JP2005148799A - Method for processing information, method for deriving service time and method for adjusting number of processing unit - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To derive the service time of a processing unit in an operating system under operation. <P>SOLUTION: A method for processing information includes: a step of acquiring, for a processing unit executing a predetermined service, a processing time to be measured during operation and being the sum of a queuing time and a service time while the queuing time and the service time are separated from each other and of storing the acquired processing time into a storage device; a simulation execution step of calculating the queuing time corresponding to an estimated service time stored in a service time storing part and set based on a predetermined rule by queue simulation, and of storing the queuing time into a queue storing part; and a repetition step of resetting the estimated service time based on a predetermined rule and storing the estimated service time into a service time storing part until a difference between the sum of the estimated service time and the queuing time stored in the queuing time storing part and the processing time stored in the storage device falls within an allowable error range, or until specific conditions are satisfied, and of executing the simulation execution step. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a technique for deriving a service time of a processing unit in an operating system and a technique for using the derived service time. In the present application, the processing unit refers to a unit of hardware that executes a predetermined process such as a server constituting a cluster, a memory and CPU set included in a computer, or a processor.

  In recent years, there are an increasing number of cases in which a cluster configuration is used in a system that provides services via the Internet. The system that provides services via the Internet includes, for example, a web (Web) front server that accepts requests from users and authenticates, an application server that performs application processing, and a DB server that accesses a database (DB). And are included. When a cluster configuration is applied to such a system, for example, the Web front server is configured by a plurality of servers having similar functions, and processing is executed in a form in which the load is distributed. In addition, for example, a plurality of servers included in the Web front server are divided into a processing group and a standby group, and the cluster configuration takes into account switching at the time of failure and addition of servers at high load. In some cases. Further, for example, a server included in the Web front server also has an application processing function, and in some cases, the server may be switched to a server constituting the application server.

There are various techniques for increasing the processing efficiency in a system using a cluster configuration that includes various forms. For example, there is a technique for selecting the server that can be processed earliest and executing the process (see, for example, Patent Document 1). In other words, an information processing apparatus including a plurality of servers and clients, a prediction request unit that requests a plurality of servers to predict the time required for processing, and a processing time predicted in response to a request from the prediction request unit Evaluation means for evaluating the above, means for selecting the server with the shortest processing time from the processing times of the plurality of servers predicted by the evaluation means, and means for sending the processing to the server selected by the selection means.
Japanese Patent Laid-Open No. 7-134684

  On the other hand, for example, when the service content (program to be executed) is changed or the amount of data to be handled is increased, it is necessary to update the service time of the queue model used for determining the resource amount. The service time in the queuing model is obtained by subtracting the waiting time from the processing time, and conventionally, the processing time during non-load has been used as a value. Also in the above-described prior art (Patent Document 1), the processing of the server including the waiting time using the service time calculated based on the hardware processing capacity measured in advance, that is, the processing time when there is no load. Derived time.

  FIG. 1 shows a schematic diagram of processing time in the queuing model. In FIG. 1, the queue model includes a queue 102 and a processing element 104. It is assumed that data to be processed such as a transaction (including processing contents and instructions) flows from the left to the right in the figure. For example, when processing is performed by the first-in first-out method (first-in first-out), the transaction is first placed at the end of the queue 102. In the example of FIG. 1, for example, five transactions are already included in the queue 102, so the queue 102 is arranged in the sixth (leftmost in the figure). The processing is performed in order from the top of the queue 102. For example, the time from when the sixth queue is arranged until the previous five transactions are processed and the turn is the waiting time 106. The time required for actual processing is the service time 108. The processing time 100 is a time from when a transaction is input until the processing is completed, and is a total of the waiting time 106 and the service time 108. Note that, when there is no load, the queue 102 does not exist, and the processing time 100 is equal to the service time 108. In the prior art, the waiting time 106 and the processing time 100 are derived based on the processing time 100 (= service time 108) when there is no load.

  However, the operating system is usually loaded, and it is difficult to measure the service time, that is, the processing time when there is no load. For this reason, in order to update the service time of the queuing model, it is necessary to temporarily set the operation status of the system to a non-load state. On the other hand, it is not realistic to temporarily stop a system that is required to continue operating, such as a system that operates 24 hours, and measure the processing time when there is no load.

  Therefore, an object of the present invention is to provide a technique for enabling derivation of the service time of a processing unit in an operating system.

  Another object of the present invention is to provide a technique for appropriately setting a cluster resource amount in an operating system.

  The method according to the first aspect of the present invention is the sum of waiting time and service time for a processing unit that executes a predetermined service, and the waiting time and service time are measured during operation without being separated. A waiting time corresponding to the estimated service time stored in the service time storage unit and set based on a predetermined rule is calculated by queue simulation, and stored in the queue. Until the difference between the simulation execution step stored in the storage unit and the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device falls within an allowable error range or a specific condition The estimated service time is reset based on the predetermined rule and stored in the service time storage unit until And a repetition step of performing the steps.

  In this way, services that are inherently difficult to measure while the system is operating by repeating the queuing simulation based on the estimated service time based on the actual processing time (including the waiting time) measured while the system is operating. Time can be derived. In other words, when the estimated service time was set and the queue simulation was executed, the sum of the estimated service time and the waiting time obtained as a result of the queue simulation (simulation result processing time) was actually measured. If the processing time is sufficiently close, the set estimated service time can be regarded as sufficiently close to the actual service time. If the processing time of the simulation result is significantly different from the actually measured processing time, the estimated service time is also significantly different from the actual service time (correct answer), so the estimated service time is reset.

  The method according to the second aspect of the present invention is a method for adjusting the number of processing units for adjusting the number of processing units constituting a cluster of one or a plurality of clusters constituting the system, which is measured during operation of the system. A first queuing simulation is performed using the processing time of the processing unit, and the service time of the processing unit is derived and stored in the service time storage unit; and the processing unit stored in the service time storage unit Based on the service time, the step of executing the second queue simulation for each cluster and the processing time for each cluster calculated by the second queue simulation are summed to calculate the processing time of the entire system. Stored in the storage unit and calculated by the second queuing simulation Configure a cluster based on the step of storing the operating rate for each processing unit in the operating rate storage unit, the processing time of the entire system stored in the system processing time storage unit, and the operating rate stored in the operating rate storage unit Determining an appropriate value for the number of processing units and storing it in the processing unit number storage unit.

