient 285 measured by the load measuring unit 290. More specifically, kl is a coefficient value stored in the load coefficient 285, and S.I. l-kl × (S.l-S′.l) is set as the load target value.

  FIG. 4 shows a detailed procedure of the load information providing process in step Z140 of FIG. The load information request transmitted at step A220 in FIG. 3 is received by the load information exchanging means 280 of the other server 200. When the load information exchanging means 280 receives the load information request (step B110), the load measuring means 290 measures the load of its own server (step B120). The load information exchanging unit 280 transmits the load information measured by the load measuring unit 290 to the server that issued the load information request (step B130).

  FIG. 5 shows a detailed procedure of the load accommodation request process in step Z150 of FIG. The server 200 receives the load accommodation request transmitted from the other server in Step A240 in FIG. 3 (Step C110). The load information exchanging means 280 determines whether or not to accept the load acceptance based on the server state (step C120). In this acceptance determination, the server load measured by the server load measuring unit 290 is referred to. For example, when the server load is light (when it is lower than a predetermined threshold), it is determined that the load acceptance is accepted. However, even in that case, if the application used by the client is not yet installed in consideration of the server state and the installation time is taken into consideration, it is rejected if it is not a good idea to accept the load accommodation request. When the load information exchanging means 280 determines that it is acceptable, it returns a permission response to the server that issued the request (step C130), and when it determines that it is not acceptable, it returns a rejection response to the server that issued the request (step C130). C140).

  FIG. 6 shows a detailed procedure of the client request process in step Z160 of FIG. The transport stack 230 receives the processing request from the client (step D110). The processing request from the client received by the transport stack 230 is passed to the load accommodation unit 220. The load accommodation unit 220 refers to the state flag 340 and checks whether the state flag 340 is “normal state” or “load accommodation state” (step D120).

  When the load accommodation unit 220 determines that the status flag 340 is “normal state”, the load accommodation unit 220 determines that the load is not transferred, and passes the request from the client to the dispatch unit 210. The dispatch unit 210 selects an appropriate server group from the application group 300 based on the received request from the client, and activates the application (step D160). The activated application executes appropriate processing according to the processing request, and returns a response (processing result) to the dispatch unit 210 (step D170). The response created by the application is returned to the client via the dispatch unit 210, the load accommodation unit 220, and the transport stack 230 (step D180).

  When the load accommodation unit 220 determines in step D120 that the status flag 340 is “load accommodation state”, the load measurement unit 290 measures the load of the server 200 (step D130). The load accommodation unit 220 compares the measured load of the server 200 with the load target value 330, and determines whether or not the load of the server 200 is equal to or greater than the load target value 330 (step D140). When determining that the server load is not equal to or higher than the load target value 330, the load accommodation unit 220 determines not to transfer the load, and resets the status flag 340, the load target value 330, and the transfer destination server name 335 (step D155). . Thereafter, the process proceeds to step D160, the request from the client is passed to the dispatch unit 210, the application is started by the dispatch unit 210, and a response to the request is returned to the client.

  As a result of the comparison in step D140, when the load of the server 200 is equal to or higher than the load target value 330, it is determined to transfer the load. In this case, the load accommodation unit 220 refers to the transfer destination server name 335 and creates a redirect response with the server indicated by the transfer destination server name 335 as the redirect destination (step D150). The redirect response created by the load accommodation unit 220 is returned to the client via the transport stack 230.

  FIG. 7 shows an operation procedure of the client. The client operates according to the procedure shown in FIG. 7 as defined by redirection such as HTTP. That is, the client issues a processing request to the predetermined server 200 (step E110). Upon receiving this processing request, the server 200 returns a processing result or a redirect response according to the procedure shown in FIG. The client receives the response from the server 200 (step E120), and determines the type of the received response (step E130). If the response received from the server 200 is a processing result, the processing result is received (step E140) and the processing is terminated. If the response is a redirect response, the client changes the success server to the designated redirect destination (step E150), returns to step E110, and transmits a processing request to the server at the redirect destination.

