JP4472944B2 - Request load adjusting apparatus and method, program, and performance data collecting method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a request load adjusting apparatus and method for adjusting a load on a server due to a request, a program, and performance data for collecting data related to server performance while giving a constant load to the server by the request load adjusting apparatus or method. It relates to the collection method.
[0002]
[Prior art]
The technology that predictively evaluates the performance of online services provided by Web servers, etc. is the realization of autonomous computing aiming at self-configuration, self-optimization, etc. that does not involve humans, and performance that improves website performance. This is a basic technology required for services. In order to perform performance evaluation based on the system model, as shown in FIG. 6, a model of the target system is created (step 61), the performance of the system is measured (step 62), and the model is based on the measurement data. The parameters are estimated (step 63), and the completed model is analyzed by a queue or simulator (step 64).
[0003]
Conventionally, services for finding optimal parameters in the above-described procedure have been implemented for Web servers. However, when implementing services, the main parameters of the server software are set using the actual machine in which the target Web application is installed. While changing, the performance is actually measured by a load test, and the optimum parameter is found by an empirical rule based on the result.
[0004]
As a technique for performing a load test on a Web server, for example, a test server for testing the Web server to be tested is connected, and the test server generates a number of sessions forged with different port numbers. In this case, an HTTP request and an HTTP response are transmitted and received with a Web server (see, for example, Patent Document 1). In this technique, performance such as response time required for request / response and the fact that the object specified by the request is included without error in the response are measured. At that time, parameters such as arguments to be given to the CGI can be dynamically changed.
[0005]
On the other hand, methods for determining an optimal system configuration and server parameters without using an actual machine by queuing analysis and model analysis by simulation are also being put into practical use. However, even in this case, the contents of the Web application are Java (registered trademark) programs, and the contents differ for each customer. Therefore, in order to estimate the parameters and complete the model, it is necessary to measure at least once with the actual machine. There is.
[0006]
[Patent Document 1]
JP 2002-7232 A (paragraphs 0035-0038)
[0007]
[Problems to be solved by the invention]
According to the above-described conventional technology, a load test is performed at the final stage of Web application development, and sufficient measurement data can be collected, or there is an extra actual machine such as a staging server even when the system is in operation. There is no problem if measurement data can be collected by an environmental load test. However, when developing a Web application, the development period is short, and the cut-over is often performed only by checking the operation for debugging the application without performing a sufficient load test. Also, only a limited number of customers have staging servers and backup servers. In many cases, a server that is currently in operation is used as a development machine, so that it can be put into production. Don't have an extra real machine. Therefore, it is difficult to realize a sufficient load test at present. For this reason, a service request is received only when a performance problem in the system becomes apparent during operation. However, stopping a running system for a long time for a load test leads to lost opportunities, and it goes without saying that customers want to avoid it as much as possible.
[0008]
The most ideal solution that can be considered is to collect performance data such as the response time and resource usage of a live system and estimate the model parameters for analysis based on that data. It is done. However, the amount of requests to a Web server that is actually operating generally has a fairly complicated distribution of the number of arrivals per unit time, the average number of arrivals itself varies with time, and the mixing ratio of various requests also varies with time. It has the property of fluctuating. Collecting system performance data in such a state is of little use for estimating parameters for the model with current analysis techniques.
[0009]
An object of the present invention is to provide a technique capable of adjusting a load on a server due to a request without stopping a service by the server in view of the problems of the related art. Another object of the present invention is to provide a technology that can apply a steady load equivalent to that in a normal load test to a server without stopping the service provided by the server.
[0010]
[Means for Solving the Problems]
In order to achieve this object, the request load adjusting apparatus and method according to the present invention adjusts the load on the server due to the request by adding a pseudo request to the request transmitted to the server. It is characterized by that. A program according to the present invention causes a computer to function as a request load adjustment device according to the present invention, or causes a computer to execute the request load adjustment method according to the present invention. Further, the performance data collection method according to the present invention is characterized in that the request load adjustment device or the request load adjustment method according to the present invention collects data related to server performance while adjusting the load on the server due to the request. To do.
