JP4453823B2 - Performance bottleneck analysis system and performance bottleneck analysis method - Google Patents

Description

  The present invention relates to an information processing system and an information processing method, and more particularly, to a performance bottleneck analysis system and a performance bottleneck analysis method for a computer system for identifying a cause of performance deterioration or performance improvement of a computer system as expected.

  An example of a performance bottleneck analysis system of a conventional computer system is described in Patent Document 1. As shown in FIG. 20, the conventional performance bottleneck analysis system 10 includes an input device 12, a processing device 14, and a storage device 16. The processing device 14 includes a measurement target setting unit 18, a semaphore wait state start detection unit 20, a semaphore wait state end detection unit 22, a semaphore information search unit 24, and a semaphore wait accumulated time calculation unit 26. The storage device 16 includes semaphore waiting information list storage means 28.

  The conventional performance bottleneck analysis system 10 having the above configuration operates as follows. First, a list of semaphore identifiers to be measured is designated by the user through the input device 12 and transmitted to the measurement target setting means 18. The measurement target setting unit 18 assigns to each semaphore identifier a semaphore waiting accumulated time indicating the accumulation of the time required to wait for the semaphore, and stores it in the semaphore waiting information list storage unit 28. At this time, the accumulated semaphore waiting time is initialized with a value of zero.

  The semaphore wait state start detection means 20 detects that a process (task) operating on the computer system 30 has changed to a semaphore wait state for a semaphore, and acquires the time at the time of detection (wait start time). Next, the semaphore wait state start detection means 20 gives the detected semaphore identifier and the wait start time to the semaphore information search means 24.

  The semaphore information search means 24 uses the list of semaphore wait information stored in the semaphore wait information list storage means 28 provided in the storage device 16 to search for the semaphore identifier received from the semaphore wait state start detection means 20, and Check whether the identifier is a measurement target. If the received semaphore identifier is a measurement target, a waiting start time is given to the semaphore identifier, and the semaphore waiting information list stored in the semaphore waiting information list storage unit 28 is updated.

  The semaphore wait state end detection means 22 detects that the process has ended the semaphore wait state for a semaphore and started another process, and acquires the time at detection (wait end time). Thereafter, the semaphore wait state end detection means 22 gives the detected semaphore identifier and the wait end time to the semaphore information search means 24.

  The semaphore information retrieval unit 24 retrieves the semaphore identifier received from the semaphore wait state end detection unit, and checks whether the semaphore identifier is a measurement target. At this time, if the received semaphore identifier is a measurement target, the waiting start time given to the semaphore identifier and the waiting accumulated time at that time are read, and the waiting start time, waiting end time, and waiting accumulated time are calculated. This is given to the waiting accumulated time calculating means 26.

  The semaphore waiting accumulated time calculating means 26 calculates the current waiting time from the received waiting start time and waiting end time, and adds this waiting time to the waiting accumulated time received together. Thereafter, the accumulated waiting time as a result of the addition is given to the semaphore information search means 24.

  The semaphore information search means 24 gives the received new accumulated waiting time together with the semaphore identifier to the semaphore wait information list storage means 28 provided in the storage device 16 to update the semaphore wait information list. At this time, the waiting start time assigned to the semaphore identifier is deleted.

  After the program ends, the semaphore with the longest accumulated semaphore waiting time is identified as the bottleneck.

JP-A-10-63516

  However, for some computer systems, semaphores with long latencies are not always a bottleneck. This is because a long waiting state is intended, and the waiting time does not decrease the performance efficiency of the system as long as other related processes effectively use hardware resources such as a CPU.

  Also, if there are multiple processes waiting for a specific semaphore, the true bottleneck may not be identified. Because the calculation of the waiting time is executed for each process, there are many processes waiting for a specific semaphore, and when each waiting time is relatively short, the semaphore is not specified as a bottleneck. Because.

  Furthermore, since the time at which the process state changes must be measured, a special mechanism is required in the program execution environment in order to detect a change to the process wait state and the end of the wait state. .

