JP3910982B2 - Computer system, service load balancing method and program - Google Patents

Description

  The present invention relates to a computer system including a plurality of computers and capable of executing a plurality of types of services, and more particularly to a computer system and a program suitable for load distribution of services in an asymmetric resource environment.

  Conventionally, in order to efficiently process service execution requests from a large number of client terminals, a server load distribution system that distributes these execution requests to a plurality of server computers is known (for example, Non-Patent Document 1 or 2). reference). This server load distribution system is generally composed of a plurality of server computers having a symmetric (uniform) resource environment and a load distribution apparatus. When the load distribution apparatus receives a service execution request (request) from a client terminal via a network (external network), the load distribution apparatus selects a server computer to execute the service specified by the request. The selection of the server computer is performed in consideration of not applying a great load only to the specific server computer. That is, the load balancer distributes the execution of the same type of service to a plurality of computers.

  In the server load balancing system, in order to select a computer that executes a service, that is, in a service schedule, (1) round robin method, (2) weighted round robin method, (3) minimum connection method, or (4) fastest The law is generally applied. The round robin method is a method of selecting each server computer equally in a certain order. The weighted round robin method changes the frequency with which each computer is selected according to the capabilities of each server computer, by giving each computer a weight (ease of being selected) according to the capability (based on the round robin method). Is the method. The minimum connection method is a method of selecting a computer having the smallest number of connections (sessions) at that time. The fastest method is a method of selecting a computer that can respond the fastest at that time.

  When the load distribution device selects a server computer to execute a service by any one of the methods (1) to (4), the load distribution device sends an execution request from the client terminal to the selected server computer via a network (internal network). ) To send via. The server computer executes a service based on this execution request and sends a response to the load balancer. The load balancer returns the response sent from the server computer to the requesting client terminal.

  The load balancer monitors a response from the server computer. Then, the load balancer detects a failure of the server computer based on a timeout in which a response from the server computer does not return within a certain time. The failure of the server computer includes a failure of the server computer itself and a failure related to service execution by the server computer. When the load balancer detects a failure in the server computer, the load balancing device causes the system to perform a degenerate operation by not allocating the service to the failure computer.

  On the other hand, in recent years, a computer system called a cluster system has appeared (for example, see Non-Patent Document 3). The cluster system is composed of a plurality of computers having an asymmetric resource environment. In a cluster system, functionally different services (heterogeneous services) are assigned to a plurality of computers having an asymmetric resource environment in a form carefully planned by a user in advance. Each computer in the cluster system communicates with each other via a network to detect a failure of the computer executing the service. In this case, in the cluster system, rescheduling is performed to reassign (fail over) the service that was being executed on the computer in which the failure was detected. As a result, service (business) downtime can be shortened, and high availability (HA availability, business availability) called HA (High Availability) can be realized. So such a cluster system. This is called an HA cluster system.

Generally, rescheduling of services in a cluster system is performed for a standby computer. In such a cluster system, the load status of the computer is not taken into consideration in service scheduling. In addition, by setting the processing capability for each computer in the system by the user's operation, and setting the processing capability (ticket) necessary for execution of the service for each service that can be provided by the system, a specific computer can be set. 2. Description of the Related Art A static ticket type cluster system that prevents a service exceeding its processing capacity from being assigned is known.
Rajkumar Buyya, “High Performance Cluster Computing: Architecture and Systems (Volume 1)”, 1999, Prentice-Hall, Inc., p.340-363 Tony Bourke, “Server Load Balancing”, O'Relly & Associates, Inc., p.3-31, December 2001 Tetsuo Kaneko, Yoshiya Mori, "Cluster Software", Toshiba Review, Vol.54 No.12 (1999), p.18-21

  As described above, the conventional server load balancing system can dynamically load balance a plurality of server computers having a symmetric resource environment. However, the conventional server load distribution system can dynamically distribute loads to multiple server computers with complex asymmetric resource environments, that is, it can perform reliable execution control of services that operate in complex asymmetric resource environments. Absent. Further, in the conventional server load distribution system, since the failure of the computer is detected due to a response timeout from the computer, the computer failure cannot be detected quickly.

  On the other hand, load balancing in a conventional cluster system with an asymmetric resource environment is realized by careful planning of functional load balancing by the user, or a static ticket that assigns a fixed ticket for each service. It is realized by the method. For this reason, the conventional cluster system cannot dynamically distribute the load, and in the static ticket method, there is a possibility that services different from the actual load situation may be arranged.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a computer system, a service load distribution method, and a program capable of dynamically distributing service loads in an asymmetric resource environment.

  According to one aspect of the present invention, a computer system including a plurality of computers and capable of executing a plurality of types of services is provided. The computer system includes a service load measuring unit that measures a load of a service executed on each computer as a service load, a node load measuring unit that measures a load on each of the computers as a node load, the service load measuring unit, Based on the measurement result of the node load measuring means, the most suitable computer for service execution among the plurality of computers and the service to be relocated to the optimum computer are determined, and the service is relocated to the optimum computer. And an optimum service arrangement means.