  In this way, by further executing a queuing simulation using the service time derived based on the actual processing time measured during system operation, the processing time for each cluster constituting the system is calculated, and each cluster is calculated. The number of processing units (composing each cluster) is determined based on the processing time. Thereby, an appropriate number of processing units can be determined based on the service time even during system operation.

  In the second aspect of the present invention, when the determination step is performed when the processing time of the entire system stored in the system processing time storage unit is shorter than a predetermined shortest processing time, the operation stored in the operation rate storage unit Based on the rate, a step of identifying a cluster including a processing unit having the minimum operation rate from among the clusters constituting the system, and a second queuing simulation with the number of processing units reduced by 1 from the identified cluster Steps may be included.

  Further, in the second aspect of the present invention, the processing unit connection switching device allows the number of processing units constituting the cluster to be equal to an appropriate value of the number of processing units constituting the cluster stored in the processing unit number storage unit. A step of switching the connection may be further included. As a result, the number of processing units can be dynamically adjusted even during system operation.

  The method according to the third aspect of the present invention is the sum of waiting time and service time for a processing unit that executes a predetermined service, and the waiting time and service time are measured during operation without being separated. A service time deriving method for deriving a service time from a processing time, wherein the processing time of the processing unit measured during the operation of the processing unit and stored in the processing time storage unit is used as the estimated service time in the service time storage unit. A step of storing, a simulation execution step of performing a queuing simulation based on the estimated service time stored in the service time storage unit, and storing the waiting time calculated by the queue simulation in the waiting time storage unit, and storing the waiting time Waiting time stored in the service and estimated service time stored in the service time storage The error calculation step of calculating the difference between the sum of the two and the processing time stored in the processing time storage unit and storing the difference in the error storage unit as the first difference, and the first difference stored in the error storage unit A reset step of resetting the estimated service time and storing the estimated service time in the service time storage unit when the error is outside the predetermined allowable error range. Then, when the first difference stored in the error storage unit is within a predetermined allowable error range, the estimated service time stored in the service time storage unit is regarded as the service time of the processing unit.

  In this way, by first setting the actual processing time to the estimated service time and performing the queue simulation, the estimated service time may be reset from a relatively close value depending on the queue status. it can. For example, if there is no queue such as a low transaction occurrence rate, the estimated service time set first becomes equal to the actual service time without resetting.

  In the third aspect of the present invention, the reference value storage unit includes a first reference value having a value less than or equal to the maximum service time, a value smaller than the first reference value and greater than or equal to the minimum service time. A second reference value having a value is stored, and the resetting step includes a step in which the first difference stored in the error storage unit is outside a predetermined error allowable range and the first difference is positive. If the value is a value, the first reference value is rounded down and stored in the reference value storage unit, and a new estimated service time is set between the new first reference value and the second reference value. The step of storing in the time storage unit and the second reference value when the first difference stored in the error storage unit is outside a predetermined error allowable range and the first difference is a negative value Is rounded up and stored in the reference value storage unit, and the new second reference value and first reference value It may include a step of storing the service time storage unit by setting a new estimated service time during.

  In this way, the estimated service time can be brought closer to the actual service time (correct answer) by gradually narrowing the reset range of the estimated service time. That is, when the processing time of the simulation result is longer than the actually measured processing time, the estimated service time is set to be shortened by reducing the first reference value, and the processing time of the simulation result is If it is shorter than the actually measured processing time, the second reference value is rounded up to reset the estimated service time longer. Since the magnitude relationship between the first reference value and the second reference value does not reverse, the reset range of the estimated service time becomes narrower.

  In the third aspect of the present invention, the step of storing the processing time of the processing unit in the maximum value storage unit as an initial value of the service time maximum value, and the initial value of the service time minimum value as zero in the minimum value storage unit A step of storing, wherein the resetting step is when the first difference stored in the error storage unit is outside a predetermined error tolerance range and the first difference is a positive value. The service time maximum value is reset using the estimated service time and stored in the maximum value storage unit, and the estimated service time is reset using the average value of the estimated service time and the minimum service time, and the service time is reset. Storing in the storage unit.

  In the third aspect of the present invention, the difference between the service time maximum value reset and stored in the maximum value storage unit and the estimated service time reset and stored in the service time storage unit When the difference between the two is outside the predetermined approximate range, the simulation execution step and subsequent steps may be re-executed based on the estimated service time.

  It is also possible to create a program for causing a computer to execute the method according to the present invention, and the program is a storage medium such as a flexible disk, a CD-ROM, a magneto-optical disk, a semiconductor memory, and a hard disk. Alternatively, it is stored in a storage device. Moreover, it may be distributed as a digital signal via a network. Note that data being processed is temporarily stored in a storage device such as a computer memory.

  According to the present invention, it is possible to grasp the service time of a processing unit even during system operation. Also, the cluster resource amount (number of processing units) in the operating system can be set appropriately.

  A system configuration diagram according to an embodiment of the present invention is shown in FIG. For example, the network 1 that is the Internet is connected to the server switching device 5 and one or a plurality of user terminals 3 that are personal computers, for example, wirelessly or by wire. The user terminal 3 has a web (Web) browser function (not shown) and can browse web pages. Note that the user terminal 3 may be a mobile phone having a Web browser function, another mobile information terminal, or the like. A cluster 50, a cluster 54, a processing time storage unit 7, and a management server 21 are connected to the server switching device 5. In addition, one or a plurality of clusters indicated by “...” Are connected. The cluster 50 includes a server 51 and a server 53, and the cluster 54 includes a server 55. Each cluster may also include one or more servers. Each server includes a processing time log generation unit. The server 51 includes a processing time log generation unit 510, the server 53 includes a processing time log generation unit 530, and the server 55 includes a processing time log generation unit 550. These processing time log generation units generate a processing time log for each server and store it in the processing time storage unit 7.