[0011]
Here, the server means a system or computer that provides a corresponding service in response to various requests. An example of the server is a web server. A request means a command or message for requesting a service from a server.
[0012]
According to the present invention, since the load is adjusted by adding a pseudo request to the original request transmitted to the server, the server performs processing for the original request, The load is adjusted by responding to a pseudo request while operating as in the case. At that time, by adjusting the amount of pseudo requests to be added, a constant amount of load without fluctuation can be given to the server. Therefore, it is possible to acquire data relating to the performance of the server while applying a certain load without stopping normal operation.
[0013]
In a preferred aspect of the present invention, while adding a pseudo request to the original request transmitted to the server, each request is transferred to the server at a timing such that it arrives with a constant arrival time interval distribution. By adding or subtracting pseudo requests, it is possible to maintain a certain arrival time interval distribution without delaying the arrival of the original request. According to this, since the arrival time interval distribution of requests is constant, a constant load can be given to the server. For example, a Poisson distribution can be used as the arrival time interval distribution.
[0014]
In this case, a pseudo request may be added for each request type, and the transfer to the server may be performed at a timing that has a unique arrival time interval distribution for each request type. According to this, it is possible to adjust the load for each type of request.
[0015]
In another aspect of the present invention, the requests transmitted to the server are temporarily stored in a queue, and if there are requests in the queue, the requests are generated, and if they do not exist, pseudo-generated requests are sequentially By transferring to the server at a predetermined timing, the load on the server due to the request is adjusted. Thereby, it is possible to adjust the server load by adding a pseudo request to the original request transmitted to the server with a simple configuration.
[0016]
In this case, a timing at which a request to be transferred to the server arrives at the server with a constant arrival time interval distribution can be adopted as the predetermined timing. Such timing can be determined based on random numbers generated according to a predetermined probability distribution, for example.
[0017]
In another aspect of the present invention, the requests transmitted to the server are distributed according to the type, temporarily stored in a plurality of request queues for each classification according to the type, and a separate request for each request queue. At each timing, for each request queue, if there is a request in the request queue, if the request does not exist, the corresponding type of request that is generated in a pseudo manner is sequentially transferred to the server. The load on the server is adjusted.
[0018]
In this case, when generating a pseudo-request, for a type of request that needs to include session information, generate the request as including the session information acquired by transferring the request to the server in advance. Can do. As a result, it is possible to individually adjust the load for a request that requires session information without any problem.
[0019]
Generation of a pseudo request including session information, for example, accumulates a thread that generates a corresponding type of request in each possible state for each state in a plurality of thread queues, and requests are received by each thread. This can be done by holding the session information acquired by transferring it to the server and transmitting a request including the session information.
[0020]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 is a block diagram showing a module configuration of a load adjustment filter according to an embodiment of the present invention. As shown in the figure, the load adjustment filter 1 is provided between the Web server 2 and the Internet 3. When the load adjustment filter 1 collects data related to the performance of the Web server 2, the load adjustment filter 1 adds a pseudo request to the request transmitted to the Web server 2 via the Internet 3. This is to adjust the load.
[0021]
The load adjustment filter 1 receives a request sent to the server 2 via the Internet 3 and temporarily stores a reception queue 4, a pseudo client 5 that generates a pseudo request, a random number generator 6, and a random number A gate 7 is provided that sends a request from the queue 4 or the pseudo client 5 to the Web server 2 at a timing according to a random number generated by the generator 6.
[0022]
The pseudo client 5 generates a request in response to a request from the gate 7. The random number generator 6 generates a random number according to a given probability distribution in response to a request from the gate 7. If a random number generator 6 using a sampling algorithm used in Monte Carlo simulation or the like is used, it is possible to generate random numbers according to almost arbitrary probability distribution. Each of the modules 4 to 7 can be configured as a combination of software and hardware.