  In such a configuration, based on the service load (service load) executed on each computer in the system and the load (node load) on each computer, the optimal computer for service execution and the optimal The service to be relocated to the computer is determined. As a result, even if the resource environment of each of the computers is asymmetric, service load distribution can be performed dynamically.

  Here, the service load measuring means calculates the dynamic service ticket value representing the dynamic load of the service as the service load based on the amount of resources consumed for executing the service on each computer. And each node load measuring means based on the dynamic service ticket value of the service being executed by each computer calculated by the service load measuring means. A total service ticket value calculating means for calculating a total service ticket value representing the node load of the computer, a total service ticket value of each of the computers calculated by the total service ticket value calculating means, and a preset value of each of the computers Based on the static node ticket value that represents the processing capacity, each of the above computers can use a new ticket. A dynamic node ticket value calculating means for calculating a dynamic node ticket value as a dynamic node ticket value, and a selection means for the service optimal arrangement means so that the dynamic node ticket value can be equal to or less than a predetermined value. It is preferable that a service to be relocated to the optimal computer is selected by the selection means from among the services being executed on a compatible computer. In this case, optimal load distribution is possible.

  Further, if the optimum computer is configured to be searched by the service optimum arrangement unit based on the dynamic node ticket value of each computer, more optimal load distribution is possible.

  Furthermore, the dynamic node ticket value of each computer, the static service ticket value that represents the resource amount that is predicted in advance to be executed by each computer, and the computer that executes the selected service. A computer whose dynamic node ticket value exceeds the larger value is searched as the optimum computer based on the larger value of the selected dynamic service ticket value of the selected service. If so, even more optimal load distribution becomes possible.

  In addition, parallel execution type service execution means for executing a predetermined parallel execution type service in parallel on at least two of the plurality of computers is added, and the optimum number of services is calculated in the optimum service arrangement means. Means for adjusting the number of services, and in the optimum service number calculation means, the optimum service number representing the optimum execution number of the parallel execution type service executed in parallel in the system It is calculated based on the dynamic service ticket value of the execution type service and the static service ticket value that represents the resource amount that is predicted in advance to be executed on each computer. In the service number adjusting means, the optimum service number calculated by the optimum service number calculating means and the current time Execution of a parallel execution type service that is actually executed by the parallel execution type service execution means in accordance with the magnitude relationship with the current service number that is the number of executions of the parallel execution type service that is actually executed in the system. It is preferable that the number is adjusted. In this way, even if the computer system is a cluster system and the resource environment of each computer is asymmetric, the optimal number of parallel execution services executed in parallel on the system is determined. Can be adjusted.

  According to the present invention, on the basis of the load of services executed on each computer in the system and the load on each computer, the optimal computer for service execution and the service to be relocated to the optimal computer are determined. Therefore, even if the resource environment of each computer is asymmetric, service load distribution can be performed dynamically.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a high availability cluster system (HA cluster system) according to an embodiment of the present invention. The cluster system shown in FIG. 1 includes four computers (server computers) 10-1 to 10-4. The computers 10-1 to 10-4 are interconnected by a network (internal network) 20 used for communication between the computers 10-1 to 10-4. In FIG. 1, a network (external network) used for communication between the computers 10-1 to 10-4 and the client terminal is omitted. A service execution request from the client terminal is transmitted to the cluster system of FIG. 1 via this external network. The computers 10-i (i = 1 to 4) in the cluster system execute the service specified by the request from the client terminal, and return a response indicating the execution result to the requesting client terminal via the external network. . The communication between the computers 10-1 to 10-4 and the communication between the computers 10-1 to 10-4 and the client terminal may be performed via the same network. However, it increases communication traffic in the network.

  In the computers 10-1 to 10-4, operating systems (hereinafter referred to as OSs) 11-1 to 11-4 operate. In the cluster system constituted by the computers 10-1 to 10-4, the cluster control mechanism 12 operates. The cluster control mechanism 12 is a virtual machine realized by a cluster control unit (not shown) provided in each of the computers 10-1 to 10-4 operating in synchronism with each other while communicating with each other. It is. Therefore, it can be considered that the cluster control mechanism 12 exists across the computers 10-1 to 10-4. The cluster control unit is realized by the computer 10-i reading and executing a cluster control program (cluster software) including service load balancing. The cluster control program is a storage medium (a magnetic disk represented by a floppy (registered trademark) disk, an optical disk represented by a DVD, a semiconductor represented by a flash memory) that can be read by the computer 10-i. It can be stored in advance in a memory or the like and distributed. Further, this program may be downloaded (distributed) via a network. The cluster control mechanism 12 can quickly detect a failure of a computer when the cluster control units on the computers 10-1 to 10-4 operate synchronously while communicating with each other.