  A processing time storage unit 7 and a service time storage unit 11 are connected to the service time determination device 9, and the service time determination device 9 executes processing with reference to the processing time storage unit 7 and displays the processing result. Stored in the service time storage unit 11. The performance evaluation device 15 is connected to a service time storage unit 11, a system requirement DB 13, a server / service time DB 17, and an evaluation result storage unit 19. The performance evaluation device 15 executes processing with reference to the service time storage unit 11 and the system requirement DB 13, and stores the processing result in the server / service time DB 17 and the evaluation result storage unit 19. The evaluation result storage unit 19 is also connected to the management server 21, and the management server 21 performs processing with reference to the evaluation result storage unit 19.

  In addition, the servers constituting each cluster are each connected to or connectable to the server switching device 5. The cluster resources (the number of servers constituting the cluster) are managed by the management server 21, and the management server 21 outputs a resource switching instruction to the server switching device 5 as necessary. For example, an instruction is issued to increase the number of servers included in the cluster 50.

  The outline of the processing on the server side (cluster 50 or the like) shown in FIG. 2 will be described with reference to FIG. The server processing outline diagram of FIG. 3 includes a Web front server 300, an application server 310, and a DB server 320. These are server groups that functionally classify servers. The Web front server 300 includes a server A (302) and a server F (304). The server A (302) and the server F (304) are each composed of a plurality of servers (processing units), and are indicated by circle numbers. That is, the server A (302) and the server F (304) are clusters each composed of three servers.

  The application server 310 includes a server B (312), a server D (314), and a server E (316). Each server is composed of two servers. The DB server 320 includes a server C (322). The server C (322) is composed of one server. In addition, the square of the vertical line or horizontal line pattern shown on the left or top of each server (server A (302) etc.) represents the queue. 2 correspond to the server A (302) and the server B (312) in FIG. 3, and the increase or decrease in the resource (number of servers) is the server A (302) or the server B ( 312) is indicated by the increase or decrease of the circled numbers.

  Data to be processed (including processing contents and instructions) such as a transaction is input from the upper left of the figure, and is first processed by the server A (302). Server A (302) outputs the processing result to server B (312), and server B (312) outputs the processing result to at least one of server D (314) and server C (322). At least one of the server D (314) and the server C (322) outputs the processing result to the server E (316), and the server E (316) outputs the processing result to the server F (304). The server F (304) transmits the processing result to the user terminal 3 (FIG. 2) via the network 1, for example. A series of processing is performed in this way, and the total processing time including the waiting time in all servers that have performed processing is the processing time (response time) of the entire system.

  FIG. 4 shows an example of the table configuration of the system requirement DB 13 (FIG. 2) and stored data. The example of FIG. 4 includes a requirement item column 400 and a set value column 402. In this table, a system administrator or the like stores data in advance. For example, in the first line, the requirement item column 400 indicates “response time of the entire system”, and the setting value column 402 indicates “maximum value 2 sec. Minimum value 1 sec”. That is, the system is set such that the processing time of the entire system needs to be between 1 second and 2 seconds including the waiting time. If the processing time of the entire system is 3 seconds, for example, it is necessary to take measures such as increasing the resources of any cluster in order to shorten it by 1 second. In the example of FIG. 4, the maximum operating rate of each server is registered in the second and subsequent lines. For example, if there is a server with an operation rate exceeding this value, there is a risk of malfunction, so it is necessary to take measures such as increasing the resources of the cluster including the server. Although not shown, the system requirement DB 13 is provided with a table that stores identification data of servers constituting the cluster, model type, and setting data related to software (program) to be executed.

  FIG. 5 shows an example of the table structure of the server / service time DB 17 (FIG. 2) and stored data. The example of FIG. 5 includes a model type 500 column, a software / OS (Operating System) column 502, and a service time column 504. The data is stored as a processing result of the performance evaluation device 15 (FIG. 2). Since the service time mainly depends on the server model type, the software (program) to be executed, and the OS, storing such data can be used for processing time simulation when designing a new system. It becomes possible.

  FIG. 6 shows a functional block diagram of the service time determination device 9 (FIG. 2). The service time determination device 9 shown in FIG. 6 includes an initial setting unit 600, a queue simulator 610, an error calculation unit 620, a maximum error extraction unit 630, a first comparison unit 640, a set value adjustment unit 650, and a second comparison unit 660. And a processing server control unit 670 are included. Details of the processing will be described later in the description of the processing flow. 6 includes a processing time storage unit 6007 and a service time storage unit 6011, which are equivalent to the processing time storage unit 7 and the service time storage unit 11 shown in FIG.

  Although not shown in FIG. 2, the service time determination device 9 has a work memory area. The work memory area stores a first server setting value storage unit 601 and a second server setting value. Unit 602, third server set value storage unit 603,..., Nth server set value storage unit 605, first server waiting time storage unit 611, second server waiting time storage unit 612, third server waiting time. Storage unit 613,..., And n-th server waiting time storage unit 615, first server error storage unit 621, second server error storage unit 622, third server error storage unit 623,. A server error storage unit 625, a specific number storage unit 631, and a reset value storage unit 651 are included. Here, n is the number of servers included in the system, and there are n server setting value storage units, server waiting time storage units, and server error storage units.

The first server setting value storage unit 601, the second server setting value storage unit 602, the third server setting value storage unit 603,..., And the nth server setting value storage unit 605 each have a processing time. (T _i (i is an integer from 1 to n, for example)), minimum service time (smin _i ), maximum service time (smax _i ), and estimated service time (s _i ) can be stored. It has become.

  Furthermore, the service time determination device 9 includes a request frequency storage unit 608, a first approximate parameter storage unit 638, and a second approximate parameter storage unit 658. In these, predetermined values are stored in advance.

The initial setting unit 600 refers to the processing time storage unit 6007 and stores the processing results (T _i , smin _i , smax _i and s _i ) as the first server setting value storage unit 601, the second server setting value storage unit 602, The three server setting value storage units 603,... And the nth server setting value storage unit 605 are stored in corresponding storage units.

The queue simulator 610 includes a first server setting value storage unit 601, a second server setting value storage unit 602, a third server setting value storage unit 603,..., And an nth server setting value storage unit 605 and a request frequency storage. Section 608, each processing result (w _i ) is stored in the first server latency storage unit 611, the second server latency storage unit 612, the third server latency storage unit 613, and the nth server. Store in the corresponding storage unit of the waiting time storage unit 615.