[0023]
FIG. 2 shows a hardware configuration of the load adjustment filter 1. As shown in the figure, the hardware configuration includes a CPU 21 that performs data processing and control of each unit based on a program, a memory 22 that stores programs and data, an input means 23 that performs operation input, and data by the CPU 21. A display interface 24 that performs display based on the processing result and functions as a GUI (graphic user interface), a communication interface 25 for communication with the Web server 2 and communication via the Internet 3, A bus 26 and the like for connecting the elements are provided. The memory 22 includes a ROM, a RAM, a hard disk, and the like, and stores an OS and various application programs. The application program includes a program that allows the hardware configuration to function as the load adjustment filter 1.
[0024]
In order to evaluate the performance of the system model in the Web server 2, as shown in FIG. 6, a target system model is created (step 61), the system performance is measured (step 62), and the measurement data is used. The parameters in the model are estimated (step 63), and the completed model is analyzed by a queue or simulator (step 64). The load adjustment filter 1 determines the parameters of the system model in this procedure. In performing system performance measurement (step 62), an ideal load is applied to the system model.
[0025]
The ideal load at the time of system performance measurement is a steady load, that is, a load with a constant distribution of the number of request arrivals per unit time. In order for the load to be constant, it is only necessary that the time distribution of request arrival intervals is constant. For example, Poisson arrival often assumed in queuing analysis can be obtained by making the arrival time interval distribution an exponential distribution. Poisson arrival is the simplest to analyze, but it is often possible to analyze or artificially reproduce other distributions if they are stationary. For this reason, in this embodiment, the specific distribution shape is not limited as long as the condition that the request arrival time interval distribution is constant is satisfied.
[0026]
FIG. 3 is a flowchart showing a procedure of load adjustment processing by the gate 7. In the random number generator 6, parameters are set in advance so that the arrival distribution of requests arriving at the Web server 2 is a constant desired distribution by the following processing. When the load adjustment process is started, the gate 7 first acquires a random number from the random number generator 6 in step 31. Next, in step 32, a time corresponding to the acquired random number is waited.
[0027]
Thereafter, in step 33, the queue 4 is examined to determine whether a request exists. If it is determined that the request exists, the request is transmitted to the Web server 2 in step 34, and the process proceeds to step 37. If it is determined that the request does not exist, the request is acquired from the pseudo client 5 in step 35, the acquired request is transmitted to the web server 2 in step 36, and the process proceeds to step 37.
[0028]
In step 37, it is determined whether or not to end the load adjustment process based on the end of the load adjustment period or the load test period for the Web server 2. If it is determined not to end, the process returns to step 31 and the above processing is repeated. If it is determined to end, the load adjustment process ends.
[0029]
In this way, the gate 7 requests a random number from the random number generator 4, and after a time corresponding to the random number has elapsed, if there is a request sent via the Internet 3 in the queue 4, the request is sent to the Web server 2. If not transferred, the process of transferring the request obtained by requesting the pseudo client 5 to the Web server 2 is repeated.
[0030]
According to this, when there is no request from the Internet 3, since the request from the pseudo client 5 is supplied instead, the request can always be transmitted to the Web server 2 when the gate 7 opens. Therefore, it is possible to give the Web server 2 a load according to the arrival time interval distribution corresponding to the random number based on the probability distribution given to the random number generator 4.
[0031]
Since the arrival rate of requests transmitted via the Internet 3 fluctuates with time, the load when a request is given directly to the normally operating Web server 2 without going through the load adjustment filter 1 Can only be considered stationary for a very short time. On the other hand, according to the present embodiment, the load adjustment filter 1 adjusts the arrival time interval distribution of requests arriving at the Web server 2 to be constant, and thus continues during the period of performing the load adjustment process. Therefore, a constant load can be given to the Web server 2.
[0032]
At that time, since all requests from the Internet 3 are accepted without being rejected, a load with an arrival rate lower than the average arrival rate cannot be given, but the load at the maximum arrival rate during the period of load adjustment processing If it exceeds the amount, an arbitrary load amount can be set by adjusting a parameter of the random number generator 6. As a matter of fact, Web sites that are constantly subjected to high loads are rare, and the number of requests decreases considerably during a certain day of the day, such as at dawn, and the average resource usage rate is often several percent. If load adjustment processing is performed during such a time period, the degree of freedom of load that can be set is large.