  The cluster control mechanism 12 includes a service optimal arrangement mechanism 121 and a service control mechanism 122. The service optimum arrangement mechanism 121 is realized by a service optimum arrangement unit (not shown) provided in each of the computers 10-1 to 10-4 operating synchronously and integrally while communicating with each other. . The service optimal arrangement mechanism 121 has a function of determining an optimal computer for executing the service when a failure of the computer executing the service occurs or when the load status of the service changes. The service optimal arrangement mechanism 121 also has a function of rearranging services to the determined optimal computer. The service optimum arrangement mechanism 121 further has a function of adjusting the number of executions of the same type of service (parallel execution type service PSVC) executed in parallel by the parallel execution type service execution unit 13 described later to an optimal number.

  The service control mechanism 122 is realized by a service control unit (not shown) provided in each of the computers 10-1 to 10-4 operating synchronously and integrally while communicating with each other. The service control mechanism 122 has a function of switching over the service to the computer determined by the service optimum arrangement mechanism 121 under the control of the service optimum arrangement mechanism 121.

  In the cluster system of FIG. 1, a parallel execution type service execution unit 13 managed by the cluster control mechanism 12 operates. The parallel execution type service execution unit 13 is realized by the computers 10-1 to 10-4 similarly to the cluster control mechanism 12, and exists across the computers 10-1 to 10-4. The parallel execution type service execution unit 13 has a function of executing the service PSVC in parallel on a plurality of computers (nodes) among the computers 10-1 to 10-4. The type of service PSVC that can be executed in parallel by the parallel execution type service execution unit 13 is referred to as a parallel execution type service. Here, the number of executions of the parallel execution type service PSVC executed in parallel by the parallel execution type service execution unit 13, that is, the number of services (= number of nodes) is determined in the cluster control mechanism 12 based on the service ticket value described later. This is determined by the service optimum arrangement mechanism 121. FIG. 1 shows a case where the number of executions (number of services) of the parallel execution service PSVC is two. Here, regarding the execution of the service PSVC, the computers 10-3 and 10-4 are in the operating state, and the computers 10-1 and 10-2 are in the standby state. That is, in FIG. 1, the parallel execution type service PSVC is executed in parallel on the two computers 10-3 and 10-4 by the parallel execution type service execution unit 13.

A parameter value called a static service ticket value SST _PSVC is set in advance in the parallel execution service PSVC (an application that realizes the service). This static service ticket value SST _PSVC represents a resource amount that is predicted in advance to be necessary for executing the parallel execution type service PSVC on the computer 10-i (that is, a static load of the service PSVC). . On the other hand, the parallel execution type service execution unit 13, the number of execution of the parallel execution type service PSVC should by minimally parallel execution in the system (hereinafter, referred to as the minimum service number N _min) in advance by the input operation of the user settings Is done. The minimum service number N _min also represents the number of computers (the number of nodes) that execute the parallel execution type service PSVC in parallel at the minimum in the system.

The parallel execution type service execution unit 13 includes service load monitors 131-1 to 131-4 operable on the computers 10-1 to 10-4. The service load monitor 131-i (i = 1 to 4) operates only when the computer 10-i is in the execution state of the parallel execution type service PSVC, and the resource in the execution state of the service PSVC in the computer 10-i. Measure consumption. The service load monitor 131-i predicts the resource amount necessary for executing the parallel execution type service PSVC on the computer 10-i based on the measured current resource consumption, and is necessary for executing the service PSVC. A dynamic service ticket value DST _PSVCi representing a large resource amount (that is, a dynamic load of the service PSVC) is _obtained . The service load monitor 131-i notifies the cluster control mechanism 12 of the dynamic service ticket value DST _PSVCi .

  In the computers 10-1 to 10-4, the HA type service execution units 141-1 to 141-4 that execute the HA type service SVC1 and the HA type service execution units 142-1 to 142-4 that execute the HA type service SVC2 are used. Are operable. The HA type service execution units 141-1 to 141-4 and 142-1 to 142-4 are managed by the cluster control mechanism 12.

  The HA type service is a service (application) that is failed over by the control of the cluster control mechanism 12, and can be executed only by any one of the computers 10-1 to 10-4 in the same time zone. Service. In FIG. 1, regarding execution of the HA type service SVC1, only the computer 10-1 (inside the HA type service execution unit 141-1) is in operation, and the other computers 10-2 to 10-4 (inside HA type) The service execution units 141-2 to 141-4) are in a standby state. In FIG. 1, regarding execution of the HA type service SVC2, only the computer 10-2 (inside the HA type service execution unit 142-2) is in operation, and the other computers 10-1, 10-3, 10- 4 (of which the HA type service execution units 142-1, 142-3, 142-4) are in a standby state.

Static service ticket values SST _SVC1 and SST _SVC2 are set in advance in the HA type services SVC1 and SVC2 (each application that realizes them). The static service ticket values SST _SVC1 and SST _SVC2 are parameter values representing the amount of resources necessary for the HA type service execution unit 141-i in the computer 10-i to execute the HA type services SVC1 and SVC2.

