JP2002123403A - Program storage medium recording program for improving response time of computer system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for determining a system resource parameter to be changed to a specified user or group of users for improving system performance defined by response time. SOLUTION: In this method, a system resource to control a parameter is specified and the response time of services by means of these resources is improved. In a representative execution style, the measured results of the response time concerning the services provided by various system resources are totaled concerning the specified user or group of users. The average of this total indicates which resource consumes excessive time for a process to be used by the user or group of users. Then, the value of the parameter related to the system resource is usefully changed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

FIELD OF THE INVENTION The present invention relates generally to computer systems, and more particularly, to adjusting parameters, as reflected in service levels provided to various users, to values that lead to improved system performance. .

[0002]

2. Description of the Related Art Modern computing systems generally have a large number of processes running on computer systems simultaneously, often belonging to different users. In order to improve system performance, it is often desirable to provide different levels of service for different processes. Administrators of modern computing systems generally have access to a variety of parameters that can be adjusted and set to improve system performance. Examples of such adjustable parameters include swap space size, memory dedicated to buffer cache, maximum allowed number of processes, and maximum application data size.

One method of determining system parameters that can be used to manipulate service levels is through guesswork and experimentation based on operator experience. Disadvantages of this method include: (1) time intensive for the computer operator; (2) its effectiveness is limited by the capabilities of the operator; (3) its effectiveness under changing system loads and environments. The point that must be done continuously to keep

[0004] Another more general approach is to change only a very limited number of parameters, usually those to which the operator is most familiar. This method has the disadvantage that it is often not as effective as one might think.

[0005]

Therefore, there is a need for a method that can improve a computer system more effectively and efficiently. In particular, there is a need for a method of automatically and continuously adjusting system parameters to improve system performance, even when the workload on the system changes.

[0006]

SUMMARY OF THE INVENTION The present invention provides a computer system which includes adjusting a computer system parameter.
A new method for improving system performance. According to this method, appropriate adjustment can be performed even when the workload of the system changes. Conventional methods of improving system performance have relied on estimating the optimal system configuration and controlling a limited number of system parameters.

[0007] Modern computer systems generally serve many users. These users each have one or more processes running on the computer at different times. Each process requires a variety of system resources, often including the same resources required by another process. In the very limited case where only one process requests service by system resources, the time delay to service a request by a process is the time required for the resource to fulfill the request. In a multi-user computer system, the time is further increased, causing a time delay because the process must wait until resources are available to the process. The sum of these two times is the total time that the process experiences when requesting service from the resource. For a single user with one running process, this sum is also referred to as the response time (ie, the time the user receives the service he requested from the process). If the user has more than one process, the sum for all processes started by the user is needed.

When performing different processes on the same system with multiple workloads of the same or different users, or trying to provide different service levels to different users with the same workload. , (A) group processes according to the user or the workload they are performing, and (b) assign resources to separate groups of processes using the controls (knobs) provided by the system, Needs the ability to control the level of service that end users receive.
Various controls are available in modern computing systems, but it is difficult for system operators to determine how to set or change "knob" settings to control the level of service experienced by the end user There are many. The exemplary embodiments disclosed herein monitor which systems are running and which "knobs" to manipulate service levels for each workload.
Informs the operator if the can be used most effectively.

In order to avoid having to double calculate the latency during the time the user has more than one process waiting for service by the resource, the method calculates that part of the latency only once. . The sum assumes the case of a single microprocessor. With multiple microprocessors, as well as processing time, some of the latencies occur in parallel, and the response time at various levels is less than or equal to the time indicated by the equation.

In an exemplary embodiment, the total time delay experienced by all processes for the work done for a particular user during a request for service is aggregated. This time delay is further aggregated and averaged for users requesting multiple services. This aggregation can also be done for a group of related users.

The decision to change or not change the system parameter value for the selected system resource is made based on the resulting delay time results.
Those system resources with longer delays are easily identified and appropriate changes are made to improve the response time for the user or user group in question.