  In addition, since the user must specify the semaphore to be observed, the user needs to know the contents of the program to be analyzed.

  Therefore, the present invention has been made in view of the above-mentioned problems of the performance bottleneck analysis system in the conventional computer system, and the object of the present invention is to provide a semaphore and a file for a computer system whose software contents are not well understood. A performance bottleneck analysis system in a new and improved computer system capable of detecting performance degradation caused by waiting for software resources such as waiting for lock and waiting for data arrival via a network using a general-purpose execution environment and tools, and It is to provide a performance bottleneck analysis method.

  Another object of the present invention is to provide a performance bottleneck analysis system and a performance bottleneck analysis method in a new and improved computer system capable of estimating the amount of hardware resources such as a CPU optimal for software operation.

In order to solve the above problems, according to an aspect of the present invention, a plurality of processes at least one hardware resource (hereinafter referred to as HW resource.) In the performance bottleneck analysis system of a computer system that utilizes the For each of a plurality of processes, waiting for acquisition of one of the plurality of software resources (hereinafter referred to as SW resources) that are in the HW resource use state in which the HW resource is being used or that are mutually exclusive. A process state measuring unit for measuring a process state indicating a SW resource waiting state and a type of SW resource waiting for the SW resource waiting process, and the number of processes in the HW resource use state Unused status detecting means for detecting an unused status of HW resources smaller than the number of HW resources, and detecting the unused status of HW resources. In this case, the counting means for counting the number of processes waiting for SW resources (hereinafter referred to as the number of waiting processes) for each of the plurality of SW resources, and the number of waiting processes obtained for each SW resource are set as performance bottles. There is provided a performance bottleneck analysis system comprising an output means for outputting as a neck analysis result .

By adopting such a configuration, the process state measuring unit measures the state of each process operating on the external target system, and from the measurement result, the process in which the unused state detecting unit is using the HW resource is determined. When an HW unused status less than the number of HW resources is detected, and the HW resource unused status is detected, the counting means can calculate the number of waiting processes for each software resource when the HW unused status is detected . In other words, the configuration is such that the process that is using the HW (hardware) resource can detect the HW unused state where the number of HW resources is less than the number of HW resources and can calculate the number of SW resource waiting processes when the HW is not used. By considering a SW resource having a large number of processes as a bottleneck, it is possible to identify a SW (software) resource that is a bottleneck in performance.

At this time, the process state measuring means measures a plurality of times the process state, it is also possible that the counted number of the process further comprises an average value calculating means for calculating a number of processes waiting for the number average process by they said counting means .

  With this configuration, the average value of the number of SW resource waiting processes is calculated based on a plurality of measurements, so that it is possible to more accurately identify the SW resource that is a bottleneck in performance. it can.

Further, at this time, the measurement time acquisition means for calculating the time required for the measurement by the process state measurement means, and the time required for the measurement are compared with a predetermined threshold value to determine whether the bottleneck is on the HW resource side. or further comprising a determining HW Bottoleneck means, time required for measurement is a bottleneck if more than a predetermined threshold is determined to be in HW resource side, the output means is HW resource bottlenecks If the time required for the measurement is within a predetermined threshold value, it is determined that the bottleneck is not on the HW resource side, and the counting means may count the number of waiting processes.

  By adopting such a configuration, it is possible to calculate the time required for the measurement based on the fact that the process state measurement itself uses the HW resource, so it is possible to determine that the performance bottleneck is on the HW resource side. .

Further, at this time, when the HW bottleneck determining unit determines that the bottleneck is on the HW resource side, it may further include an HW configuration changing unit that changes the number of HW resources of the computer system .

  With such a configuration, the measurement time acquisition unit calculates the time required for the process state measurement means to measure the state of each process operating on the external system, and the HW bottleneck is calculated from the calculation result. The determination unit determines whether or not the HW resource is a bottleneck. If the bottleneck is a bottleneck, the determination unit increases the number of HW resources. If not, the determination unit operates to decrease the HW resource by a certain number. The For this reason, since the number of HW resources can be increased or decreased based on the location of the bottleneck in performance, the number of HW resources can be automatically adjusted, and the system can be operated with the optimum number of HW resources.