[0033]
Therefore, according to the present embodiment, it is possible to perform a load test for collecting performance data related to the system model of the Web server 2 in a state where a desired steady load is applied without interrupting processing for a request from the Internet 3. it can.
[0034]
FIG. 4 shows another embodiment of the present invention. In general, there are multiple types of requests, so to estimate the amount of hardware resources used by each type of request, you can change the average arrival rate of various requests and the mix ratio of requests multiple times. It is necessary to measure and acquire performance data. Actually, when collecting performance data by a conventional normal load test, several types of performance data are collected while changing the mixing ratio of loads according to various requests. Therefore, in the present embodiment, as shown in the figure, load adjustment filters 41a to 41d are provided for each of various requests so that the arrival time intervals of the various requests can be adjusted independently.
[0035]
That is, requests from the Internet are distributed to the load adjustment filters 41a to 41d corresponding to the various requests by the network dispatcher 42, and transferred to the Web server 40 at the timing of each load adjustment filter 41a to 41d. ing. Each load adjustment filter 41a to 41d shares one pseudo client 43 extended so that various requests can be supplied, and can acquire a corresponding type of request as needed. Thereby, the mixing ratio of various requests can be adjusted.
[0036]
FIG. 5 shows the configuration of the load adjustment filters 41a to 41d. Each load adjustment filter 41a to 41d receives a request corresponding to itself among requests sent to the Web server 40 via the Internet via the dispatcher 42, and temporarily stores the reception queue 51, A random number generator 52 set to generate a random number with a probability distribution corresponding to the type of request stored in the queue 51, and a request from the queue 51 or the pseudo client 43 at a timing according to the random number generated by the random number generator 52 Is sent to the Web server 40. The queue 51, the random number generator 52, and the gate 53 have the same configuration as the queue 4, the random number generator 6, and the gate 7 in the load adjustment filter 1 of FIG.
[0037]
The gates 53 of the load adjustment filters 41a to 41d share one pseudo client 43 that is expanded so that various requests can be supplied. This is because session information can be given to the request transferred to 40. In other words, depending on the application program that constitutes the system model in the Web server 40, there is one that holds the session state on the Web server 40 side. In this case, the request is normal only when it includes appropriate session information. To be processed. In other words, there is no point in sending an inappropriate request. As a simple example, even if a logout request is sent without the session information that should be returned at the time of login, the processing in the Web server 40 for the request results in an error, and the normal processing is Not done. Therefore, the pseudo client 43 is employed so that appropriate session information can be given to each request.
[0038]
The pseudo client 43 defines a state transition according to a state transition graph that transitions between states corresponding to the state transition on the Web server 40 side, and includes a queue corresponding to each state. In each queue, a thread for transmitting a request including session information is stored in advance. When each type of request is requested, the pseudo client 43 takes out the thread that transmits the request of the type from the queue, transmits the request by the thread, and puts the thread in the queue of the next state according to the state transition graph. Move.
[0039]
As a specific example, consider a case where the system model in the Web server 40 is configured by a Web application of a simple shopping site. As shown in FIG. 4, the pseudo client 43 indicates, as a state transition graph, between each state of the state 44 before login, the state 45 during login, and the state 46 after logout corresponding to the state transition on the Web server 40 side. Are defined, and queues 47 to 49 corresponding to the states 44 to 46 are provided. The type of request is A. Login, B. Catalog display, C.I. Settlement, D. There are four types of logout. The request that can be sent in the pre-login state 44 is login (A), and the request that can be sent in the logged-in state 45 is catalog display (B), settlement (C), or logout (D). There is no request that can be sent in the state 46 after logout.