  Next, details of the determination of the load coefficient 285 will be described. FIG. 8 shows the procedure. The load coefficient 285 is used to determine a load target value, and is used to determine how much load is transferred to another server. The load coefficient determination unit 260 reads the network configuration information 250 at a predetermined timing, for example, when the system is started up (step F110). The load coefficient determination unit 260 refers to the read network configuration information 250 and creates a connection matrix representing the network configuration (connection state between servers) (step F120).

上記接続行列Ａにおいて、Ｎ（ｉ）は、サーバｉと接続されているサーバの集合である。すなわち、システムを構成するｎ台のサーバうちのｉ番目のサーバとｊ番目のサーバとが接続関係にある場合には、（ｉ，ｊ）成分及び（ｊ，ｉ）成分を「−１」にし、対角成分（ｉ，ｉ）をｉ番目に接続されたサーバに接続されたサーバの数とする。接続行列Ａでは、任意の要素α_i,jについて、

が成立する。 The connection matrix A is defined as follows.

In the connection matrix A, N (i) is a set of servers connected to the server i. That is, when the i-th server and the j-th server among n servers constituting the system are connected, the (i, j) component and the (j, i) component are set to “−1”. , The diagonal component (i, i) is the number of servers connected to the i-th connected server. In the connection matrix A, for any element α _{i, j} ,

Is established.

  The load coefficient determination unit 260 obtains an eigenvalue of G = (− 1A + E) (E is a unit matrix) for the connection matrix A. In general, an n × n square matrix has n eigenvalues. Assuming that the eigenvalue is k (l), the value of l in the range of 0 <l <1 such that all the eigenvalues k (l) are larger than −1 and 1 or less is obtained, and the maximum value among them is determined. Obtained (step F130). The maximum value of l in the range of 0 <l <1 obtained in this way is stored in the load coefficient 285 as a load transfer coefficient (step F140).

  The effect of this embodiment will be described. In this embodiment, all requests from the client are transmitted to each server without passing through a specific device. For this reason, there is no device that becomes a bottleneck, and it is possible to cope with a large load. Moreover, in this embodiment, servers exchange load information periodically by the load information exchanging means 280 to set the load target value 330. For this reason, the load information of the server does not become information older than the cycle. Therefore, in this embodiment, load distribution based on the latest load information is possible as compared with DNS round robin in which old server information remains in the cache.

  Further, in the present embodiment, for the connection matrix A, the load coefficient 285 is obtained so that the eigenvalue of G (−1A + E) is greater than −1 and equal to or less than 1, and the load target value is obtained using this load coefficient 285. 330 is determined. By doing in this way, when load transfer is repeated with other servers, it is possible to realize load distribution in which the server load is leveled. Hereinafter, the leveling of the server load will be described in detail.

とおく。ｔ＋１回目の負荷移管によるサーバｉの負荷は、ｔ回目の負荷移管によるサーバｉの負荷ｘ_i,tを用いて、

と表すことができる。ｌは、負荷係数２８５の値である。これを変形すると、ｔ＋１回目の負荷と、ｔ回目の負荷との差は、

と表すことができる。接続行列を用いると、各サーバにおけるｔ＋１回目の負荷と、ｔ回目の負荷との差は、

と表すことができる。すなわち、

である。 The load of the server i resulting from the t-th load transfer is represented by x _{i, t} . The load resulting from the tth load exchange protocol on each server is

far. The load of the server i by the t + 1-th load transfer is calculated by using the load x _{i, t} of the server i by the t-th load transfer,

It can be expressed as. l is the value of the load coefficient 285. When this is transformed, the difference between the t + 1th load and the tth load is

It can be expressed as. Using the connection matrix, the difference between the t + 1th load and the tth load on each server is

It can be expressed as. That is,

It is.