The error calculation unit 620 includes a first server setting value storage unit 601, a second server setting value storage unit 602, a third server setting value storage unit 603, ..., an nth server setting value storage unit 605, Referring to the server waiting time storage unit 611, the second server waiting time storage unit 612, the third server waiting time storage unit 613, ..., and the nth server waiting time storage unit 615, each processing result (| (s _i + w _i −T _i ) / T _i |) is stored in the first server error storage unit 621, the second server error storage unit 622, the third server error storage unit 623,. Store in the corresponding storage.

  The maximum error extraction unit 630 refers to the first server error storage unit 621, the second server error storage unit 622, the third server error storage unit 623,. j) is stored in the specific number storage unit 631. “J” is a number corresponding to a specific server.

  The first comparison unit 640 includes a specific number storage unit 631 that stores a specific server number “j”, a jth server setting value storage unit that corresponds to the specific server number “j”, and a specific server number “j”. The setting value adjustment unit 650 or the processing server control unit 670 is caused to execute processing according to the processing result with reference to the jth server waiting time storage unit corresponding to the first and the first approximate parameter storage unit 638.

The set value adjustment unit 650 refers to the specific number storage unit 631 that stores the specific server number “j” and the jth server set value storage unit that corresponds to the specific server number “j”, and determines the processing result (s _j and smax _j or smin _j ) are stored in the reset value storage unit 651 and data (s _j and smax) stored in the j-th server set value storage unit corresponding to the specific server number “j” _j or smin _j ) is updated.

The second comparison unit 660 refers to the reset value storage unit 651 and the second approximate parameter storage unit 658 and causes the processing server control unit 670 or the queue simulator 610 to execute processing according to the processing result. The processing server control unit 670 causes the maximum error extraction unit 630 to execute processing according to the processing result, or stores the first server setting value storage unit 601, the second server setting value storage unit 602, and the third server setting value storage. The data (s _i ) is stored in the service time storage unit 6011 with reference to the units 603,... And the nth server set value storage unit 605.

  Next, a processing flow of processing (service time calculation processing) performed by the service time determination device illustrated in FIG. 6 will be described with reference to FIG.

First, the initial setting unit 600 sets each processing time (T _i (i is an integer from 1 to n (number of servers), for example) from the processing time storage unit 6007 to a number corresponding to each server. )), And corresponding storage units of the first server setting value storage unit 601, the second server setting value storage unit 602, the third server setting value storage unit 603, ..., and the nth server setting value storage unit 605 (Step S1). The processing time storage unit 6007 stores, for example, the average processing time per transaction of each server generated by the processing time log generation units 510 to 550 (FIG. 2), and reads this value. When only the log data is stored in the processing time storage unit 6007, the initial setting unit 600 calculates, for example, the average processing time per transaction of each server based on the log data, and stores each server setting value. Store in the department.

Further, the initial setting unit 600 stores “0” in the service time minimum value storage area for all servers, and stores the processing time read above in the estimated service time storage area and the service time maximum value storage area (steps). S3). That is, in each server setting value storage unit, “0” is set in smin _i , and the value of T _i is set in s _i and smax _i .

Then, the queue simulator 610 performs a queue simulation using the estimated service time (s _i ) and the predetermined request frequency λ stored in the request frequency storage unit 608 for all servers. The time (w _i ) is calculated and the first server waiting time storage unit 611, the second server waiting time storage unit 612, the third server waiting time storage unit 613,. Store in the corresponding storage unit (step S5). For specific methods of queuing simulation, please refer to the prior art (for example, ““ Simulation of Queue ”, Internet <URL: http://www.cis.nagasaki-u.ac.jp/labs/oguri/CompSimExm .pdf>"), and a detailed description is omitted.

Then, the error calculation unit 620 uses, for each server, the processing time (T _i ), the estimated service time (s _i ), and the waiting time (w _i ) stored in each server waiting time storage unit, s _i + w _i −T _i ) / T _i | is calculated, and the first server error storage unit 621, the second server error storage unit 622, the third server error storage unit 623,. The data is stored in the corresponding storage unit of the error storage unit 625. The maximum error extraction unit 630 refers to each of the server error storage units, specifies | (s _j + w _j −T _j ) / T _j | having the largest value, and stores “j” in the specific number storage unit 631. (Step S7). That is, the server (j) where the maximum error occurs is specified.

Then, the first comparison unit 640 uses the estimated service time (s _j ), waiting time (w _j ), and processing time (T _j ) of the identified server (j) to determine s _j + w _j and T _j (Step S9), and using ε1 which is a predetermined approximate parameter stored in the first approximate parameter storage unit 638, it is determined whether s _j + w _j is sufficiently close to T _j (step S11). . That is, it is determined whether the difference between the sum of the estimated service time and the waiting time and the actually measured processing time is within an allowable error range. Specifically, if the absolute value of the difference between the values (for example, | s _j + w _j −T _j |) is smaller than ε1, which is a predetermined approximate parameter, it is determined that the values are sufficiently close. When it is determined that the distance is sufficiently close (step S11: Yes route), the process proceeds to step S25 described later. On the other hand, when it is determined that it is not sufficiently close (step S11: No route), the setting value adjustment unit 650 instructed to process by the first comparison unit 640 determines whether s _j + w _j is greater than T _j . Determination is made (step S13).

If it is not greater (Step S13: No route), the setting value adjustment unit 650 is instructed to process the first comparing unit 640 sets the s _j in smin _j, s _j to s _j and smax _j And the average value ((s _j + smax _j ) / 2) is set and stored in the reset value storage unit 651, and the first server set value storage unit is used by using the reset smin _j and s _j 601, the second server setting value storage unit 602, the third server setting value storage unit 603,..., And the corresponding storage unit (server (j)) of the nth server setting value storage unit 605 are updated ( Step S15). That is, the processing time calculated as a result of the simulation is less than or equal to the actually measured processing time means that the estimated service time (s _j ) used in the simulation is shorter than the actual (sought) value. Since it is considered that it was set, smin _j is rounded up and the estimated service time is set longer. If s _j and smin _j is reset, the second comparing section 660 compares the s _j and smin _j stored in the re-set and re-set value storage unit 651 (step S17). And it transfers to the process of step S23 mentioned later.