The HA type service execution units 141-1 to 141-4 and 142-1 to 142-4 include service load monitors 151-1 to 151-4 and 152-1 to 152-4, respectively. The service load monitors 151-i and 152-i (i = 1 to 4) indicate that the HA type service execution units 141-i and 142-i in the computer 10-i are in the execution state of the HA type services SVC1 and SVC2. Only the resource consumption in the execution state of the services SVC1 and SVC2 in the computer 10-i is measured. The service load monitors 151-i and 152-i predict the resource amount necessary for the execution of the services SVC1 and SVC2 in the computer 10-i based on the measured current resource consumption, and the service SVC1, The dynamic service ticket values DST _SVC1i and DST _SVC2i representing the amount of resources necessary for executing the SVC 2 (that is, the dynamic load of the services _{SVC 1} and _{SVC 2)} are obtained. The service load monitors 151-i and 152-i notify the cluster control mechanism 12 of the dynamic service ticket values DST _SVC1i and DST _SVC2i .

In the computers 10-1 to 10-4, node load monitors 16-1 to 16-4 operate. The computer 10-1 to 10-4, a static node ticket value SNT ₁ ~SNT ₄ representing a processing capability of the computer (node) (resource amount) is set in advance. In the present embodiment, it is assumed that the computer 10-1 to 10-4 has an asymmetric resource environment, thus static nodes ticket value SNT ₁ ~SNT ₄ are different. Node load monitor 16-1 to 16-4, the sum of the ticket value of all services that are running on the computer 10-1 to 10-4 (hereinafter, referred to as total service ticket value) TST ₁ ~TST ₄ The dynamic node ticket values DNT _{1 to} DNT ₄ are calculated from the static node ticket values SNT _{1 to} SNT ₄ every time a predetermined inspection time comes. The dynamic node ticket values DNT _{1 to} DNT ₄ are ticket values representing resource amounts that can be newly used in the computers 10-1 to 10-4. Node load monitor 16-1 to 16-4 notifies the dynamic node ticket value DNT ₁ ~DNT ₄ to the cluster control mechanism 12.

Next, the operation of the cluster system of FIG. 1 will be described.
The service load monitor 131-i (i = 1 to 4) receives a predetermined inspection time when the parallel execution service PSVC is executed on the computer 10-i by the parallel execution service execution unit 13. For example, it operates periodically, for example. The service load monitor 131-i measures the resource consumption in the execution state of the service PSVC in the computer 10-i. The service load monitor 131-i is a dynamic service ticket value DST _PSVCi that represents a predicted value of the resource amount necessary for executing the parallel execution service PSVC in the computer 10-i based on the measured current resource consumption. Calculate Three prediction functions f (x), g (y), and h (z) are used for the calculation of the dynamic service ticket value DST _PSVCi . The prediction functions f (x), g (y), and h (z) are consumption amounts of three types of resources in the computer 10-i, for example, CPU usage x, memory usage y, response time z (from the client terminal). The time from receiving the execution request of the service s until the response of the execution result is returned). Here, the dynamic service ticket value DST _PSVCi is _expressed by the following equation: DST _si = f (x) + g (y) + h (z) (1)
Is calculated according to However, s = PSVC. The dynamic service ticket value DST _PSVCi calculated by the service load monitor 131-i is notified to the cluster control mechanism 12.

Similarly, the service load monitors 151-i and 152-i are preliminarily stored when the computer 10-i (within the HA-type service execution units 141-i and 142-i) is in the execution state of the HA-type services SVC1 and SVC2. Each time a predetermined inspection time arrives, for example, it operates periodically. The service load monitors 151-i and 152-i use the dynamic service ticket values DST _SVC1i and DST _SVC2i representing the predicted values of the resource amounts necessary for executing the services SVC1 and SVC2 in the computer 10-i as current resources. It is calculated according to the above formula (1) based on the consumption amount. Here, s = SVC1 and s = SVC2. The dynamic service ticket values DST _SVC1i and DST _SVC2i calculated by the service load monitors 151-i and 152-i are notified to the cluster control mechanism 12.

Next, the calculation operation of the dynamic node ticket value DNT _i by the node load monitor 16-i (i = 1 to 4) will be described with reference to the flowchart of FIG.
The node load monitor 16-i obtains the service ticket value ST _si for all the services s being executed on the computer 10-i by the following formula: service ticket value ST _si
= MAX (static service ticket value SST _s , dynamic service ticket value DST _si )
(2)
(Steps S1, S2). The service ticket value ST _si is the maximum value of the amount of resources consumed or predicted to be consumed by the service s being executed on the computer 10-i, that is, the possibility of being consumed by the service s. Represents the maximum resource amount with

Node load monitor 16-i, when obtaining the service ticket value ST _si for all services s running on the computer 10-i, the sum of the service ticket value ST _si, that is, the total service ticket value TST _i, following Formula Total service ticket value TST _i
= Σ (service ticket value ST _si of the service being executed on the computer 10-i) (3)
(Step S3). The total service ticket value TST _i represents the maximum resource amount that can be consumed by the entire service being executed on the computer 10-i, that is, the load (node load) of the entire computer 10-i.