は収束するので、負荷係数決定手段２６０が、固有値がこの範囲に収まるようにｌを設定すれば、

は収束する。また、このとき、全ての固有値が０以上、かつ、１以下であれば、負荷は単調に収束する。固有値については、

のため、固有値のうち、ひとつは１であることがわかっており、その固有ベクトルは、

である。初期負荷ベクトルｘ_０を固有ベクトルの一次結合に分解し、ｔを無限大に近付けたとき負荷ベクトルｘ_ｔのリミットを考えると、１以外の固有値は−１よりも大きく、かつ、１よりも小さいので０に収束し、最終的に、固有ベクトルの一次結合のうちの

の成分のみが残る。従って、負荷移管を繰り返すことで、負荷は平準化する。 In general, if the eigenvalue of G = (− 1A + E) is greater than −1 and less than or equal to 1,

Converges, so if the load coefficient determining means 260 sets l so that the eigenvalue falls within this range,

Converges. At this time, if all eigenvalues are 0 or more and 1 or less, the load converges monotonously. For eigenvalues,

Therefore, one of the eigenvalues is known to be 1, and its eigenvector is

It is. The initial load vector x ₀ is decomposed into a linear combination of eigenvectors, given the limits of the load vector x _t when close to t to infinity, 1 except eigenvalue greater than -1 and is smaller than 1 Converges to 0 and finally out of the linear combination of eigenvectors

Only the ingredients remain. Therefore, the load is leveled by repeating the load transfer.

上記接続行列Ａでは、１行目及び１列目がサーバＡ２００−ａに対応し、２行目及び２列目がサーバＢ２００−ｂに対応している。また、３行目及び３列目は、サーバＣ２００−ｃに対応している。 Hereinafter, a specific example will be described. FIG. 9 illustrates the configuration of the load distribution system according to the first embodiment. In the present embodiment, as the server 200, three servers, a server A 200-a, a server B 200-b, and a server C 200-c are considered. Further, it is assumed that the server A 200-a and the server B 200-b, and the server B 200-b and the server C 200-c mutually recognize each other and regularly exchange load information. In this configuration, the connection matrix A created by the load coefficient determination unit 260 by reading the network configuration information 250 (FIG. 1) in step F110 of FIG. 8 is as follows.

In the connection matrix A, the first row and the first column correspond to the server A 200-a, and the second row and the second column correspond to the server B 200-b. The third row and the third column correspond to the server C200-c.

の固有値を求める。その結果、ｋ_０＝１−３ｌ、ｋ_１＝−ｌ＋１、ｋ_２＝１となる。単調に収束させるために、全ての固有値を０以上でかつ１以下にするには、０≦ｌ≦（１／３）かつ０≦ｌ≦１なので、これらを満たすｌの最大値は１／３である。負荷係数決定手段２６０は、図８のステップＦ１４０で、「１／３」を負荷係数２８５に格納する。 The load coefficient determination means 260 uses the connection matrix A created in step F130,

Find the eigenvalue of. As a result, k ₀ = 1-3l, k ₁ = -l + 1, k ₂ = 1. In order to converge monotonously, to make all eigenvalues 0 or more and 1 or less, 0 ≦ l ≦ (1/3) and 0 ≦ l ≦ 1, so the maximum value of l satisfying these is 1/3 It is. The load coefficient determining means 260 stores “1/3” in the load coefficient 285 in step F140 of FIG.

  Next, it is assumed that the clients 400-a and 400-b are accessing the server A 200-a, and the clients 400-c and 400-d are accessing the server B 200-b. Further, it is assumed that the clients 400-e and 400-f are accessing the server C200-c. Initially, each client is accessing at 10 accesses / second. It is assumed that the access frequency of the client 400-c accessing the server B 200-b has increased to 30 accesses / second. Then, the load on the server B 200-b increases from 20 access / second to 40 access / second.

  The server A 200-a and the server B 200-b regularly exchange load information with each other by the load information exchanging means 280 (step Z130 in FIG. 2). The server B 200-b obtains the load information from the server A 200-a to know that the load of the server A 200-a is 20 accesses / second (step A220 in FIG. 3). Assuming that the accommodation threshold 270 is D = 1.5, the ratio of the load on the server B 200-b to the load on the server A 200-a is (40/20)> 1.5. The server B 200-b requests the server A 200-a for load accommodation.