On the other hand, when it is determined that s _j + w _j is larger than T _j (step S13: Yes route), the set value adjustment unit 650 sets s _j to smax _j , and sets s _j to s _j and smin _j . An average value ((s _j + smin _j ) / 2) is set and stored in the reset value storage unit 651, and the first server set value storage unit 601 is used by using the reset smax _j and s _j . The data of the storage unit (server (j)) corresponding to any one of the second server set value storage unit 602, the third server set value storage unit 603,... And the nth server set value storage unit 605 is updated ( Step S19). In other words, the processing time calculated as a result of the simulation is longer than the actually measured processing time, which means that the estimated service time (s _j ) used in this simulation is set longer than the actual (to be obtained) value. Smax _j is rounded down and the estimated service time is shortened. When s _j and smax _j are reset, the second comparison unit 660 compares the reset s _j with smax _j (step S21).

Then, the second comparison unit 660 determines whether s _j and smin _j or s _j and smax _j are sufficiently close based on the comparison result of step S17 or step S21 (step S23). Specifically, if the absolute value of the value difference (for example, | s _j −smin _j |) is smaller than a predetermined approximate parameter ε2, it is determined that the difference is sufficiently close. In such a case, since it is difficult to obtain a desired result even if further processing is performed, the processing for this server is terminated. When it is determined that it is not sufficiently close (step S23: No route), the process returns to the process of step S5, and the queue simulator 610 instructed to process by the second comparison unit 660, in step S15 or step S19. Perform a queuing simulation based on the reset and updated value (s _j ).

On the other hand, when it is determined that the processing is sufficiently close (step S23: Yes route), the processing server control unit 670 instructed to process by the second comparison unit 660 determines whether the processing for all the servers has been completed (step S25). ). When it is determined that the processes for all the servers have not been completed (step S25: No route), the maximum error extraction unit 630 determines | (s _j + w _j −T _j ) among the servers that have not been processed yet. The server (j) having the maximum value of / T _j | is specified (step S27). Then, the process returns to step S9.

On the other hand, when it is determined that the processing for all the servers has been completed (step S25: Yes route), the processing server control unit 670 includes the first server setting value storage unit 601, the second server setting value storage unit 602, the third The estimated service time (s _i ) of all the servers stored in the server set value storage unit 603,..., And the nth server set value storage unit 605 is used as the service time calculation process result data. (Step S29). Then, the service time calculation process ends.

  In this way, the estimated service time is set, and the service time of each server is calculated by a method of gradually narrowing down the reset range of the estimated service time based on the simulation result. Thereby, it is possible to know the service time of each server even when the system is operating (without putting the system in a non-load state).

Next, a processing flow of processing (resource amount adjustment processing) performed by the performance evaluation device 15 (FIG. 2) will be described with reference to FIGS. Hereinafter, the main subject of the processing is the performance evaluation device 15 (FIG. 2), and the description may be omitted. First, the performance evaluation device 15 reads the resource amount of each cluster (the number of servers constituting the cluster) from the system requirement DB 13 (FIG. 2), for example, in a work memory area (not shown). The data is stored in the storage unit “X” and the storage unit “Y” (FIG. 8: step S41). The resource amount of each cluster is expressed as R _ki in the following description. For example, ki is an integer from 1 to kn (number of clusters), and is a number corresponding to each cluster.

Further, the performance evaluation device 15 reads the maximum and minimum values of the response time of the entire system and the operating rate upper limit of each server from the system requirement DB 13 (FIG. 2), and stores them in the work memory area (step S43). ). In the following description, the maximum response time of the entire system is expressed as RPmax, and the minimum value is expressed as RPmin. Similarly, the operating rate upper limit of each server is expressed as Kmax _i . i is an integer from 1 to n (the number of servers), for example, and is a number corresponding to each server.

The performance evaluation device 15 reads the service time of each server stored in the service time storage unit 11 (FIG. 2) and stores it in the work memory area (step S45). In the following description, the service time of each server represented as S _i. Based on the resource amount (R _ki ) of each cluster and the service time (S _i ) of each server, a queue simulation is performed for each cluster (step S47). As a specific method of queuing simulation, a conventional technique may be used, and detailed description is omitted, but the resource amount (R _ki ) is used as the number of service windows.

Then, the performance evaluation device 15 specifies the operating rate of each server and the response time of the entire system based on the result of the queue simulation, and stores each in the work memory area (step S49). The response time of the entire system is calculated by adding the processing time of each cluster. In the following description, the response time of the entire system RP, the operation rate of each server is expressed as K _i.

  Then, the performance evaluation device 15 determines whether the response time of the entire system calculated by the queue simulation is within the range of the maximum value and the minimum value of the response time of the entire system defined as system requirements, that is, RPmin ≦ It is determined whether RP ≦ RPmax is satisfied (step S51). When it is determined that RPmin ≦ RP ≦ RPmax is satisfied (step S51: Yes route), the resource amount does not need to be adjusted, so the processing shifts to the processing in FIG. To do.

  On the other hand, when it is determined that RPmin ≦ RP ≦ RPmax is not satisfied (step S51: No route), the response time of the entire system calculated by the queue simulation is the maximum of the response time of the entire system defined as the system requirements. It is determined whether it is longer than the value, that is, RP> RPmax (step S53). When it is determined that RP> RPmax is not satisfied (step S53: No route), the process proceeds to the process of FIG.

On the other hand, when it is determined that RP> RPmax (step S53: Yes route), the server (j) having the maximum operation rate and the operation based on the operation rate (K _i ) of each server specified in step S49 A server exceeding the rate upper limit, that is, a server (m) where K _m > Kmax _m is identified (step S55). “J” and “m” are numbers corresponding to a specific server. Then, the performance evaluation device 15 increases the resource amount (R _kj and R _km ) of the cluster including the specified server by 1 and stores it in X (step S57). “Kj” and “km” are numbers corresponding to a specific cluster. Then, the process returns to step S47, and the queue simulation is re-executed using the reset resource amount.