When the node load monitor 16-i obtains the total service ticket value TST _i of all services s being executed on the computer 10-i, the ticket representing the amount of resources that can be newly used by the computer 10-i at that time. Value, that is, the dynamic node ticket value DNT _i is expressed by the following formula: dynamic node ticket value DNT _i
= (Static node ticket value SNT _i ) − (Total service ticket value TST _i ) (4)
(Step S4).
Thus, the dynamic node ticket value DNT _i is calculated by subtracting the total service ticket value TST _i from the static node ticket value SNT _i . The node load monitor 16-i repeats the above processing (steps S1 to S4) periodically (that is, at regular time intervals).

  Next, the operation for adjusting the number of executions (number of services) of the parallel execution type service PSVC executed in parallel by the parallel execution type service execution unit 13 to the optimum number by the service optimal arrangement mechanism 121 in the cluster control mechanism 12 Will be described with reference to the flowchart of FIG.

The service optimum arrangement mechanism 121 includes the dynamic service ticket value DST _PSVCi of the parallel execution type service PSVC, the static service ticket value SST _PSVC of the service PSVC, and the minimum of the service PSVC in each computer (node) 10-i in the system. Based on the service number N _min , the execution number (hereinafter referred to as the optimum service number) OSN of the parallel execution type service PSVC necessary for the current system is calculated as follows (step S11). First International optimal deployment mechanism 121, the sum of the dynamic service ticket value DST _PSVC1 ~DST _PSVC4 at each computer 10-1 to 10-4, that is, the total dynamic service ticket value TDST, the following equation total dynamic service ticket value TDST
= Σ (Dynamic service ticket value DST _PSVCi of computer 10-i) (5)
(Step S11a).

Next, the service optimum arrangement mechanism 121 determines the number of executions of the currently required parallel execution service PSVC based on the total dynamic service ticket value TDST and the static service ticket value SST _PSVC of the parallel execution service PSVC. The following formula is used as the number of services TSN.
= (Total dynamic service ticket value TDST / Static service ticket value SST _PSVC )
Integer part (when remainder is 0)
Or provisional service number TSN
= (Total dynamic service ticket value TDST / Static service ticket value SST _PSVC )
Integer part of +1 (when remainder is not 0) (6)
(Step S11b).

Next, the service optimum arrangement mechanism 121 calculates the optimum service number OSN based on the provisional service number TSN and the preset minimum service number N _min as follows:
= MAX (temporary service number TSN, minimum service number N _min ) (7)
(Step S11c).

Next, the service optimal arrangement mechanism 121 compares the optimal service number OSN with the execution number (hereinafter referred to as the current service number) CSN of the parallel execution type service PSVC currently being executed by the parallel execution type service execution unit 13. If the optimum service number OSN is larger than the current service number CSN (step S12), the computer 10-j (j is a computer that can newly execute the parallel execution type service PSVC among the computers 10-1 to 10-4 in the system). It is determined whether there is any one of 1-4) (step S13). If there is a computer 10-j capable of executing the service PSVC, the service optimal arrangement mechanism 121 determines whether the static node ticket value SNT _j and the dynamic node ticket value DNT _j out of the computer 10-j. The computer having the largest difference is selected, and the service PSVC is executed by the selected computer (step S14). Thereafter, the service optimum arrangement mechanism 121 returns to step S11. That is, the service optimal arrangement mechanism 121 has a large difference between the static node ticket value SNT _j and the dynamic node ticket value DNT _j among the computers 10-j that can execute the service PSVC. The operation of sequentially selecting and starting the service PSVC by the selected computer is repeated until the optimum service number OSN reaches the current service number CSN. On the other hand, if there is no computer 10-j that can execute the service PSVC (step S13), the service optimum arrangement mechanism 121 waits for a certain period of time (sleeps) (step S15), and then returns to step S11.

Further, if the optimum service number OSN is smaller than the current service number CSN (step S16), the service optimum arrangement mechanism 121 can stop the parallel execution type service PSVC in the computers 10-1 to 10-4 in the system. It is determined whether or not there is a computer 10-j (j is any one of 1 to 4) (step S17). If there is a computer 10-j capable of stopping the service PSVC, the service optimum arrangement mechanism 121 determines whether the static node ticket value SNT _j and the dynamic node ticket value DNT _j out of the computer 10-j. The computer with the smallest difference is selected, and the execution of the service PSVC in the selected computer is stopped (step S18). Thereafter, the service optimum arrangement mechanism 121 returns to step S11. That is, the service optimal arrangement mechanism 121 has a small difference between the static node ticket value SNT _j and the dynamic node ticket value DNT _j among the computers 10-j that can stop the service PSVC. The operation of sequentially selecting and stopping the execution of the service PSVC in the selected computer is repeated until the optimum service number OSN reaches the current service number CSN. On the other hand, if there is no computer 10-j that can stop the service PSVC (step S17), the service optimum arrangement mechanism 121 waits for a certain time (step S15), and then returns to step S11. If the optimum service number OSN matches the current service number CSN, the service optimum arrangement mechanism 121 waits for a predetermined time (step S15) and then returns to step S11.