[0012] As shown herein, the exemplary embodiment is advantageous in that only a small amount of time or work is required on the part of the system operator to achieve an improved system configuration. In addition, as usage characteristics and loads in the system change, the method discovers new improvements to the system without requiring the operator to continuously monitor and redo previous adjustments made manually. Recommend. Another advantage is that this mechanism can be used for a subset of the parameters available in the system. Thus, the operator may choose to use this method to search for obscured or poorly understood parameter improvements, or may choose to make macro adjustments himself.

The operator has a great deal of visibility on how to best manipulate the service level experienced by a user or group of users. This method is time efficient, and the actions performed by the operator can be effective for improving response time.

[0014] Other aspects and advantages of the present invention will become apparent from the following description, taken in conjunction with the accompanying drawings, illustrating by way of example the principles of the invention.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION Introduction As shown in the drawings for illustration, the present invention relates to a new method of improving the performance of a computer system by adjusting computer system parameters. This method can appropriately adjust even when the workload of the system changes. Previous methods of improving system performance have relied on estimating the optimal system configuration and controlling a limited number of system parameters. In the following detailed description and drawings, identical elements are identified by identical reference numerals.

When performing different processes on the same system with multiple workloads of the same or different users, or providing different service levels to different users with the same workload. , (A) group processes according to the user or the workload they are performing, and (b) assign resources to separate groups of processes using the controls (knobs) provided by the system, Needs the ability to control the level of service that end users receive. Various controls are available in modern computing systems, but it is difficult for system operators to determine how to set or change "knob" settings to control the level of service experienced by the end user There are many. The exemplary embodiments disclosed herein monitor the running system and inform the operator which "knobs" can be most effectively used to operate the service level of each workload. I do.

2. Computer System FIG. 1 is a diagram illustrating an equation for a time delay 115 experienced by a process requesting service by a resource as described in various representative embodiments herein. Modern computer systems generally serve a large number of users, each of which has one or more processes running on the computer at different times. Each process requires a variety of system resources, often including the same resources required by another process.
In the very limited case where only one process requests service by a system resource, the time delay 115 for servicing the request by the process is the time 105 that the resource needs to fulfill the request.
It is. In a multi-user computer system, a time delay 110 results from the waiting of a process until resources are available to the process. These two
The two times add up to the total time 115 that the process experiences when requesting service from the resource. For a single user with one running process, this sum is also referred to as response time 115 (ie, the time the user receives the service he requested from the process). If the user has more than one process, the sum for all processes started by the user is needed.

FIG. 2 is a diagram illustrating the interrelationship of users, processes, and resources described in various representative embodiments of the present invention. The computer system is composed of a number of resources 205, of which k-1, k, k are shown in FIG.
Only three of +1 are shown. Other resources are represented as triple dots above and below the three resources shown. Services may be requested by various processes 210 operable on a computer system. The computer system further includes a number of processes 210, while other processes 210 are represented as triple dots above and below the three processes shown. Various processes 210 are illustrated in FIG.
It is activated at different times by a second user, a user 215 of the computer system identified as "user n". In general, the computer system is
Includes a number of users 215 not shown. Although FIG. 2 illustrates that any process 210 can access any resource 205, those skilled in the art will recognize that some resources 205 may not be accessible by any particular process 210. Similarly, FIG. 2 appears to imply that any user 215 can launch any process 210. However, those skilled in the art will recognize that some processes 210 may not be accessible by a particular user 215.

3. Definitions and Formulas The following definitions and formulas describe the time delay experienced by various users 215 of the computer system. Service level for a user making a given request = response time experienced for that request

The following symbols are used as indices and represent integers identifying the designated item. n = user n (the nth user of the computer system) i = identification number associated with the ith service request by a given user j = identification number associated with the _jth process _Pj k = kth identification number T _a associated with the resource _{R k (n, i, P} j, R k) between = by user N215 i-th service request, k th resource R _k is for the j-th process P _j Time spent in active work T _W (n, i, P _j , R _k ) = j-th process P _j spends waiting for k-th resource R _k during i-th service request by user n 215 time

During an i-th request by an n-th user 215, a single process 210, a j-th process 210, k initiated by an n-th user 215 requesting service from a single resource 205 For the second resource 205, the response time or time delay experienced is calculated by the following equation:

[0022]

(Equation 1)

During the i-th service request by user n 215, the total response time experienced by user n 215 for all processes P _j 210 using resource R _k 205 is calculated as follows:

[0024]

(Equation 2)

To avoid counting the latency twice, while user n 215 has more than one process 210 waiting for service by resource R _k ,

(Equation 3) Counts that portion of the wait only once.