At this time, a filtering means for sorting and / or extracting the number of waiting processes on a predetermined basis may be provided.

  With such a configuration, SW resources with a large number of waiting processes can be extracted, so that SW resources that are bottlenecks in performance can be identified earlier.

  As described above, according to the present invention, a performance bottleneck caused by waiting for software resources such as waiting for semaphores, waiting for file locks, waiting for data arrival via the network, etc. Can be detected using the execution environment and tools.

  In addition, by automatically adjusting the number of HW resources, it is possible to estimate the amount of hardware resources such as a CPU that is optimal for the operation of software.

  Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, the duplicate description is abbreviate | omitted by attaching | subjecting the same code | symbol.

(First embodiment)
First, the configuration of the performance bottleneck analysis system for a computer system according to the first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a performance bottleneck analysis system for a computer system according to a first embodiment of the present invention.

  As shown in FIG. 1, the performance bottleneck analysis system 100 of this embodiment includes a processing device 102, a storage device 104, and an output device 106.

  The processing apparatus 102 includes a process state measuring unit 108, a hardware (hardware) resource use process number comparison unit 110, a SW (software) resource waiting process count unit 112, and an output unit 114. The storage device 104 also includes an HW resource number storage unit 116 that stores the number of HW resources included in the external system 120 and an SW resource queue length storage unit 118.

  Next, an outline of the operation of each component described above provided in the performance bottleneck analysis system 100 of the present embodiment will be described below.

  The process state measuring means 108 provided in the processing apparatus 102 acquires information on the execution state of a process operating on the external system 120, and a part of the acquired information is a HW resource use process number comparison unit provided in the processing apparatus 102. 110 and SW resource waiting process count unit 112.

  The HW resource utilization process number comparison unit 110 inputs data related to the process that is using the HW resource from the process state measurement unit 108, and inputs the number of HW resources of the external system 120 from the HW resource number storage unit 116 provided in the storage device 104. Then, the number of processes using the HW resource is compared with the number of HW resources. Then, the SW resource waiting process count unit 112 is informed of the magnitude relationship between the number of processes using the HW resource obtained from the comparison result and the number of HW resources.

  The SW resource wait process count unit 112 receives data related to a process in the SW resource wait state from the process state measuring unit 108, and the number of processes using the HW resource and the number of HW resources from the HW resource use process number comparison unit 110. Enter the magnitude relationship. If the number of processes using the HW resource is smaller than the number of HW resources, the type of SW resource that the process waiting for the SW resource waits is extracted, and for each SW resource, how many processes are waiting for the SW resource. The number of waiting processes related to each SW resource obtained as a result is stored in the SW resource queue length storage unit 118.

  The output unit 114 receives the number of waiting processes related to each SW resource from the SW resource queue length storage unit 118 and outputs the number to the output device 106 such as a display device, a file, or a disk.

  Next, the operation of the performance bottleneck analysis system 100 of the present embodiment will be described in detail with reference to FIGS. 1 and 2. FIG. 2 is a flowchart showing the operation of the performance bottleneck analysis system 100 of the computer system according to the present embodiment.

  First, in the present embodiment, as advance preparation, the number of HW (hardware) resources of the external system 120 is set in the HW resource number storage unit 116 (step A11). Here, the HW resource refers to a physical resource used by a plurality of users (processes) based on a scheduling policy represented by a time division method, such as a CPU (Central Processing Unit). A computer system can generally include a plurality of CPUs. The number of CPUs of a computer system can be known from system specifications, firmware such as BIOS (basic input output system), commands and functions provided by the OS, and the like.

  Next, the number of waiting processes QSi for each software resource Si stored in the SW resource queue length storage unit 118 is initialized to 0 (step A12). At this time, the order of step A11 and step A12 may be reversed.

  Next, the process state measuring unit 108 measures the current execution state of the process operating on the external system 120 (step A13). The execution state of the process includes the following two types of information. One is the distinction between whether the process is currently using the HW resource (executable state) or waiting for a specific SW (software) resource to become available (sleep state). The other is the type of SW resource that a dormant process waits for.