  When the server A 200-a receives the load accommodation request in step C110 of FIG. 5, the server A 200-a checks whether or not the load can be accepted in step C120. If there is no problem due to load transfer, the server A 200-a makes a permission response to the load accommodation request in step C130. When the server B 200-b receives the permission response from the server A 200-a in step A250 of FIG. 3, the load information exchanging unit 280 sets the status flag 340 to “load accommodation state” and sets the transfer destination server name 335 as the destination server name 335. Set to server A. In step A270, the load target value 330 is set to 40− (1/3) × (40−20) = 33.

  When the server B 200-b receives the request processing from the client in step D110 of FIG. 6, it checks the status flag 340 in step D120. Since the status flag 340 is “load accommodation state”, the process proceeds to step D130 and the load measurement unit 290 measures the load on the server B 200-b. As a result, for example, it can be seen that the load is 40 accesses / second. The load accommodation unit 220 compares the load (40 accesses / second) with the load target value 330 (33 accesses / second). Since the load is larger than the target value, the process proceeds from step D140 to step D150, and the server B 200-b creates a redirect response to the server A 200-a indicated by the transfer destination server name 335 by the load accommodation unit 220. A redirect response is returned to the client through the transport stack 230.

  The client receives a redirect response from the server B 200-b in step E120 of FIG. In step E150, the client that has received the redirect response changes the access destination to the instructed redirect destination server A 200-a, returns to step E110, and transmits a processing request to the server A 200-a. In this way, load distribution is realized by redirecting a processing request from a client to the server B 200-b to the server A 200-a until the load drops to the target value.

Next, another specific example will be described. FIG. 10 illustrates the configuration of the load distribution system according to the second embodiment. In the present embodiment, as the server 200, four servers, a server A 200-a, a server B 200-b, a server C 200-c, and a server D 200-d are considered. These mutually recognize each other and regularly exchange load information. In this configuration, the connection matrix A that the load coefficient determination unit 260 reads and creates the network configuration information 250 (FIG. 1) in step F110 of FIG. 8 is as follows.

  When the load coefficient determination unit 260 sets l = 0.4, among eigenvalues excluding 1 in the matrix G = (− 1A + E), the eigenvalue having the maximum absolute value is −0.84, and this value is Greater than -1 and less than or equal to 1. Therefore, the load converges. FIG. 11 shows how the load on each server changes. In this example, the accommodation threshold D is 1.2. In the figure, the vertical axis represents the load on the server, and the horizontal axis represents time. Referring to the figure, when a load is applied to one of the four servers, the load is balanced with the other three servers, so that the load is finally leveled between the servers. It can be seen that As a comparative example, FIG. 12 shows how the server load changes when l = 0.8 under the same conditions. In this case, the eigenvalue having the maximum absolute value is −2.7. Therefore, even if load transfer is repeated, the load on each server fluctuates and is not leveled as shown in FIG.

  Regarding the accommodation threshold value 270, since the load transfer to other servers is performed when the load ratio exceeds the accommodation threshold value, the accommodation threshold value is finally equalized load It shows the variation among servers. If this value is 1, load transfer frequently occurs and efficiency is lowered. Therefore, the accommodation threshold is preferably set to a value of about 1.1 to 1.5.

  As described above, the present invention has been described based on the preferred embodiments. However, the load distribution system, method, and program of the present invention are not limited to the above embodiments, and various configurations are possible from the configurations of the above embodiments. Those modified and changed as described above are also included in the scope of the present invention.

  The present invention can be applied to a use such as a load balancer that processes a large amount of load from a client. Moreover, it is applicable also to uses, such as a load balancer which processes a lot of loads from a sensor or RFID.