For example, a user n21 who has a total of I requests
During multiple service requests according to 5, the average response time experienced by user n 215 for all processes P _j 210 using resource R _k 205 is calculated by the following formula:

[0027]

(Equation 4)

During the ith service request by user n 215, all processes P using all resources R _k 205
_The total response time experienced by user n for _j 210 is calculated by the following equation:

[0029]

(Equation 5)

For a total of I requests by user n 215, the average response time for user n is calculated by the following equation:

[0031]

(Equation 6)

For all N users 215, the average response time for each user is calculated by the following equation.

[0033]

(Equation 7)

The above sum assumes a single microprocessor. If there are multiple microprocessors, as well as the processing time, some of the latencies will occur in parallel, and the response times at various levels will be less than or equal to the above sum.

4. Exemplary Embodiment FIG. 3 is a flowchart of a process for accumulating data for the delay calculation method described in various exemplary embodiments of the present invention. At block 305, user 215
(Eg, user n 215) requests service by process 210 (eg, process j 210). Then block 305
Transfers control to block 310.

At block 310, process j210 requests service of system resources 205 (eg, resource k205). Subsequently, block 310 transfers control to block 315.

At block 315, resource k205 records the time at which process j210 requested its service. This time is referred to herein as the first time. Subsequently, block 315 transfers control to block 335.

At block 320, another user 215 (eg, user n-1 215) requests service by the same process j210. Block 320 then controls block 325
Move to

At block 325, process j-1 210 requests service for the same system resource k205 (eg, resource k205). Subsequently, block 325 transfers control to block 330.

At block 330, resource k205 records the time at which process j-1 210 requested its service. Subsequently, block 330 transfers control to block 335.

At block 335, resource k205 selects the next service request from the two competing service requests. For example, a request from the process j210 is selected. Subsequently, block 335 transfers control to block 340.

At block 340, resource k205 records salient system resource control parameter values for resource k205. Subsequently, block 340 transfers control to block 345.

At block 345, resource k records the time at which system resource k 205 begins to service process j 210. This time is referred to herein as the second time. Block 345 then controls block 350
Move to

At block 350, the system resource k205
Performs the service required by process j210. Subsequently, block 350 transfers control to block 355.

At block 355, resource k records the time at which system resource k 205 ends service to process j 210. This time is referred to herein as the third time. Subsequently, block 355 transfers control to block 360.

At block 360, resource k is assigned to process j2
Return to the control of 10.

As previously defined, the time that process j 210 spends waiting for service by resource k 205 during the i-th service request by user n 215, T _W (n,
i, P _j , R _k ) is _obtained as the clock time recorded at block 345 (the time at which the service was started) minus the clock time recorded at block 315 (the time at which the service was requested). As previously defined, T _A (n, i, P _j , R _k ), the time that resource k 205 spends on active work on process j 210, is the clock time recorded in block 355 (the time when the service ended). ) Minus the clock time recorded at block 345 (the time the service started). During the i-th request by user n215, for process j210 initiated by user n215 and requesting service from resource k205, the response time or time delay experienced is T _S as calculated in equation (1). (n, i, P _j , R _k ).

In the exemplary embodiment, the time delay experienced by all processes for the work performed for a particular user 215, eg, user n 215, during the ith service request is given by equation (2). And for a plurality of service requests, further aggregated and averaged for user 215, eg, user n 215, and aggregated as shown by equation (3). This aggregation operation can also be performed for a group of related users 215.

The system resources 205
The decision whether to change the parameter value or not is generally made based on the result of equation (3). Those system resources with longer delays are easily identified and appropriate changes are made to improve the response time for the intended user or user group. Note that other aggregation and averaging methods are possible, as specified in equations (4)-(6).