  The SW resource includes a semaphore and lock for performing exclusive control, and a flag indicating that transmission / reception of data to / from the network or disk has been completed. In the case of a UNIX (registered trademark, the same shall apply hereinafter) OS such as Linux, these pieces of information relating to the process state can be obtained by issuing a ps command on the external system 120, for example.

  Further, the performance bottleneck analysis system 100 of the present embodiment may be operated on the same system as the external system 120.

  Next, the HW resource utilization process number comparison unit 110 counts the number of processes R (the number of executable processes) that are using the HW resource measured by the process state measurement unit 108 (step A14), and the HW resource number storage unit. Referring to 116, the number of executable processes R is compared with the number of HW resources (step A15).

  What is desired to be detected here is a situation in which the number of executable processes R is less than the number of HW resources (HW unused status). When such a situation occurs frequently, there is a high possibility that performance degradation has occurred because the HW resources of the system 120 are not fully utilized. Most of the causes of such a situation are waiting for the SW resource of the process.

  In general, many wait states occur in a process in the course of calculation. In the wait state, other than waiting to acquire a semaphore for prohibiting simultaneous execution of critical sections by multiple processes and file lock for prohibiting simultaneous update of files and entries, network waits for transmission / reception and disk input / output There is waiting etc.

  Assuming that the system is a server that returns a result in response to a request from a client, as long as the content of the request is uniform and the usage rate of the HW resource is constant, It is almost constant. This is true regardless of the arrival interval of requests. This is because the amount of calculation required for processing a certain number of requests is constant unless the processing generated by the request performs a wasteful calculation such as an empty loop.

  Therefore, if the number of request arrivals is within the processing capability range, the response time of the request will be within a certain time. However, if the number of request arrivals exceeds the limit of the system processing capability, the response time will deteriorate rapidly. Begin to. This is due to a vicious circle in which the response time of requests deteriorates when requests waiting to be processed start to stay in the system, and as a result, the number of staying requests further increases.

  That is, until the utilization rate of the HW resource reaches the limit, the response time of the request does not usually deteriorate so much, so in the conventional performance bottleneck analysis, the HW resource utilization rate such as the CPU load is monitored, When the utilization rate became very high, the HW with the high utilization rate was identified as the bottleneck. However, since the process waiting for the SW resource does not consume the HW resource, it is difficult to identify the cause of the performance deterioration from the HW resource utilization rate such as the CPU load that is a normal analysis target. An object of the present invention is to detect performance degradation caused by waiting for SW resources and to identify the SW resource that is the cause.

  The response time of the system can be expressed as the sum of the HW resource use time and the SW resource wait time. Here, “using HW” means that the HW resource is shared while being used for a short time between the calculation subjects. To be precise, the HW usage time is the HW resource usage time. Is the sum of the time for physically using and the HW resource waiting time.

  Since the HW resources can be almost occupied when the HW is not being used, the HW resource waiting time is a relatively small value, while the HW resource utilization time is considered to be substantially constant. However, regarding SW resources that are bottlenecks, it takes a long time to acquire the resources, and the processes waiting for the SW resources stay, so the SW resources that the dormant process waits for are examined, and the SW resources that have many waiting processes are found. It is reasonable to regard it as a bottleneck.

  Therefore, as the next operation, when the HW underutilization state is observed, the SW resource waiting process count unit 112 checks each SW resource that the process waits from the dormant process information measured by the process state measuring unit 108, and The number of sleep processes waiting for the SW resource is counted one by one and stored in the SW resource queue length storage unit 118 (step A16 and step A17).

  When all the sleep processes are counted, the output unit 114 outputs the number of waiting processes related to each SW resource to the output device 106 such as a display device, a file, or a disk. Further, the result may be transmitted to an external system or apparatus. Further, when the HW resource utilization process number comparison unit 110 does not observe the HW unused state, it can also output that fact.