The block diagram which shows the structure of the load distribution system of 1st Embodiment of this invention. The flowchart which shows the flow of the whole process of a load distribution system. The flowchart which shows the detailed procedure of a load information exchange process. The flowchart which shows the detailed procedure of a load information provision process. The flowchart which shows the detailed procedure of a load accommodation request | requirement process. The flowchart which shows the detailed procedure of a client request process. The flowchart which shows the operation | movement procedure of a client. The flowchart which shows the procedure of the determination of a load coefficient. 1 is a block diagram illustrating a configuration of a load distribution system according to a first embodiment. FIG. 3 is a block diagram illustrating a configuration of a load distribution system according to a second embodiment. 9 is a graph showing a change in load on each server in Example 2. The graph which shows the mode of the load change of each server in a comparative example. The block diagram which shows the structure of the conventional load distribution system. The block diagram which shows the structure of the conventional load distribution system (DNS round robin). The block diagram which shows the structure of the conventional load distribution (decentralized type load distribution) system. The block diagram which shows the structure of the conventional load distribution system.

Explanation of symbols

100: Load distribution system 200: Server 210: Dispatch means 220: Load accommodation means 230: Transport stack 240: CPU
250: Network configuration information 260: Load coefficient determination means 270: Interchange threshold 280: Load information exchange means 285: Load coefficient 290: Load measurement means 300: Application group 310: Network 320: Timer 330: Load target value 335: Transfer Destination server name 340: Status flag 400: Client

Claims (35)