FIG. 9 shows a process after the transition through the terminal A (step S53: No route (FIG. 8)). The performance evaluation device 15 specifies the server (p) having the minimum operating rate based on the operating rate (K _i ) of each server specified in step S49 (FIG. 8) (FIG. 9: step S71). “P” is a number corresponding to a specific server. Here, the performance evaluation device 15 stores the current resource amount (R _ki ) of each cluster in the “X” (step S73). Then, the resource amount (R _kp ) of the cluster including the server (p) specified in step S71 is reduced by 1 and stored in the “Y” (step S75). “Kp” is a number corresponding to a specific cluster.

Then, the performance evaluation device 15 performs a queuing simulation based on the resource amount (R _ki ) of each cluster (value stored in the “Y”) and the service time (S _i ) of each server (step S77). ). Then, based on the result of the queue simulation, the response time (RP) of the entire system and the operating rate (K _i ) of each server are specified and stored in the work memory area (step S79).

Further, the performance evaluation device 15 determines whether or not the response time of the entire system calculated by the queue simulation is equal to or greater than the maximum value of the response time of the entire system defined as the system requirements, that is, RP ≧ RPmax ( Step S81). When it is determined that RP ≧ RPmax (step S81: Yes route), the resource amount (R _ki ) of each cluster stored in the “X” is stored in the evaluation result storage unit 19 (FIG. 2) as a processing result. Store (step S83). Then, the resource adjustment process ends.

On the other hand, when it is determined that RP ≧ RPmax is not satisfied (step S81: No route), it is determined whether RP ≧ RPmin is satisfied (step S85). That is, it is determined whether the response time of the entire system calculated by the queue simulation is within the range of the maximum value and the minimum value of the response time of the entire system defined as system requirements. If it is determined that RP ≧ RPmin is not satisfied (step S85: No route), the process returns to step S71. On the other hand, if it is determined that RP ≧ RPmin (step S85: Yes route), the evaluation result storage unit 19 (FIG. 2) uses the resource amount (R _ki ) of each cluster stored in “Y” as a processing result. (Step S87). Then, the resource amount adjustment process ends.

  In this way, the resource amount adjustment process is performed. That is, an appropriate resource amount for each cluster can be calculated based on the service time for each server stored in the service time storage unit 11 (FIG. 2).

  The appropriate resource amount stored in the evaluation result storage unit 19 (FIG. 2) is referred to by the management server 21 (FIG. 2), and is used to actually adjust the resource amount of each cluster. Specifically, the management server 21 first reads a resource amount (optimum resource amount) for each cluster from the evaluation result storage unit 19 and stores it in a storage device such as a work memory area. Then, for example, the current resource amount (current resource amount) of each cluster stored in a predetermined storage unit is read and stored in a storage device such as a work memory area. Note that the management server 21 may have a resource amount check function and check the current resource amount of each cluster. Then, the management server 21 compares the optimal resource amount with the current resource amount for each cluster, and if there is a different cluster, identifies the cluster and the optimal resource amount, and sends them to the server switching device 5 (FIG. 2). Instruct the switching process.

  The server switching device 5 performs cluster resource switching processing designated by the management server 21. That is, the resource amount of the cluster in which the optimum resource amount and the current resource amount are different is changed to the optimum resource amount. The server switching device 5 is a switching hub that supports, for example, VLAN (Virtual LAN), and logically reduces the amount of resources by changing the servers constituting the cluster from a connected state to a disconnected state, or a server in a disconnected state. Is connected and added to the cluster to increase the amount of resources. Moreover, the management server 21 switches software (program) operated on each server as necessary.

  Thereby, even when the system is in operation, the resource amount of each cluster can be dynamically controlled and kept in an appropriate state. Note that the processing may be partially stopped when the resource amount is changed (switched).

  The performance evaluation device 15 (FIG. 2) also performs server / service time registration processing in addition to the resource amount adjustment processing described above. The server service time registration process is performed periodically or in accordance with a request from a user such as an administrator. First, the performance evaluation device 15 reads the setting data related to the model type of the server constituting the system and the software (program) executed by the server from the system requirement DB 13 and stores it in a storage device such as a work memory area. . Further, the performance evaluation device 15 reads the service time of each server from the service time storage unit 11, and stores the service time of each server, the server model type and software (program) stored in the work memory area above. The server / service time record is generated by associating with the setting data relating to the server / service time DB 17 and stored in the server / service time DB 17. The data stored in the server / service time DB 17 can be used for a processing time simulation when designing a new system.

  Although one embodiment of the present invention has been described above, the present invention is not limited to this. For example, the schematic diagrams shown in FIGS. 1 and 3 are examples, and the queues, the length of processing time, the number of servers, and the like are not limited to the cases shown in the drawings. Further, the system configuration diagram shown in FIG. 2 is an example, and each device such as the management server 21 and the service time determination device 9 may be configured by a plurality of servers and computers. In the present embodiment, the resource amount is the number of servers constituting the cluster. However, the cluster is replaced with one server, for example, and the server constituting the cluster is a processing unit such as a memory and CPU set. It is also possible to apply the present invention in place of.

  The table configurations shown in FIGS. 4 and 5 are examples, and other configurations may be adopted as long as they store similar data, and items are added or deleted as necessary. May be. Further, the functional block configuration shown in FIG. 6 is an example, and may differ from the actual program / module configuration or memory configuration. Furthermore, the processing flows shown in FIG. 7 to FIG. 9 are also examples, and the processing order may be changed within a range in which similar processing results can be obtained, and steps may be added or deleted as necessary. .

(Appendix 1)
For a processing unit that executes a predetermined service, the processing time that is the sum of the waiting time and the service time and that is measured while the waiting time and the service time are not separated is acquired and stored in the storage device Steps,
A simulation execution step of calculating a waiting time corresponding to the estimated service time stored in the service time storage unit and set based on a predetermined rule by a queue simulation, and storing it in the queue storage unit;
Until the difference between the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device falls within an allowable error range or until a specific condition is satisfied Re-setting the estimated service time based on the predetermined rule, storing the estimated service time in the service time storage unit, and repeating the simulation execution step;
An information processing method executed by a computer.