  As described above in detail, in this embodiment, it is possible to detect a situation where the number of processes using HW resources is less than the number of HW resources, and to calculate the number of processes waiting for each SW resource in such a situation. Therefore, it is possible to identify the SW resource that is a bottleneck in performance.

(Second Embodiment)
Next, the configuration of a performance bottleneck analysis system for a computer system according to a second embodiment of the present invention will be described with reference to the drawings. FIG. 3 is a block diagram showing a configuration of a performance bottleneck analysis system 200 for a computer system according to the second embodiment of the present invention.

  As shown in FIG. 3, the performance bottleneck analysis system 200 according to the second embodiment of the present invention has a filtering device that includes a processing unit 202 as compared with the performance bottleneck analysis system 100 according to the first embodiment shown in FIG. 1. The difference is that 220 is provided.

  The filtering unit 220 inputs the number of waiting processes related to each SW resource from the SW resource queue length storage unit 218 and gives the SW resource extracted or sorted under a certain condition to the output unit 214.

  Next, the overall operation of the performance bottleneck analysis system 200 of this embodiment will be described in detail with reference to FIGS. 3 and 4. FIG. 4 is a flowchart showing the operation of the performance bottleneck analysis system 200 of the computer system according to this embodiment.

  The operations of the process state measuring unit 108, the HW resource utilization process number comparison unit 110, and the SW resource waiting process count unit 112 in the present embodiment indicated by steps A21 to A27 in FIG. 4 are the operations of the corresponding parts in the first embodiment. Therefore, the description of these operations is omitted.

  In the first embodiment, the process is terminated when the number of waiting processes for each SW resource is obtained. In this embodiment, the filtering unit 220 outputs the SW resources extracted or sorted using the number of waiting processes. To the unit 214 (step A28 in FIG. 4). As an extraction or sorting method, only SW resources whose number of waiting processes exceeds a certain number are extracted, or SW resources are sorted in order of increasing or decreasing number of waiting processes, or SW resources sorted using the number of waiting processes are sorted. For example, only those in which the number of waiting processes exceeds a certain number are extracted in the sorted order. The output unit 214 may be given not only the extracted or sorted SW resources but also the number of waiting processes.

  Thus, in this embodiment, since it is comprised so that SW resource with many number of waiting processes can be extracted, SW resource used as the bottleneck in performance can be identified earlier.

(Third embodiment)
Next, the configuration of the performance bottleneck analysis system in the computer system according to the third embodiment of the present invention will be described with reference to the drawings. FIG. 5 is a block diagram showing the configuration of a performance bottleneck analysis system 300 for a computer system according to the third embodiment of the present invention.

  As illustrated in FIG. 5, the performance bottleneck analysis system 300 according to the third embodiment of the present invention has an average value of the processing device 302 as compared with the performance bottleneck analysis system 100 according to the first embodiment illustrated in FIG. 1. The difference is that the calculation unit 320 is included, and the storage device 304 includes an unused event occurrence number storage unit 322.

  The average value calculation unit 320 refers to one or more waiting process numbers related to each SW resource stored in the SW resource queue length storage unit 318, calculates an average value of the number of waiting processes related to each SW resource, and outputs an output unit 314.

  Next, the overall operation of the performance bottleneck analysis system 300 of this embodiment will be described in detail with reference to FIGS. 5 and 6. FIG. 6 is a flowchart showing the operation of the performance bottleneck analysis system 300 of the computer system according to the present embodiment.

  The operations of the process state measuring unit 108, the HW resource utilization process number comparison unit 110, and the SW resource waiting process count unit 112 in the present embodiment indicated by steps A31 to A37 in FIG. 6 are the operations of the corresponding parts of the first embodiment. Therefore, the description of these operations is omitted.

  In the first embodiment, the process state is measured only once, but in the present embodiment, measurement is performed a plurality of times. For this reason, after step A32, a variable C for counting the number of occurrences of the HW underutilization situation is initialized to 0 and stored in the unutilized event occurrence number storage unit 322 (step B31 in FIG. 6). Step B31 may be performed before step A32 or before step A31.