Load coefficient determining means for determining a load coefficient with reference to network configuration information storing a network configuration of a network system having a plurality of servers;
Load information exchange means for obtaining load information from another server, and setting a load target value based on the load of the own server, the load of the other server, and the load coefficient;
The load of the own server is compared with the load target value, and when the load of the own server is equal to or less than the load target value, the processing for the processing request from the client is executed in the own server. And a load accommodation unit that causes the other server to execute processing corresponding to the processing request when the value exceeds the value.   The load information exchanging means sets, as the load target value, a value obtained by subtracting a value obtained by multiplying the load of the own server and the load of another server by the load coefficient from the load of the own server. The load distribution system according to claim 1.   When the load information exchanging means determines that the ratio of the load of the own server and the load of the other server exceeds a predetermined threshold, a status flag indicating whether or not to perform load accommodation is set to “load accommodation state”. The load target value is set by setting the load target value, and the load accommodation unit compares the load of the server with the load target value when the status flag is "load accommodation state". Or the load distribution system of 2.   The load accommodation means sets the state flag to a “normal state” indicating a state in which load accommodation is not performed when the load of the server is equal to or less than the load target value as a result of the comparison. Load balancing system described in 1.   5. The load distribution system according to claim 3, wherein the load information exchanging unit acquires load information from the other server when the state flag is not “load accommodation state”.   The load information exchanging means stores the information indicating the other server in the transfer destination server storage unit when the status flag is set to “load accommodating state”. 6. The process according to claim 3, wherein when the target value is exceeded, the process corresponding to the processing request is executed by a server stored in the transfer destination server storage unit with reference to the transfer destination server storage unit. Load balancing system.   When the load information exchanging means determines that the ratio of the load of the local server and the load of the other server exceeds a predetermined threshold, the load interchange request for inquiring whether or not the other server can accommodate the load. When the response indicating that the request is granted is received, the status flag is set to “load accommodation state”, and information indicating the other server is stored in the transfer destination server storage unit. The load distribution system according to claim 6.   When the load information exchange unit receives the load accommodation request from another server, the load information exchanging means determines whether or not to grant the license by referring to the load information of the own server. The load distribution system according to claim 7, wherein   When the load of the own server exceeds the load target value, the load accommodation unit transmits a redirect response to the client as a redirect destination in response to the processing request, and performs processing for the processing request to another The load distribution system according to any one of claims 1 to 8, which is executed by a server.   The load accommodation unit activates an application corresponding to the processing request by a dispatch unit when a load on the server is equal to or less than the load target value, and transmits a processing result of the application as a response to the client. Item 10. The load distribution system according to any one of Items 1 to 9.   The load factor determination means corresponds to each server with a row and a column based on the network configuration information, corresponds to a server corresponding to a server having a connection relationship between servers, and corresponds to a server having no connection relationship. A connection matrix A is created in which the element is 0, and the diagonal element is the number of servers connected to the server corresponding to the row. For the created connection matrix A, l is a variable and E is a unit matrix. Eigenvalue k (l) of G = (− lA + E) is obtained, and the maximum value of l of 0 <l <1 where all eigenvalues k (l) are greater than −1 and 1 or less is obtained. The load distribution system according to claim 1, wherein the value is determined as a load coefficient. In a network system having a plurality of servers, a method of distributing the server load,
A load factor determining step in which the server refers to network configuration information storing a network configuration of the network system and determines a load factor;
A load target value setting step in which the server exchanges load information with another server and sets a load target value based on the load of the own server, the load of the other server, and the load coefficient. When,
Processing server determination in which the server compares the load of its own server with the load target value, and determines whether to execute processing for a processing request from a client on its own server or to execute the other server And having steps
When the server decides that the server is to be executed by the server in the processing server determination step, the server performs the process on the other server when the server is determined to be executed in the server and executed by another server. A load balancing method characterized by the above.   In the processing server determination step, the server compares the load of the server with the load target value, and determines that the process is executed in the server if the load of the server is equal to or less than the load target value. The load distribution method according to claim 12, wherein when the load of the own server exceeds the target value, the other server is determined to execute.   In the load target value setting step, the server subtracts a value obtained by subtracting a value obtained by multiplying the load between the own server and the load of another server by the load coefficient from the load of the own server. The load distribution method according to claim 12 or 13, wherein the load distribution method is set as a value. In the load target value setting step, when the server determines that the ratio of the load of the own server and the load of another server exceeds a predetermined threshold, a status flag indicating whether or not to perform load accommodation is displayed. Set to "Load accommodation state", set the load target value,