(Appendix 2)
The reference value storage unit includes a first reference value having a value less than or equal to the maximum value of the service time, and a second reference value having a value smaller than the first reference value and greater than or equal to the minimum value of the service time; Is stored,
The repeating step comprises:
When the difference between the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device is a positive value, the first reference value And setting the new estimated service time between the new first reference value and the second reference value and storing the new estimated service time in the service time storage unit;
When the difference between the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device is a negative value, the second reference value Rounding up and storing in the reference value storage unit and setting a new estimated service time between the new second reference value and the first reference value and storing in the service time storage unit;
The information processing method according to appendix 1, including:

(Appendix 3)
For a processing unit that executes a predetermined service, a service time that is the sum of the waiting time and the service time and that derives the service time from the processing time measured during operation without the waiting time and the service time being separated. A derivation method comprising:
Storing the processing time of the processing unit measured during operation of the processing unit and stored in the processing time storage unit in the service time storage unit as an estimated service time;
A simulation execution step of performing a queuing simulation based on the estimated service time stored in the service time storage unit, and storing the waiting time calculated by the queue simulation in a waiting time storage unit;
A difference between the sum of the waiting time stored in the waiting time storage unit and the estimated service time stored in the service time storage unit and the processing time stored in the processing time storage unit is calculated. An error calculation step of storing in the error storage unit as the first difference;
A resetting step of resetting the estimated service time and storing in the service time storage unit when the first difference stored in the error storage unit is outside a predetermined error tolerance;
Including
If the first difference stored in the error storage unit is within a predetermined error tolerance, the estimated service time stored in the service time storage unit is regarded as a service time of the processing unit;
A service time deriving method executed by a computer.

(Appendix 4)
A method for adjusting the number of processing units for adjusting the number of processing units constituting a cluster for one or a plurality of clusters constituting the system,
Deriving a service time of the processing unit by performing a first queuing simulation using the processing time of the processing unit measured during operation of the system, and storing the service time in a service time storage unit;
Performing a second queuing simulation for each of the clusters based on the service time of the processing unit stored in the service time storage;
Totaling the processing time for each cluster calculated by the second queuing simulation to calculate the processing time of the entire system and storing it in the system processing time storage unit;
Storing an operation rate for each of the processing units calculated by the second queue simulation in an operation rate storage unit;
Based on the processing time of the entire system stored in the system processing time storage unit and the operating rate stored in the operating rate storage unit, an appropriate value of the number of processing units constituting the cluster is determined and stored. A decision step to store in the department;
A method for adjusting the number of processing units executed by a computer.

(Appendix 5)
The determining step comprises:
When the processing time of the entire system stored in the system processing time storage unit is longer than a predetermined longest processing time, based on the operating rate stored in the operating rate storage unit, among the clusters constituting the system Identifying a cluster including a processing unit having a maximum operating rate or a processing unit exceeding a predetermined maximum operating rate;
Performing the second queuing simulation with one processing unit added to the identified cluster;
The processing unit number adjusting method according to appendix 7, including:

(Appendix 6)
For a processing unit that executes a predetermined service, the processing time that is the sum of the waiting time and the service time and that is measured while the waiting time and the service time are not separated is acquired and stored in the storage device Steps,
A simulation execution step of calculating a waiting time corresponding to the estimated service time stored in the service time storage unit and set based on a predetermined rule by a queue simulation, and storing it in the queue storage unit;
Until the difference between the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device falls within an allowable error range or until a specific condition is satisfied Re-setting the estimated service time based on the predetermined rule, storing the estimated service time in the service time storage unit, and repeating the simulation execution step;
A program that causes a computer to execute.

(Appendix 7)
For a processing unit that executes a predetermined service, a service time that is the sum of the waiting time and the service time and that derives the service time from the processing time measured during operation without the waiting time and the service time being separated. A derivation program,
Storing the processing time of the processing unit measured during operation of the system and stored in the processing time storage unit in the service time storage unit as an estimated service time;
A simulation execution step of performing a queuing simulation based on the estimated service time stored in the service time storage unit, and storing the waiting time calculated by the queue simulation in a waiting time storage unit;
A difference between the sum of the waiting time stored in the waiting time storage unit and the estimated service time stored in the service time storage unit and the processing time stored in the processing time storage unit is calculated. An error calculation step of storing in the error storage unit as the first difference;
A resetting step of resetting the estimated service time and storing in the service time storage unit when the first difference stored in the error storage unit is outside a predetermined error tolerance;
To the computer,
If the first difference stored in the error storage unit is within a predetermined error tolerance, the estimated service time stored in the service time storage unit is regarded as a service time of the processing unit;
Service time derivation program.

(Appendix 8)
A processing unit number adjustment program for adjusting the number of processing units constituting a cluster for one or a plurality of clusters constituting the system,
Deriving a service time of the processing unit by performing a first queuing simulation using the processing time of the processing unit measured during operation of the system, and storing the service time in a service time storage unit;
Performing a second queuing simulation for each of the clusters based on the service time of the processing unit stored in the service time storage;
Totaling the processing time for each cluster calculated by the second queuing simulation to calculate the processing time of the entire system and storing it in the system processing time storage unit;
Storing an operation rate for each of the processing units calculated by the second queue simulation in an operation rate storage unit;
Based on the processing time of the entire system stored in the system processing time storage unit and the operating rate stored in the operating rate storage unit, an appropriate value of the number of processing units constituting the cluster is determined and stored. A decision step to store in the department;
For adjusting the number of processing units for causing a computer to execute.

(Appendix 9)
For a processing unit that executes a predetermined service, the processing time that is the sum of the waiting time and the service time and that is measured while the waiting time and the service time are not separated is acquired and stored in the storage device Means,
A simulation execution means for calculating a waiting time corresponding to the estimated service time stored in the service time storage unit and set based on a predetermined rule by a queue simulation, and storing the waiting time in the queue storage unit;
Until the difference between the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device falls within an allowable error range or until a specific condition is satisfied Repetitive means for resetting the estimated service time based on the predetermined rule, storing it in the service time storage unit, and executing the simulation execution means;
An information processing apparatus.