  Next, it is determined whether or not to continue the measurement (step B32). When the number of times of measurement reaches a predetermined number of times, when an end command arrives from the outside, or when a certain time has passed, the measurement is ended.

  When the measurement is continued, if the HW unused state is not observed in step A35, the execution after step B32 is repeated. When the HW unused state is observed in step A35, the HW resource utilization process number comparison unit 110 increases the variable C stored in the unused event occurrence number storage unit 322 by 1 (step B33).

  Thereafter, the number of waiting processes measured this time is added to the number of waiting processes QSi for each SW resource Si by the loop processing of steps A36 and A37, and the SW resource queue length storage unit 318 is updated. When the addition of the number of waiting processes is finished, the execution of the processes after step B32 is repeated again.

  When the measurement is completed in step B32, the average value calculation unit 320 first compares the variable C stored in the unutilized event occurrence count storage unit 322 with 0 to determine whether an HW unutilized situation has occurred. (Step B34).

  When the HW unused status has occurred (C ≠ 0), the number of waiting processes QSi for each SW resource Si holds the total value by one or more measurements, so that each QSi is not used in HW Dividing by the number of occurrences C of the situation, the average value of the number of waiting processes for each SW resource Si is calculated (step B35).

  If no HW unused state has occurred, the process is terminated. Further, the fact that it has not occurred may be transmitted to the output unit 314 and output to the output device 106.

  Finally, the output unit 314 outputs the average value of the number of waiting processes for each SW resource Si to the output device 106. At this time, the result of extraction or sorting using the average value of the number of waiting processes may be output as executed by the filtering unit (220 in FIG. 3) in the second embodiment.

  As described above, in the present embodiment, the measurement is performed a plurality of times and the average value of the number of waiting processes for the SW resource is calculated. By reducing the influence of the general state, it is possible to more accurately identify the SW resource that is a bottleneck in performance.

(Fourth embodiment)
Next, the configuration of a performance bottleneck analysis system for a computer system according to a fourth embodiment of the present invention will be described with reference to the drawings. FIG. 7 is a block diagram showing the configuration of a performance bottleneck analysis system 400 for a computer system according to the fourth embodiment of the present invention.

  As shown in FIG. 7, the performance bottleneck analysis system 400 according to the fourth embodiment of the present invention requires measurement by the processing device 402 as compared with the performance bottleneck analysis system 100 according to the first embodiment shown in FIG. The difference is that the time acquisition unit 422 and the HW bottleneck determination unit 420 are provided.

  The measurement required time acquisition unit 422 acquires the measurement start time and end time of the process state measurement unit 108, calculates the time required for the measurement, and provides the time to the HW bottleneck determination unit 420.

  The HW bottleneck determination unit 420 inputs the measurement required time from the measurement required time acquisition unit 422 and determines whether the HW resource is a bottleneck by comparing the measurement required time with a threshold value. If the HW resource is not a bottleneck, the number of waiting processes related to each SW resource is input from the SW resource queue length storage unit 418 and provided to the output unit 414.

  Next, the overall operation of the performance bottleneck analysis system 400 of the present embodiment will be described in detail with reference to FIGS. FIG. 8 is a flowchart showing the operation of the performance bottleneck analysis system 400 of the computer system according to the present embodiment.

  The operations of the process state measuring means 108, the HW resource utilization process number comparison unit 110, and the SW resource waiting process count unit 112 in the present embodiment indicated by steps A41 to A47 in FIG. 8 are the operations of the corresponding parts in the first embodiment. Therefore, the description of these operations is omitted.

  In the first embodiment, after the process state is measured (step A13), the number of processes that are using the HW resource is counted immediately (step A14). However, in this embodiment, the measurement required time acquisition unit 422 is counted. However, the time taken for the measurement is calculated in cooperation with the measurement operation by the process state measurement means 108.

  Next, the HW bottleneck determination unit 420 determines whether or not the required measurement time calculated by the required measurement time acquisition unit 422 is within a certain range (step C41 in FIG. 8).