As described above, in this embodiment, the optimum service number OSN representing the optimum number of executions of the parallel execution service PSV executed in parallel in the cluster system (computer system) of FIG. It is calculated based on the dynamic service ticket value DST _PSVCi of the parallel execution type service PSVC in 10-i, the static service ticket value SST _PSVC of the service PSVC, and the minimum service number N _min of the service PSVC. Then, the execution of the parallel execution service is performed according to the magnitude relationship between the calculated optimum service number OSN and the current service number CSN (the number of executions of the parallel execution service PSV actually executed in the system at that time). The number is adjusted. Thus, as in the present embodiment, even if the computer system of FIG. 1 is a cluster system and the resource environments of the computers 10-1 to 10-4 in the system are asymmetric, It is possible to adjust the number of parallel-execution services that are executed in an optimal number.

  In the system of FIG. 1, it is assumed that the type of executable parallel execution service is one type of PSVC. However, a plurality of types of parallel execution services that can be executed are also possible. In this case, the optimum service number OSN may be obtained for each type of parallel execution service.

Next, the optimum arrangement of the HA type service or the parallel execution type service by the service optimum arrangement mechanism 121 will be described with reference to the flowchart of FIG.
The service optimal arrangement mechanism 121 is a computer 10-j in which the value of (dynamic node ticket value DNT _j -Δ) is equal to or less than a predetermined value among the computers 10-1 to 10-4, that is, the dynamic node ticket value DNT. _The computer 10-j in which j may be a certain value or less is searched (step S21). Here, delta is also not less than a predetermined value is the dynamic node ticket value DNT _j, a margin for the DNT _j retrieves the computer 10-j that can be a fixed value or less. In the present embodiment, the constant value is zero. In addition, a configuration may be adopted in which the computer 10-j in which the dynamic node ticket value DNT _j may be less than a certain value is searched.

If there is no computer 10-j that may cause the dynamic node ticket value DNT _{j to} become a certain value or less (step S21), the service optimum arrangement mechanism 121 waits for a certain time (sleeps) (step S21). S22), the process returns to step S21. When an event such as a computer failure occurs, the service optimal arrangement mechanism 121 returns to step S21 without waiting for a certain time.

On the other hand, if there is a computer 10-j in which the dynamic node ticket value DNT _j may be equal to or less than a certain value (step S21), the service optimum arrangement mechanism 121 may select from among the computers 10-j. The computer executing the service s with the lowest priority and the service s are selected (step S23). Next, the service optimal arrangement mechanism 121 determines whether or not the selected service s can be switched over to another computer in the system (step S24). In the present embodiment, services that can be switched over are determined in advance. That is, in this embodiment, whether or not switchover is possible is set in advance for each service. In this case, the determination in step S24 is realized by determining whether the selected service s is predetermined as a service that can be switched over. Note that it is possible to determine whether the switchover is possible depending on the execution status of the service s, for example, whether the service s is in the critical area. The critical area process is, for example, a process that requires response performance or a process that requires consistency (primitiveness), and is a process that requires a cost for backtracking. Specific examples include transaction processing and database update processing.

First, it is assumed that the selected service s can be switched over to another computer. Further, it is assumed that the computer capable of executing the service s is the computer 10-k (k is any one of 1 to 4). In this case, the service optimum arrangement mechanism 121 searches the computer 11-k capable of executing the selected service s for the optimum computer for switching over the service s as follows (step S25). . That is, the service optimal arrangement mechanism 121 has a larger one of the dynamic node ticket value DNT _k , the static service ticket value SST _s, and the dynamic service ticket value DST _sk of each computer 10-k that can execute the service s. value, i.e. MAX (SST _s, DST _sk) on the basis of the, DNT _k searches for computer 10-k of greater than _{MAX (SST s, DST sk)} . If a plurality of computers 10-k are searched, the service optimum arrangement machine 121 selects one of the plurality of computers 10-k as an optimum computer for switching over the service s. Here, the computer 10-k having the largest dynamic node ticket value DNT _k may be selected as the optimum computer. _{Furthermore, MAX (SST s, DST sk} ) of the DNT _{k exceeding, MAX (SST s, DST sk} ) computer 10-k having the closest DNT _k in may be configured to be selected.

When the optimal computer for switching over the selected service s can be searched (step S26), the service optimal arrangement mechanism 121 starts execution of the service s with the optimal computer (step S27). Later, the process returns to step S21. On the other hand, when the optimal computer cannot be searched (step S26), the service optimal arrangement mechanism 121 determines that the dynamic node ticket value DNT _j may be a certain value or less from among the computers 10-j. Next, the computer that is executing the service s with the lower priority and the service s are selected (step S28). Thereafter, the service optimum arrangement mechanism 121 returns to step S24.

