JP2007179246A - Method, program and server for computer management - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately measure performance of servers used in a load balancing system without stopping the working load balancing system. <P>SOLUTION: A computer performs load balancing according to a priority in which the priority of a performance measurement target server is set high and the priorities of other servers are set low, and the maximum limit of the amount of processing that is distributed to the performance measurement target server is raised gradually by checking CPU loads of the performance measurement target server. The amount of processing distributed to the performance measurement target server becomes throughput for the server when CPU loads reach a prespecified permitted maximum limit, so that the value is recorded. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a technique for a computer management method, a computer management program, and a computer management server.

  Load distribution is a technique in which a plurality of servers process requests in parallel by distributing requested requests to a plurality of servers. As a result, the total number of requests executed by multiple servers is expected to increase per unit time, rather than one server executing requests, so the number of requests that can be processed per unit time is increased. be able to.

  As a selection algorithm of a server to which a requested request is allocated, a load situation as described in Patent Document 1 is started, starting with a static load distribution algorithm (see Non-Patent Document 1) such as round robin, weighting, and priority. A method of distributing processing by seeing has been developed. These load distribution methods are intended to distribute the processing so that the loads on all servers are equal or the total number of processing cases is maximized.

  Furthermore, in order to perform load distribution efficiently, a technique for dynamically increasing / decreasing the load distribution target server according to the request amount at each time point (hereinafter referred to as dynamic load distribution) has been proposed. ing. Specifically, the load status of the server is monitored, and if the load increases and exceeds a predetermined upper limit threshold, a server that is not used at that time is added to deal with the load increase. Control is performed. Conversely, when the load decreases and falls below the lower threshold, control is performed so that at least one server being used is reduced and the reduced server can be used for another purpose. Thereby, it can suppress operating a server excessively and can improve the utilization efficiency of a server.

  For example, in the dynamic load distribution described in Patent Document 2, immediately after adding a server, the request transmission amount to the added server is reduced, and the request transmission amount to the additional server is gradually increased with time. It is characterized by increasing it. The purpose of this is to prevent the response time from deteriorating by reducing the load on the server immediately after the addition because the server immediately after the addition has a low processing capacity because the necessary information is not cached.

In order to efficiently realize dynamic load balancing, it is necessary to measure the performance of each server to which a request is assigned by load balancing. In particular, in the grid computing system, since effective utilization of computer resources such as servers is one of the purposes, the specifications of the servers to be used are not uniform and often vary. The performance measurement method described in Patent Document 3 adjusts the distribution ratio of the load balancer by operating the load value of the server included in the feedback information transmitted from the server to the load balancer, and performs a test. Although the performance is measured by intentionally concentrating the load on the target server, the load distribution device to be used is premised on performing load distribution control by active feedback.
JP 10-31365 A JP 2005-92862 A JP 2005-100500 A IAI Technology, Inc. Nobuhiro Kawamoto, “Load Balancing Technology-Building a System That Can Withstand High Loads”, http://www.nic.ad.jp/en/materials/iw/2002/proceeding/T19.pdf

  Conventionally, server performance measurement for efficiently realizing dynamic load balancing has been performed without including a server in an active load balancing system. Therefore, the server performance could not be measured with high accuracy. Not only dynamic load balancing systems, but many systems have already been operated and operated. These operating systems are required to operate without interruption (24-hour operation).

  For example, a method of measuring the performance of hardware such as a CPU (Central Processing Unit) frequency in a stand-alone manner without connecting to an operating load distribution system with benchmark software is conceivable. However, if you do not measure the application-based performance of how much processing the relevant server can handle, it cannot be said that the server performance could be measured with high accuracy, and dynamic load balancing is efficient. Can not be realized.

  Therefore, the main object of the present invention is to measure the performance of a server used in the load distribution system with high accuracy without stopping the load distribution system in operation. The measured server performance is used to select a server that is increased or decreased in accordance with an increase or decrease in load distribution requests.