[0040]
Each arrow in the transition state graph extends from the pre-transition state to the post-transition state, and each arrow is accompanied by a symbol A to D indicating the type of request sent by the thread during the state transition indicated by the arrow. ing. That is, when a thread in the pre-login state 44 issues a login request A, the thread is moved to the queue 48. When a thread in the logged-in state 45 issues a catalog display request B or a settlement request C, the thread is returned to the queue 48. When a thread in the logged-in state 45 issues a logout request D, the thread is moved to the queue 49.
[0041]
In the queues 47 and 48, a number of threads that can respond to request requests during the load adjustment period are prepared in advance. In other words, the pseudo client 43 prepares a predetermined number of threads in the queue 47, transmits a login request A to the Web server 40 using an appropriate number of threads, acquires session information from the Web server 40, and acquires each thread. Move to queue 48. As a result, an appropriate number of threads in the pre-login state are stored in the queue 47, and threads having an appropriate number of session information in the login state are stored in the queue 48.
[0042]
When the login request A is requested, the pseudo client 43 takes out the thread from the queue 47 in the pre-login state, transmits the login request A by the thread, and moves the red to the queue 48 in the login state. . When the catalog display request B or the settlement request C is requested, the thread is taken out from the queue 48 in the login state, the catalog display request B or the settlement request C including the session information is transmitted by the thread, and the thread is queued. Return to 48. When the logout request D is requested, the thread is taken out from the queue 48 in the login state, the logout request D including the session information is transmitted by the thread, and the thread is moved to the queue 49. The thread moved to the queue 49 may be deleted or returned to the queue 47 in the pre-login state.
[0043]
When a more complicated graph is used as the state transition graph, there may be a plurality of thread states that can transmit a request for a request. In that case, a state in which the request is transmitted at random may be selected from the plurality of states.
[0044]
According to the pseudo client 43, a sufficient number of threads are prepared in the queues 47 and 48 together with necessary session information in advance, so that requests for each type of request and requests for which session information is required are provided. But you can respond. Therefore, it is possible to apply a steady load different for each type of request to the system of the Web server 40 holding the session state while maintaining the normal operation state.
[0045]
The number of threads stored in the queues 47 and 48 in advance is set in each load adjustment filter 41a to 41d for the request A and the requests B, C, and D that can be transmitted in the corresponding pre-login state and in-login state. It can be determined in consideration of the distribution of arrival time intervals. In this case, for example, the number of threads can be estimated so that the probability that the threads in the queues 47 and 48 are not exhausted during the load adjustment period is 90% or more. In that case, as a rare case, if a thread disappears from the queue, the transmission of the corresponding request may be ignored. Performance data collected by load tests is statistically processed when estimating system model parameters based on this, so in rare cases, it arrives at a time interval that deviates from the arrival time interval distribution of the set request. Even if a request occurs, it is not a big problem.
[0046]
When performing load adjustment in order to collect performance data on the system model in the Web server 40, each load is previously set so as to generate a random number corresponding to a desired arrival time interval distribution for each of the requests A to D. Parameters of the random number generator 52 in the adjustment filters 41a to 41d are set. When performing the load adjustment, the gate 53 of each of the load adjustment filters 41a to 41d performs the same process as in FIG. That is, after acquiring a random number from the random number generator 52 and waiting for a time corresponding to the random number, if the request exists in the queue 51, the request is transferred to the Web server 40, and if not, the pseudo client 43 responds. The process of acquiring any of the types of requests A to D to be transferred and transferring them to the Web server 40 is repeated.
[0047]
As a result, the gates 53 of the load adjustment filters 41a to 41d add the requests A to D from the pseudo client to the requests A to D from the Internet, and the requests 53 to 41 correspond to the setting of the probability distribution in each random number generator 52. The arrival time interval distribution can be supplied to the Web server 40. Therefore, the load amount differs for each request type, and a steady load can be applied to the Web server 40.
[0048]
Therefore, according to the present embodiment, in the case of a service that improves the performance of the system model in the customer's Web server 40, a steady load can be obtained without stopping the operating system for a long time and causing a loss of transaction opportunities. The performance data can be collected and the parameters of the system model can be estimated. As a result, the threshold for performance services for customers who are afraid of losing business opportunities is lowered, which can greatly contribute to the development of the service business.