  On the other hand, if the selected service s cannot be switched over to another computer (step S24), the service optimum arrangement mechanism 121 determines whether or not the selected service s can be stopped (step S29). ). In the present embodiment, services that can be stopped are determined in advance. That is, in this embodiment, whether or not the service can be stopped is set in advance for each service. Note that it is possible to determine whether the service can be stopped depending on the execution status of the service s.

If the selected service s can be stopped (step S29), the service optimal arrangement mechanism 121 stops the selected service s (step S30), and then returns to step S21. On the other hand, if the selected service s cannot be stopped (step S29), the service optimal arrangement mechanism 121 determines that the dynamic node ticket value DNT _j is in the computer 10-j that may be a certain value or less. Then, the computer executing the service s having the next lowest priority and the service s are selected (step S31). Thereafter, the service optimum arrangement mechanism 121 returns to step S24.

As described above, in the present embodiment, the service s being executed by the computer 10-j in which the dynamic node ticket value DNT _j may be equal to or less than a certain value is represented by the dynamic node ticket value DNT _k being a static service. It can be executed by the computer 10-k that exceeds the larger one of the ticket value SST _s and the dynamic service ticket value DST _sk . Thereby, optimal load distribution is realized. That is, in the present embodiment, when a computer failure or a significant service load or node load change occurs, the service relocation is automatically performed by the service optimum arrangement machine 121.

  Note that the flowchart of FIG. 4 does not show a case where there is no optimal computer for switching over the selected service even if the loop of steps S24, S25, S26 and S28 is repeated. Similarly, FIG. 4 does not show a case where there is no service that can be stopped even if the loop of steps S24, S29, and S31 is repeated. In such a case, for example, other services may be switched over / stopped according to a predetermined user setting. When there is no optimal computer, the selected service may be stopped or nothing may be performed until the optimal computer appears.

  The above embodiment assumes a cluster system that can execute not only HA type services but also parallel execution type services. However, the present invention is not limited to a cluster system, and can be applied to a computer system (load distribution system) capable of executing a parallel execution type service although it cannot execute an HA type service.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment.

1 is a block diagram showing a configuration of a cluster system according to an embodiment of the present invention. 9 is a flowchart showing a procedure for calculating a dynamic node ticket value DNT _i by the node load monitor 16-i (i = 1 to 4) in the embodiment. 6 is an exemplary flowchart showing an operation procedure for adjusting the number of executions (number of services) of a parallel execution service PSVC applied in the embodiment; 6 is a flowchart showing an operation procedure for optimal arrangement of services (HA type service or parallel execution type service) applied in the embodiment.

Explanation of symbols

  10-1 to 10-4 ... computers, 11-1 to 11-4 ... OS (operating system), 12 ... cluster control mechanism, 13 ... parallel execution type service execution unit, 16-1 to 16-4 ... node load monitor (Node load measuring means), 20 ... network, 121 ... service optimum arrangement mechanism, 122 ... service control mechanism, 131-1 to 131-4, 151-1 to 151-4, 152-1 to 152-4 ... service load Monitor (service load measuring means).

Claims (5)