(Appendix 10)
For a processing unit that executes a predetermined service, a service time that is the sum of the waiting time and the service time and that derives the service time from the processing time measured during operation without the waiting time and the service time being separated. A deriving device,
Means for storing in the service time storage unit the processing time of the processing unit measured during the operation of the system and stored in the processing time storage unit as an estimated service time;
Simulation execution means for performing a queuing simulation based on the estimated service time stored in the service time storage unit, and storing the waiting time calculated by the queue simulation in the waiting time storage unit;
A difference between the sum of the waiting time stored in the waiting time storage unit and the estimated service time stored in the service time storage unit and the processing time stored in the processing time storage unit is calculated. Error calculation means for storing in the error storage as the first difference;
Resetting means for resetting the estimated service time and storing it in the service time storage unit when the first difference stored in the error storage unit is outside a predetermined error allowable range;
Have
If the first difference stored in the error storage unit is within a predetermined error tolerance, the estimated service time stored in the service time storage unit is regarded as a service time of the processing unit;
Service time deriving device.

(Appendix 11)
A processing unit number adjusting device that adjusts the number of processing units constituting a cluster for one or a plurality of clusters constituting the system,
Means for deriving a service time of the processing unit by performing a first queuing simulation using the processing time of the processing unit measured during operation of the system, and storing the service time in a service time storage unit;
Means for performing a second queuing simulation for each cluster based on the service time of the processing unit stored in the service time storage;
Means for calculating the processing time of the entire system by summing up the processing time for each cluster calculated by the second queuing simulation, and storing it in the system processing time storage unit;
Means for storing an operation rate for each processing unit calculated by the second queue simulation in an operation rate storage unit;
Based on the processing time of the entire system stored in the system processing time storage unit and the operating rate stored in the operating rate storage unit, an appropriate value of the number of processing units constituting the cluster is determined and stored. Determining means for storing in the department;
An apparatus for adjusting the number of processing units.

It is an outline figure of processing time in a queue model. 1 is a system configuration diagram according to an embodiment of the present invention. It is a server processing schematic diagram. It is a figure which shows an example of a system requirement DB table. It is a figure which shows an example of a server service time DB table. It is a figure which shows the functional block of a service time determination apparatus. It is a figure which shows the processing flow of a service time calculation process. It is a figure which shows the processing flow (the 1) of a resource amount adjustment process. It is a figure which shows the processing flow (the 2) of a resource amount adjustment process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Network 3 User terminal 5 Server switching apparatus 7 Processing time storage part 9 Service time determination apparatus 11 Service time storage part 13 System requirement DB 15 Performance evaluation apparatus 17 Server service time DB 19 Evaluation result storage part 21 Management server 50, 54 Cluster 51, 53, 55 Server 510, 530, 550 Processing time log generation unit

Claims (5)

For a processing unit that executes a predetermined service, the processing time that is the sum of the waiting time and the service time and that is measured while the waiting time and the service time are not separated is acquired and stored in the storage device Steps,
A simulation execution step of calculating a waiting time corresponding to the estimated service time stored in the service time storage unit and set based on a predetermined rule by a queue simulation, and storing it in the queue storage unit;
Until the difference between the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device falls within an allowable error range or until a specific condition is satisfied Re-setting the estimated service time based on the predetermined rule, storing the estimated service time in the service time storage unit, and repeating the simulation execution step;
An information processing method executed by a computer. A method for adjusting the number of processing units for adjusting the number of processing units constituting a cluster for one or a plurality of clusters constituting the system,
Deriving a service time of the processing unit by performing a first queuing simulation using the processing time of the processing unit measured during operation of the system, and storing the service time in a service time storage unit;
Performing a second queuing simulation for each of the clusters based on the service time of the processing unit stored in the service time storage;
Totaling the processing time for each cluster calculated by the second queuing simulation to calculate the processing time of the entire system and storing it in the system processing time storage unit;
Storing an operation rate for each of the processing units calculated by the second queue simulation in an operation rate storage unit;
Based on the processing time of the entire system stored in the system processing time storage unit and the operating rate stored in the operating rate storage unit, an appropriate value of the number of processing units constituting the cluster is determined and stored. A decision step to store in the department;
A method for adjusting the number of processing units executed by a computer. For a processing unit that executes a predetermined service, the processing time that is the sum of the waiting time and the service time and that is measured while the waiting time and the service time are not separated is acquired and stored in the storage device Steps,
A simulation execution step of calculating a waiting time corresponding to the estimated service time stored in the service time storage unit and set based on a predetermined rule by a queue simulation, and storing it in the queue storage unit;
Until the difference between the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device falls within an allowable error range or until a specific condition is satisfied Re-setting the estimated service time based on the predetermined rule, storing the estimated service time in the service time storage unit, and repeating the simulation execution step;
A program that causes a computer to execute. A processing unit number adjustment program for adjusting the number of processing units constituting a cluster for one or a plurality of clusters constituting the system,
Deriving a service time of the processing unit by performing a first queuing simulation using the processing time of the processing unit measured during operation of the system, and storing the service time in a service time storage unit;
Performing a second queuing simulation for each of the clusters based on the service time of the processing unit stored in the service time storage;
Totaling the processing time for each cluster calculated by the second queuing simulation to calculate the processing time of the entire system and storing it in the system processing time storage unit;
Storing an operation rate for each of the processing units calculated by the second queue simulation in an operation rate storage unit;
Based on the processing time of the entire system stored in the system processing time storage unit and the operating rate stored in the operating rate storage unit, an appropriate value of the number of processing units constituting the cluster is determined and stored. A decision step to store in the department;
For adjusting the number of processing units for causing a computer to execute. For a processing unit that executes a predetermined service, the processing time that is the sum of the waiting time and the service time and that is measured while the waiting time and the service time are not separated is acquired and stored in the storage device Means,
A simulation execution means for calculating a waiting time corresponding to the estimated service time stored in the service time storage unit and set based on a predetermined rule by a queue simulation, and storing the waiting time in the queue storage unit;
Until the difference between the sum of the estimated service time and the waiting time stored in the waiting time storage unit and the processing time stored in the storage device falls within an allowable error range or until a specific condition is satisfied Repetitive means for resetting the estimated service time based on the predetermined rule, storing it in the service time storage unit, and executing the simulation execution means;
An information processing apparatus.

Images