5. Conclusion As described above, the exemplary embodiment is advantageous in that only a small amount of time or work is required on the part of the system operator to achieve an improved system configuration. In addition, as usage characteristics and loads in the system change, the method discovers new improvements in the system without requiring the operator to continuously monitor and redo previous manual adjustments. Recommend. Another advantage is that this mechanism can be used for a subset of the parameters available in the system. Thus, the operator may choose to use this method to search for obscured or poorly understood parameter improvements, or may choose to make macro adjustments himself.

The operator has a great deal of visibility on how to best manipulate the service level experienced by the user or group of users. This method is time efficient, and the actions performed by the operator can be effective for improving response time.

Although the present invention has been described in detail with reference to particular embodiments, the described embodiments are by way of example only.
It is not intended to be limiting. Those skilled in the art will appreciate that various modifications can be made to the described embodiments to produce equivalent embodiments that do not depart from the scope of the invention.

The present invention includes the following embodiments as examples. (1) A computer readable program storage medium storing a program for executing a method step for improving computer system performance, the method step comprising:
(315) recording a first time, wherein 0) is a computer clock time requesting service from a computer system resource (205); and at least one preselected computer system resource parameter. (340) recording the value of (220);
The resource (205) is the computer process (210)
Recording a second time that is a computer clock time to start service to the server (345); and recording a third time that is a computer clock time for the resource (205) to complete the requested service. Step to do
(355) and the second time (345) to the first time (31
Subtracting 5) and assigning the result to the waiting time (110); and
5) subtracting the result and assigning the result to a processing time (105); adding the waiting time (110) and the processing time (105); and executing the waiting time (110) and the processing time (10).
5) assigning the result of the step to a response time (115); and, when the user service time determined by the response time (115) exceeds a preselected value, the value of the recorded parameter (220). ) Changing one of the above.

(2) The program storage medium according to (1), wherein the user service time is equal to the response time (115).

(3) The method step comprises the steps of:
5) for at least one additional process of requesting service, excluding the step of changing one of the recorded parameter values when the user service time exceeds a preselected value. Performing the steps described in (1) and two or more processes (21
(0) when waiting for service, the waiting time (110)
Is included in only one of the response times (115), the response time (11) determined for a plurality of processes (210).
The program storage medium of (1), further comprising: summing 5); and setting the user service time equal to the result of summing the response times.

(4) The method steps include at least one
For two additional user resource service requests,
And when at least one additional process of requesting resource (205) service for at least one of the user resource service requests, the user service time exceeds a preselected value. Performing the steps of (1) above, excluding the step of changing one of the parameter values;
When two or more processes (210) are waiting for service, the waiting time (110) is determined for a plurality of processes (210) that are included in only one of the response times (115). Summing the response time (115) and the result of the step of summing the response time (115).
(1) further comprising: equally dividing by the number of requests initiated by: and setting the user service time equal to the result of equally dividing the result of the step of summing the response times. A program storage medium according to claim 1.

(5) The method step comprises the steps of:
5) for at least one additional process requesting service, and for at least one additional resource (205) for which at least one of the processes (210) requires a resource (205) service, Performing the steps of (1) above, except for changing one of the values of the recorded parameters when the user service time exceeds a preselected value; and two or more processes. When the waiting time (110) is waiting for the service, the waiting time (110) is included in only one of the response times (115). 115) further comprising: summing the response times; and setting the user service time equal to the result of the summing the response times.
A program storage medium according to claim 1.

(6) The method includes at least one step.
For two additional user resource service requests,
Process for at least one additional process (210) requesting resource (205) service for at least one of the user resource service requests and for at least one of the user resource service requests At least one additional resource (20) for which at least one of (210) requests a resource (205) service;
Performing the steps of (1) above, except for 5) changing the one of the recorded parameter values when the user service time exceeds a preselected value. And two or more processes (21
(0) when waiting for service, the waiting time (110)
Is included in only one of the response times (115), the response time (11) determined for a plurality of processes (210).
5) summing the result of the step of summing the response time (115) by the number of requests by the user (215), and equally dividing the result of the step of summing the response time. Setting the user service time equal to the result of (1).
A program storage medium according to claim 1.