In a computer system equipped with multiple computers and capable of executing multiple types of services,
Parallel execution service execution means for executing a predetermined parallel execution service in parallel on at least two of the plurality of computers;
Service load measuring means for measuring a service load executed on each of the plurality of computers as a service load, based on the amount of resources consumed for executing the service on each of the plurality of computers, Service load measuring means including dynamic service ticket value calculating means for calculating a dynamic service ticket value representing a dynamic load of a service as the service load;
Node load measuring means for measuring the load of each of the plurality of computers as a node load, the dynamic service of the service being executed by each of the plurality of computers calculated by the dynamic service ticket value calculating means Based on the ticket value, total service ticket value calculating means for calculating the total service ticket value representing the node load of each of the plurality of computers, and the plurality of computers calculated by the total service ticket value calculating means Based on each total service ticket value and a static node ticket value representing the processing capability of each of the plurality of computers set in advance, a ticket value newly usable in each of the plurality of computers is moved. A node load measuring means including a dynamic node ticket value calculating means for calculating a dynamic node ticket value;
Based on the measurement results of the service load measuring means and the node load measuring means, the optimum computer for service execution and the service to be relocated to the optimum computer are determined from the plurality of computers, and the optimum computer is determined. Service relocation means for relocating the service to a service that is relocated to the optimal computer from among the services being executed on a computer whose dynamic node ticket value may be a predetermined value or less A selection means for selecting the system , a search means for searching for the optimum computer to be relocated of the service selected by the selection means based on a dynamic node ticket value of each of the plurality of computers, and the system The optimal number of services representing the optimal number of executions of the parallel execution type service executed in parallel in each of the plurality of computers. Column execution type total dynamic service ticket value is the sum of the dynamic service ticket value of the service, and the amount of resources is pre predicted to be required to perform the parallel execution type service in each of the plurality of computers Based on a static service ticket value that represents the optimal dynamic service ticket value calculated by dividing the total dynamic service ticket value by the static service ticket value, and an optimal service calculated by the optimal service number calculation unit The parallel execution type service execution unit actually executes the service according to the magnitude relation between the number and the current service number that is the number of executions of the parallel execution type service actually executed in the system at that time. Service optimal arrangement means including service number adjustment means for adjusting the number of executions of parallel execution type services ;
A computer system comprising: The search means is a resource that is predicted in advance to be necessary for each of the plurality of computers to execute the dynamic node ticket value of each of the plurality of computers and the service selected by the selection means. Based on the static service ticket value representing the quantity and the larger value of the dynamic service ticket value of the selected service being executed on the computer, the larger of the dynamic node ticket values 2. The computer system according to claim 1 , wherein a computer exceeding the value is searched as the optimum computer. A load balancing method for dynamically performing service load balancing in a computer system having a plurality of computers and capable of executing a plurality of types of services,
Executing a predetermined parallel execution type service in parallel on at least two of the plurality of computers;
Calculating a dynamic service ticket value representing a dynamic load of the service based on an amount of resources consumed for execution of the service in each of the plurality of computers;
Calculating a total service ticket value representing a load of each of the plurality of computers based on the dynamic service ticket value of a service being executed on each of the plurality of computers;
Newly used in each of the plurality of computers based on the total service ticket value of each of the plurality of computers and a static node ticket value representing the processing capability of each of the plurality of computers set in advance. Calculating a dynamic node ticket value, which is a possible ticket value;
Selecting a service to be relocated to a computer that is most suitable for service execution among the plurality of computers from among the services being executed by a computer that may have a dynamic node ticket value that is less than a certain value; ,
Searching for the optimal computer to which the selected service is to be relocated based on the dynamic node ticket value of each of the plurality of computers;
Relocating the selected service to the retrieved optimal computer;
A total dynamic service ticket value that is the sum of dynamic service ticket values of the parallel execution service in each of the plurality of computers, and an optimal number of services representing the optimal execution number of the parallel execution service, and the parallel execution Based on a static service ticket value that represents a resource amount that is predicted in advance to be executed on each of the plurality of computers , the total dynamic service ticket value is represented by the static service ticket value. A step of calculating by dividing ,
The parallel execution actually executed according to the magnitude relationship between the calculated optimum service number and the current service number that is the number of executions of the parallel execution type service actually executed in the system at that time Adjusting the number of executions of type services
A service load distribution method comprising : In the searching step, the dynamic node ticket value of each of the plurality of computers and a resource amount that is predicted in advance to be necessary to execute the selected service on each of the plurality of computers are calculated. Based on the static service ticket value that represents and the larger value of the dynamic service ticket value of the selected service being executed by the computer, the larger value of the dynamic node ticket value 4. The service load distribution method according to claim 3, wherein more than one computer is searched as the optimum computer. A parallel execution type service execution means for executing a predetermined parallel execution type service in parallel on at least two of the plurality of computers, and a service load executed on each of the plurality of computers Service load measuring means for measuring as a load, wherein a dynamic service ticket value representing a dynamic load of the service is obtained based on an amount of resources consumed for executing the service in each of the plurality of computers. Service load measuring means including dynamic service ticket value calculating means for calculating as a load, and node load measuring means for measuring each load of the plurality of computers as a node load, wherein the dynamic service ticket value calculating means Based on the calculated dynamic service ticket value of the service being executed on each of the plurality of computers, A total service ticket value calculating means for calculating a total service ticket value representing the node load of each of the plurality of computers; a total service ticket value of each of the plurality of computers calculated by the total service ticket value calculating means; And based on the static node ticket value representing the processing capability of each of the plurality of computers set in advance, a ticket value that can be newly used in each of the plurality of computers is calculated as a dynamic node ticket value. A plurality of computers including a node load measuring unit including a dynamic node ticket value calculating unit , applied to a computer system capable of executing a plurality of types of services, and being executed by each of the plurality of computers A program that enables dynamic load balancing of
In the calculator,
Obtaining the dynamic service ticket value of the service being executed in each of the plurality of computers from the service load measuring means;
Obtaining the dynamic node ticket value of each of the plurality of computers from the service load measuring means;
Select a service to be relocated to a computer that is most suitable for service execution among the plurality of computers from among the services that are being executed by the computers that may have the acquired dynamic node ticket value below a certain value. And steps to
Searching for the optimal computer on which the selected service is to be relocated based on the obtained dynamic node ticket value of each of the plurality of computers;
Relocating the selected service to the retrieved optimal computer ;
A total dynamic service ticket value that is a sum of the acquired dynamic service ticket values of the parallel execution service in each of the plurality of computers, and an optimal service number representing an optimal execution number of the parallel execution service; And based on a static service ticket value representing a resource amount that is predicted in advance to be executed by each of the plurality of computers, the total dynamic service ticket value is set to the static service ticket value. Calculating by dividing by the service ticket value;
The parallel execution actually executed according to the magnitude relation between the calculated optimum service number and the current service number that is the number of executions of the parallel execution type service actually executed in the system at that time For executing the step of adjusting the number of executions of the type service .

Images