  In order to solve the above problems, the present invention is a computer management method for measuring the processing capacity of a first server, which is a performance measurement target, in a plurality of servers that execute requests distributed based on a load distribution management table. A priority control procedure in which the computer changes the load distribution management table stored in the storage means so that the request is preferentially assigned to the first server over the second server that is executing the request. And a load status monitoring procedure for recording the processing amount when the performance of the first server reaches a predetermined allowable upper limit as the processing capacity of the first server. Other means will be described later.

  According to the present invention, it is possible to measure the performance of a server used in the load distribution system with high accuracy without stopping the load distribution system in operation. For example, it is possible to easily measure the processing capacity of each server required in a system that performs advanced load balancing such as grid computing.

  The best mode for carrying out the present invention will be described below in detail with reference to the drawings. FIG. 1 is a configuration diagram illustrating a load distribution system.

  A request from the user client 101 reaches the load balancer 103 via the network 102. The load balancer 103 is configured to distribute the load by distributing requests to the business processing servers (the business processing server A 104, the business processing server B 105, the business processing server C 106, and the business processing server Z 109). Yes. These business processing servers are Web servers in a Web system, for example.

  The management server 111 (computer management server) monitors the load state of the system in the calculation center, and if an overload condition has occurred, controls such as dynamically adding an unused server in the center to the system. Is realized. Note that the dynamic system addition is performed, for example, when a computer system connected in advance to the network 102 transitions from a non-operating state to an operating state in accordance with an instruction from the management server 111. This is to be a business processing server to which a request is assigned from the load balancer 103.

  FIG. 2 is a configuration diagram showing the load balancer 103. FIG. 2A shows a configuration example in which the load balancer 103 and the management server 111 are separated into separate cases, and FIG. 2B shows the same case in which the load balancer 103 and the management server 111 are the same case. The structural example accommodated in the body is shown. The load balancer 103 and the management server 111 include a main storage device 201, a CPU 202, and a communication management 203, and are connected to each other via a bus (not shown). The CPU 202 executes a program stored in the main storage device 201.

  In FIG. 2A, the load distribution processing unit 204 and the load distribution reference table 216 are stored in the main storage device 201 of the load distribution device 103, and priority control is performed in the main storage device 201 of the management server 111. A unit 207, a load status monitoring unit 205, and a load distribution management table 206 are stored. In FIG. 2B, the load distribution management table 206 also serves as the load distribution reference table 216.

  The load distribution processing unit 204 distributes a request received from the user client 101 to each business processing server while distributing the load. The load status monitoring unit 205 monitors the load status of each business processing server. The load distribution management table 206 records the load status of each business processing server and the measured processing capacity of each business processing server. The priority control unit 207 controls the priority 307 (see FIG. 3) stored in the load distribution management table 206 in order to measure the processing capability of the business processing server. The load distribution reference table 216 is a table that is referred to when executing load distribution, and at least a part of the data stored in the load distribution management table 206 is copied.

  3 to 6 are configuration diagrams showing the load distribution management table 206. In the load distribution management table 206, the load balancing destination business processing server name 301, the current processing amount 302 that is the latest load state measured by the load state monitoring unit 205, the current CPU load 303, and the server are given priority. The upper limit 304 of the processing amount when the degree is set to 1, the CPU load allowable upper limit 305 determined in advance, the processing capacity 306 obtained by performance measurement, and the priority 307 when performing load distribution according to priority are recorded. The content of each data of FIGS. 3-6 is demonstrated in detail by description of the flowchart mentioned later.

  Note that the load distribution reference table 216 has at least the current processing amount 302, processing capability 306, and priority 307. Then, the load balancer 103 selects the business processing servers in descending order of priority 307, and assigns requests to the selected business processing servers until the current processing amount 302 of the selected business processing servers reaches the processing capacity 306. Continue.

  FIG. 7 is an explanatory diagram illustrating an example of the relationship between the processing amount and the CPU load. The management server 111 measures the CPU load while gradually increasing the current processing amount 302 limited to the processing amount upper limit 304, and the CPU load allowable upper limit 305 (90% in this embodiment) of the load distribution management table 206. ) Exceeds the processing capacity 306 of the business processing server.