[0049]
Also, when autonomous computing technology is available, the load adjustment filter according to the present invention promotes such introduction because the cost of introducing the autonomous adjustment function based on the system model into the operating system is small. can do.
[0050]
Note that the present invention is not limited to the above-described embodiment, and can be appropriately modified and implemented. For example, in the above description, requests sent to the server are temporarily stored in a queue, and the load is adjusted by mixing pseudo requests when transferring the request. Instead, the request is sent to the server. Monitor incoming requests without accumulating them in the queue, and send pseudo requests directly to the server according to the monitoring results, adding pseudo requests to the original requests and adjusting the load. You may do it.
[0051]
With regard to the load adjustment filter according to each of the above-described embodiments, a particularly assumed application scene is that a web application is analyzed by analyzing the performance of a web server composed of a websphere (WebSphere) product group provided by the applicant. In addition, there are scenes of services that find server software parameters (number of threads, cache size, etc.) that are optimal for the expected request volume, and automatic control methods that dynamically adjust server software parameters. . The parameter of the system model in this case is a consumption amount of hardware resources such as a CPU, a disk I / O, and a network I / O consumed by a transaction corresponding to each request. In a typical three-tier Web server, the hardware resource consumption of transactions in the HTTP server, application server, and database tiers are individually estimated to complete the model.
[0052]
【The invention's effect】
As described above, according to the present invention, since the load applied to the server is adjusted by adding a pseudo request to the request transmitted to the server, the service provision by the server is stopped. The adjusted load can be applied to the server without causing it to occur.
[0053]
In addition, since a request in which a pseudo request is added to a request transmitted to a server is transferred to the server at a timing that arrives at a predetermined arrival time interval distribution, the load is made steady. Can do.
[0054]
In addition, for requests sent to the server, for each type of request, a pseudo request is added and transferred at a timing that arrives with its own predetermined arrival time interval distribution. The load can be made steady for each type.
[0055]
In addition, for the types of pseudo requests that need to include session information, the request is generated as including the session information acquired by transferring the request to the server in advance, so the session state is maintained. The load can be adjusted without trouble for the servers that are running.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a module configuration of a load adjustment filter according to an embodiment of the present invention.
FIG. 2 is a block diagram showing a hardware configuration of the load adjustment filter of FIG. 1;
FIG. 3 is a flowchart showing a load adjustment processing procedure by a gate of the load adjustment filter of FIG. 1;
FIG. 4 is a block diagram showing another embodiment of the present invention.
FIG. 5 is a block diagram showing a configuration of each load adjustment filter in the embodiment of FIG. 4;
FIG. 6 is a diagram illustrating a procedure for performing performance evaluation based on a system model.
[Explanation of symbols]
1: load adjustment filter, 2: Web server: Internet, 4: reception queue, 5: pseudo client, 6: random number generator, 7: gate, 21: CPU, 22: memory, 23: input means, 24: display, 25: Communication interface, 26: Bus, 40: Web server, 41a to 41d: Load adjustment filter, 42: Dispatcher, 43: Pseudo client, 44: Pre-login state, 45: Login state, 46: Post logout state, 47 ~ 49: thread queue, 51: request queue, 52: random number generator, 53: gate.