(7) The method includes at least one step.
For one additional user (215), for at least one additional user resource service request by at least one of the users (215), a user resource resource for at least one of the users (215). For at least one additional process (210) requesting resource services for at least one of the service requests, and for at least one of the user resource service requests by at least one of the users (215). The parameters recorded when the user service time exceeds a preselected value for at least one additional resource (205) for which at least one of the processes (210) requires a resource service. Performing the steps of (1) above, excluding the step of changing one of the values; When the above process (210) is waiting for service, the waiting time (110) is included in only one of the response times (115), Summing the response times (115) and at least one additional user (21
5) performing the above steps and summing the response time (115).
(1) further comprising: equally dividing by the number of requests according to 5); and setting the user service time equal to the result of equally dividing the result of the step of summing the response times. The program storage medium according to any one of the preceding claims.

(8) recording (315) a first time which is the computer clock time at which the computer process (210) of the user (215) requests service from the computer system resources (205). , At least one preselected computer system
Recording the value of the resource parameter (220) (34
0) and recording (345) a second time that is the computer clock time at which the resource (205) starts servicing the computer process (210);
Recording (355) a third time that is the computer clock time at which the resource (205) completes the requested service; and calculating the first time (315) from the second time (345). Subtracting and assigning the result to the waiting time (110); subtracting the second time (345) from the third time (355) and assigning the result to the processing time (105); The waiting time (110) and the processing time (10
5) and the response time (115) of the result of the step of adding the waiting time (110) and the processing time (105).
And changing one of the recorded parameter values (220) when the user service time determined by the response time (115) exceeds a preselected value. Computer·
A computer-implemented method for improving system performance.

(9) The method according to (8), wherein the user service time is equal to the response time.

(10) For at least one additional process (210) requesting resources (205) services:
Performing the steps of (8) except for changing one of the recorded parameter values (220) when the user service time exceeds a preselected value; When more than one process (210) is waiting for service, the waiting time (110) is determined for a plurality of processes (210) that are included in only one of the response times (115). The method of claim 8, further comprising: summing the response times (115); and setting the user service time equal to the result of the summing the response times.

[0063]

According to the present invention, during a request for a service,
The time delay experienced by all processes for the work done for a particular user is aggregated, and the decision whether to change or not change the system parameter value for the selected system resource is determined by the resulting delay time. Since it is based on the results, little time or effort is required on the part of the system operator to achieve an improved system configuration.

[Brief description of the drawings]

FIG. 1 illustrates equations for the time delay experienced by a process requesting service by a resource, as shown in various exemplary embodiments of the invention.

FIG. 2 illustrates the interaction of users, processes, and resources as shown in various representative embodiments of the invention.

FIG. 3 is a flow chart of a process for accumulating data for a delay calculation method as shown in various exemplary embodiments of the present invention.

[Explanation of symbols]

 105 Processing time 110 Latency 115 Time delay, response time 205 Resources 210 Processes 215 Users 220 System resource parameters

 ──────────────────────────────────────────────────続 き Continued on front page (72) Inventor Doug Grumman Dublin Way 6019, Citrus Heights, CA 95610, USA F-term (reference) 5B042 JJ23 LA08 MA08 MC33 MC35 5B098 GA04 GC10 JJ08

Claims (1)

[Claims] 1. A computer readable program storage medium having recorded thereon a program for performing a method step for improving computer system performance, comprising:
The method steps include: recording a first time that is a computer clock time at which a user's computer process requests service from a computer system resource; and at least one preselected computer system resource. Recording the value of the parameter; recording a second time which is the computer clock time at which the resource starts servicing the computer process; and the computer at which the resource completes the requested service. A step of recording a third time that is a clock time; a step of subtracting the first time from the second time and assigning the result to a waiting time; and a step of calculating the second time from the third time. Subtract time and assign result to processing time Adding the waiting time and the processing time; assigning the result of the step of adding the waiting time and the processing time to a response time; and selecting a user service time determined by the response time in advance. Changing one of the values of the recorded parameters when the value exceeds the set value.