The said server will refer to the said status flag, if the said status request is a "load accommodation state", and will perform the said processing server determination step, if the said processing request is received. The load balancing method described.   16. In the processing server determination step, the server sets the state flag to “normal state” indicating a state in which load accommodation is not performed when the server determines to execute processing for the processing request in its own server. The load balancing method described.   The said load target value setting step WHEREIN: The said server will store the information which points out said other server in a transfer destination server memory | storage part, if the said state flag sets to the "load accommodation state". Load balancing method.   When the server determines that another server is to execute the process for the processing request, the server stored in the transfer destination server storage unit is caused to execute the process with reference to the transfer destination server storage unit. The load distribution method according to claim 17.   In the load target value setting step, if the server determines that the ratio of the load of the local server to the load of the other server exceeds a predetermined threshold, whether or not load accommodation is possible for the other server. When receiving a response to grant permission for the request, the status flag is set to “load accommodation state”, and information indicating the other server is stored in the transfer destination server storage unit The load distribution method according to claim 17 or 18, wherein the load target value is set by storing the load target value.   When the server receives the load accommodation request from another server, the server determines whether or not to grant the permission by referring to the load information of the server, and returns a permission response when permitting or a rejection response when rejecting. The load balancing method according to claim 19.   When the server determines in the processing server determination step that the other server is to execute, the server transmits a redirect response to the client as a redirect destination in response to the processing request, and performs processing for the processing request. The load distribution method according to any one of claims 12 to 20, which is executed by another server.   The server activates an application corresponding to the processing request by a dispatch unit and transmits a processing result of the application as a response to the client when the server determines that the server executes in the processing server determination step. The load distribution method according to any one of 12 to 20.   In the load factor determination step, the server associates a row and a column with each server based on the network configuration information, and -1 indicates an element corresponding to a server that is in a connection relationship between servers, and is not in a connection relationship. A connection matrix A is created in which the element corresponding to the server is 0, the diagonal element is the number of servers connected to the server corresponding to the row, and l is a variable for the created connection matrix A, and E Is used as a unit matrix to determine eigenvalues k (l) of G = (− lA + E), and the maximum value of l among 0 <l <1 where all eigenvalues k (l) are greater than −1 and equal to or less than 1, The load distribution method according to any one of claims 12 to 22, wherein the maximum value of l is determined as a load coefficient. In a network system having a plurality of servers, there is a program for causing the server to execute a process for distributing the load on the server.
A load coefficient determination step for determining a load coefficient with reference to network configuration information storing a network configuration of the network system;
A load target value setting step for exchanging load information with another server, and setting a load target value based on the load of the own server, the load of the other server, and the load coefficient;
A processing server determination step of comparing the load of the local server with the load target value and determining whether the processing for the processing request from the client is executed by the local server or the other server is executed. Let
In the processing server determination step, when it is determined that the processing is to be executed by the own server, the processing is executed by the other server when the processing is executed in the own server and is determined to be executed by another server. Program to do.   In the processing server determination step, the load of the local server is compared with the load target value, and if the load of the local server is equal to or less than the load target value, it is determined that the processing is executed in the local server, The program according to claim 24, wherein when the load exceeds the target value, it is determined that the other server is executed.   In the load target value setting step, a value obtained by subtracting a value obtained by multiplying a difference between the load of the local server and the load of another server by the load coefficient from the load of the local server is set as the load target value. The program according to claim 24 or 25. In the load target value setting step, when it is determined that the ratio between the load of the local server and the load of the other server exceeds a predetermined threshold, a status flag indicating whether or not to perform load accommodation is set to “load accommodation state”. To set the load target value,
27. When receiving the processing request, the server refers to the status flag, and if the status flag is “load accommodation status”, causes the processing server determination step to be executed. The program according to any one of the above.   28. The program according to claim 27, wherein, in the processing server determination step, when it is determined that the processing corresponding to the processing request is to be executed by a local server, the state flag is set to a “normal state” indicating a state where load accommodation is not performed.   29. The load target value setting step according to claim 27 or 28, wherein the server stores information indicating the other server in a transfer destination server storage unit when the state flag is set to "load accommodation state". program.   When it is determined that the server is to execute processing for the processing request in another server, the processing is performed on the server stored in the transfer destination server storage unit with reference to the transfer destination server storage unit 30. The program according to claim 29, wherein the program is executed.   In the load target value setting step, when it is determined that the ratio of the load of the local server to the load of the other server exceeds a predetermined threshold, the load accommodation is made to inquire whether the load accommodation is possible to the other server. When the request is transmitted and a response indicating that the request is granted is received, the status flag is set to “load accommodation status”, and information indicating the other server is stored in the transfer destination server storage unit. The program according to claim 29 or 30, wherein the load target value is set.   When receiving the load accommodation request from another server, the server determines whether or not to grant the license by referring to the load information of the own server. 32. The program according to claim 31, wherein the program for executing a reply is executed.   When it is determined in the processing server determination step that the other server executes the server, the server transmits a redirect response to the client as a redirect destination in response to the processing request. The program according to any one of claims 24 to 32, which causes a process to be executed by another server.   Processing for starting the application corresponding to the processing request by dispatch means and transmitting the processing result of the application to the server as a response to the client when it is determined in the processing server determination step to be executed by the server. The program according to claim 24, wherein the program is executed.   In the load factor determination step, rows and columns are associated with each server based on the network configuration information, and an element corresponding to a server having a connection relationship between servers is -1, and a server having no connection relationship is associated with the server. A connection matrix A is created in which the element is 0, and the diagonal element is the number of servers connected to the server corresponding to the row. For the created connection matrix A, l is a variable and E is a unit matrix. Eigenvalue k (l) of G = (− lA + E) is obtained, and the maximum value of l of 0 <l <1 where all eigenvalues k (l) are greater than −1 and 1 or less is obtained. The program according to any one of claims 24 to 34, wherein the value is determined as a load coefficient.

Images