Claims (14)

A request load adjustment device for adjusting the load on the server by requests,
To request transmitted before Symbol server, a pseudo-request comprises a load adjusting means for adjusting the load by inter-arrival times of requests arriving into the server applies to a predetermined distribution ,
The load adjusting means is
A queue for temporarily storing requests sent to the server;
A request generating means for generating a pseudo request;
A transfer means for sequentially transferring the request generated by the request generation means when there is a request in the queue to the server at a timing when the arrival time interval is the predetermined distribution; request load adjustment device characterized by comprising a. The load adjusting means, for each type of request, to add the pseudo-request, according to claim 1, characterized in that for performing the transfer timing to be different from the arrival time distribution Request load adjustment device. 2. Random number generation means for generating random numbers according to a predetermined probability distribution, wherein the transfer means determines a timing at which the arrival time interval becomes a predetermined distribution based on the random numbers. Item 4. The request load adjustment device according to Item 1 . A request load adjusting device that adjusts a load on a server due to a request,
Load adjustment means for adjusting the load by adding a pseudo request to the request transmitted to the server so that the arrival time intervals of the requests arriving at the server have a predetermined distribution,
The load adjusting means is
A plurality of request queues for temporarily storing requests sent to the server for each classification according to the type;
A dispatcher that distributes requests sent to the server to corresponding request queues according to the type;
A request generating means for artificially generating the requested type of request;
And a transfer means provided corresponding to each request queue,
Each transfer means requests the request generation means if the request exists in the corresponding request queue when the unique arrival time interval becomes the predetermined distribution, and if the request does not exist. the type of request corresponding to the request queue to generate, features and be Brighter Quest load adjusting device that is for sequentially transferred to the server. The request generation means generates a request for a type of request that needs to include session information, including the session information acquired by transferring the request to the server in advance. The request load adjusting device according to claim 4 . The request generation means includes a plurality of thread queues that accumulate, for each state, a thread that generates a corresponding type of request in each possible state, and each thread is acquired by transferring a request to the server. 5. The request load adjusting apparatus according to claim 4 , wherein the request load adjusting apparatus is capable of holding session information and transmitting a request including the session information. A program for causing a computer to function as each means in the request load adjusting device according to any one of claims 1 to 6 . A request load adjusting method for adjusting the load on the server by requests by a computer,
The computer, to request transmitted before Symbol server, load the pseudo-request, arrival time of requests arriving into the server to adjust the load by adding to a predetermined distribution With adjustment procedures ,
The load adjustment procedure includes:
A procedure in which the computer temporarily stores a request sent to the server in a queue;
A request generation procedure in which the computer generates a pseudo-request;
If the computer has a request in the queue, the computer sequentially forwards the request generated by the request generation procedure to the server at a timing when the arrival time interval becomes the predetermined distribution. A request load adjusting method , comprising: In the load adjustment procedure, the computer, for each type of request, to add the pseudo-request, according to claim 8, characterized in that said transfer timing as the different the inter-arrival time distribution Request load adjustment method. The computer includes a random number generation procedure for generating random numbers according to a predetermined probability distribution, and in the transfer procedure, the computer determines a timing at which the arrival time interval becomes a predetermined distribution based on the random number. The request load adjusting method according to claim 8 , wherein the request load is adjusted. A request load adjustment method for adjusting a load on a server by a request by a computer,
Load adjustment for adjusting the load by adding a pseudo request to a request sent to the server so that arrival time intervals of requests arriving at the server have a predetermined distribution. With steps,
The load adjustment procedure includes:
A procedure in which the computer temporarily stores requests sent to the server in a plurality of request queues for each classification according to the type;
The computer distributes the requests sent to the server to each request queue according to the type;
A request generation procedure in which the computer artificially generates a request of a requested type;
When there is a request in the request queue for each request queue at the timing when the arrival time interval different for each request queue becomes the predetermined distribution, and the computer does not exist features and to Brighter Quest load adjusting method to have a procedure for transferring the corresponding type of the request is generated by the request generation procedure, sequentially to the server. In the request generation procedure, for the type of request that needs to include session information, the computer generates the request as including the session information acquired by transferring the request to the server in advance. The request load adjusting method according to claim 11 , wherein the request load is adjusted. The request generation procedure includes a procedure in which a thread that generates a request of a corresponding type in each possible state of the computer is accumulated for each state in a plurality of thread queues. 12. The request load adjusting method according to claim 11 , wherein the computer holds session information acquired by transferring a request to the server by each thread, and transmits a request including the session information. Collecting data relating to the performance of the server while adjusting the load on the server due to the request by the request load adjusting device according to any one of claims 1 to 6 or the request load adjusting method according to any one of claims 8 to 13. A performance data collection method characterized by

Images