  FIG. 8 is a flowchart showing the load distribution process. This flowchart operates with the start of performance measurement. Hereinafter, during the operation of the business processing server A 104, the business processing server B 105, and the business processing server C 106 (the three are examples of the second server in the claims), the business processing server Z 109 (an example of the first server in the claims) ) Will be described.

  First, the load status monitoring unit 205 refers to the load distribution management table 206 and confirms that the current processing amount 302 of the business processing server Z109 is n cases / minute (S501). For example, before the server Z is added (FIG. 3), the current processing amount 302 of the business processing server Z109 is not allocated. This is because the priority 307 of the business processing server Z109 is not defined, and the load balancer 103 does not assign a request to the business processing server Z109.

  Next, the priority control unit 207 changes the load distribution method (S502). The priority control unit 207 removes the business process server Z109 from the load distribution target so far, and starts load distribution based on the priority order in which the business process server Z109 is set to the priority level 1. For example, during the performance measurement of the server Z (FIG. 4), with respect to the priority 307, the business processing server Z109 is 1 (high priority) and the other business processing servers are 2 (low priority). Thereby, the load balancer 103 preferentially assigns requests to the business processing server Z109.

  Here, the initial value of the upper limit 304 of the processing amount of the job processing server Z109 is the current processing amount 302 obtained in S501. When the current processing amount 302 is not assigned, the initial value is 0. Then, the load distribution apparatus 103 assigns a request so that the business processing server A 104, the business processing server B 105, and the business processing server C 106 process the processing exceeding the upper limit 304 of the processing amount of the business processing server Z 109. The load distribution method in the business processing server A 104, the business processing server B 105, and the business processing server C 106 may be the load distribution method that has been performed so far, or may be another load distribution method.

  The load status monitoring unit 205 compares the CPU load of the business processing server Z109 with the CPU load allowable upper limit 305 set in the load distribution management table 206, so that the CPU 202 of the business processing server Z109 has a margin. It is determined whether or not there is (S503).

  First, when the current CPU load 303 of the business processing server Z109 is smaller than the CPU load allowable upper limit 305 set in the load distribution management table 206 and the CPU 202 has a margin (S503, Yes), the process proceeds to S503B. For example, during the performance measurement of the server Z (FIG. 4), the current CPU load 303 of the business processing server Z109 is 25%, and there is still room for 90% of the CPU load allowable upper limit 305. When the current processing amount 302 of the business processing server Z109 reaches the processing amount upper limit 304 (S503B, Yes), the processing amount upper limit 304 of the business processing server Z109 is increased (S504). The amount to be increased may be a method of increasing the percentage of the processing amount so far, or may be a method of fixedly increasing the number of cases / minute. Thereafter, the process returns to S503.

  On the other hand, when the current CPU load 303 of the business processing server Z109 becomes larger than the CPU load allowable upper limit 305 set in the load distribution management table 206 and the CPU 202 has no room (S503, No), the current processing amount 302 is recorded as the processing capability 306 of the business server (S505). For example, when the performance measurement of the server Z is completed (FIG. 5), the current CPU load 303 is 90%, reaching 90% of the CPU load allowable upper limit 305. At this time, the management server 111 copies 90 cases / minute of the current processing amount 302 to the processing capability 306.

  When the CPU load of the business processing server Z109 greatly exceeds the CPU load allowable upper limit 305 set in the load distribution management table 206, it is estimated that the management server 111 has exceeded the CPU load allowable upper limit 305. The processing amount may be obtained by proportional calculation or the like and recorded as the processing capacity 306 of the business server (see FIG. 7).

  Finally, the priority control unit 207 returns the load distribution method to the business processing server A 104, the business processing server B 105, the business processing server C 106, and the business processing server Z 109 to the load distribution method before starting the performance measurement ( S506), the performance measurement process is terminated. For example, after the server Z is disconnected (FIG. 6), for the priority 307, the business processing server Z109 is undefined and the other business processing servers are 1 (high priority). As a result, the load balancer 103 does not allocate requests to the business processing server Z109.

  Through the above processing, the processing capability of the business processing server Z109 can be obtained.

  Similar performance measurement is performed for the business processing server A 104, the business processing server B 105, and the business processing server C 106, and the processing capability 306 in the load distribution management table 206 is obtained. Thus, by comparing the total processing capability 306 of the business processing server with the current request, the business processing server B 105 can be reduced if the number of requests from the user client 101 is less than 240 requests / minute. If the rate is less than 170 cases / minute, it can be seen that the number of the business processing server B 105 and the business processing server C 106 can be reduced.

  The present invention described above can be widely modified without departing from the spirit thereof as follows.

  For example, parameters related to computer resources other than the CPU, such as the memory usage rate and the network line usage rate, are measured instead of the CPU usage rate or in combination with the CPU usage rate as parameters to be measured. It is good to do.

  Further, in the present embodiment, the business processing server is targeted, but the processing to be subject to load distribution is not limited to business processing, and may be processing of various computers such as scientific computing processing.

It is a block diagram which shows the load distribution system regarding one Embodiment of this invention. It is a block diagram which shows the load distribution apparatus regarding one Embodiment of this invention. It is a block diagram which shows the load distribution management table (before server Z addition) regarding one Embodiment of this invention. It is a block diagram which shows the load distribution management table (during the performance measurement of the server Z) regarding one Embodiment of this invention. It is a block diagram which shows the load distribution management table (Performance measurement of the server Z is completed) regarding one Embodiment of this invention. It is a block diagram which shows the load distribution management table (after server Z isolation | separation) regarding one Embodiment of this invention. It is explanatory drawing which shows the example of the relationship between the processing amount regarding one Embodiment of this invention, and CPU load. It is a flowchart which shows the load distribution process regarding one Embodiment of this invention.

Explanation of symbols

101 User client 102 Network 103 Load balancer 104 Business processing server A
105 Business processing server B
106 Business processing server C
109 Business processing server Z
111 Management Server 201 Main Storage Device 202 CPU
203 Communication Management 204 Load Distribution Processing Unit 205 Load Status Monitoring Unit 206 Load Distribution Management Table 207 Priority Control Unit 216 Load Distribution Reference Table

Claims (9)

A computer management method for measuring the processing capacity of a first server, which is a performance measurement target, in a plurality of servers executing requests distributed based on a load distribution management table,
Computer
A priority control procedure for changing the load distribution management table stored in the storage means so that the request is preferentially assigned to the first server over the second server that is executing the request;
A load status monitoring procedure for recording the processing amount at the time when the performance of the first server reaches a predetermined allowable upper limit as the processing capacity of the first server;
The computer management method characterized by performing.   The load distribution management table stores an upper limit of the processing amount for each server, and a request exceeding the upper limit of the processing amount of the first server is configured to be allocated to the second server. The computer management method according to claim 1.   3. The computer management method according to claim 2, wherein the load distribution management table uses an initial value of an upper limit of a processing amount of the first server as a processing amount of the first server before starting performance measurement.   The priority control procedure changes the load distribution management table so that a request is not assigned to the first server after recording the processing capability of the first server in the load status monitoring procedure. The computer management method according to any one of claims 1 to 3.   In the load status monitoring procedure, whether or not the performance of the first server has reached a predetermined allowable upper limit is determined based on at least one of a CPU usage rate, a memory usage rate, and a network line usage rate related to the first server. 5. The computer management method according to claim 1, wherein the computer management method is determined with reference to one usage rate. 6.   The load distribution management table is configured to increase the upper limit of the processing amount of the first server when the current processing amount of the first server reaches the upper limit of the processing amount of the first server. The computer management method according to claim 2, wherein:   The computer management method according to claim 1, wherein the first server and the second server are Web servers.   A computer management program for causing a computer to execute the computer management method according to any one of claims 1 to 7. In a plurality of servers that execute requests distributed based on a load distribution management table, a computer management server that measures the processing capacity of a first server that is a performance measurement target,
A priority control unit that changes the load distribution management table so that the request is preferentially assigned to the first server over the second server that is executing the request;
A load status monitoring unit for recording the processing amount at the time when the performance of the first server reaches a predetermined allowable upper limit as the processing capacity of the first server;
A computer management server characterized by comprising:

Images