  If the measurement is completed within a certain time, it is determined that the bottleneck does not exist on the HW resource side (step C42). The threshold value at this time is given from the outside based on data during normal operation or test operation of the system. Subsequent operations in steps A44 to A47 are the same as those in the first embodiment.

  When the measurement takes a certain time or more, the HW bottleneck determination unit 420 determines that the bottleneck is on the HW resource side (step C43).

  This determination is made because the process state measurement itself uses a HW resource such as a CPU. For example, if there are a large number of processes using HW resources on the external system 120, even if the ps command is issued for measuring the process state, the ps command processing is not scheduled and the response is very delayed. Therefore, the process state cannot be measured immediately. In such a case, it can be determined that the HW resource is a bottleneck.

  Finally, if the HW resource is a bottleneck, the output unit 414 outputs a message to that effect to the output device 106. If the HW resource is not a bottleneck, whether or not an HW unused state has been observed as in the first embodiment. Based on whether or not, the number of waiting processes for each SW resource is output to the output device 106.

  As described above, in the present embodiment, since the process state measurement itself is configured to be able to calculate the time required for the process state measurement based on the fact that the HW resource is used, there is a performance bottleneck. It can be determined that it is on the HW resource side.

(Fifth embodiment)
Next, the configuration of a performance bottleneck analysis system for a computer system according to a fifth embodiment of the present invention will be described with reference to the drawings. FIG. 9 is a block diagram showing the configuration of a performance bottleneck analysis system 500 for a computer system according to the fifth embodiment of the present invention.

  As shown in FIG. 9, the fifth embodiment of the present invention has a processing time 502 acquisition unit 522 and an HW bottle in the processing device 502 as compared with the performance bottleneck analysis system 300 according to the third embodiment shown in FIG. 5. The difference is that the neck determination unit 524 is included, and the storage device 504 includes a measurement required time and measurement count storage unit 526. The measurement required time and measurement count storage unit 526 stores the time and measurement count required for the process state measurement unit 108 to measure the process state.

  Next, the overall operation of the performance bottleneck analysis system 500 of this embodiment will be described in detail with reference to FIGS. 9 and 10. FIG. 10 is a flowchart showing the operation of the performance bottleneck analysis system 500 of the computer system according to the present embodiment.

  In the performance bottleneck analysis system 500 of the computer system according to the present embodiment indicated by steps A51 to A57 and B51 to B55 in FIG. 10, the process state measuring means 108, the HW resource utilization process number comparison unit 110, and the SW resource waiting process count unit 112 Since the operation of the average value calculation unit 320 is the same as the operation of the corresponding part of the third embodiment, description of these operations will be omitted.

  In the third embodiment, the time required for the measurement of the process state is not acquired, but in this embodiment, the time required for the measurement is stored once or more, and the average value of the time required for the measurement is obtained. To determine whether the bottleneck is on the HW resource side.

  First, after step B51, the total number TT of the number of times of measurement N and the time required for measurement is initialized to 0 (step D51 in FIG. 10). This step D51 may be performed before step B51, step A52, or step A51.

  Next, when the measurement is continued in step B52, the measurement count N is incremented by 1 after measuring the process state in step A53 (step D52). This step D52 may be performed before step A53.

  Next, the time T required for the current measurement is added to the total number TT of the measurement times N and the required measurement time (step D53). The subsequent processing (after step A54) is the same as that after step A53 of the third embodiment.

  In step B52, when the measurement is completed, TT / N that is the average of the required measurement time is calculated (step D54). At this time, if there is a possibility that the measurement has not been performed once, it may be confirmed that the number of times of measurement N is not 0 before the calculation of the average value, so that division by 0 does not occur.

  Next, it is determined whether or not the average required time TT / N is within a certain range (step D55).

  If it is within a certain range, it is determined that the bottleneck is not on the HW resource side (step C52). The subsequent operations of Steps B54 and B55 are the same as Steps B34 and B35 of the third embodiment.

  If the average measurement time TT / N is not within a certain range, it is determined that the bottleneck is on the HW resource side (step C53).

  Finally, if the HW resource is a bottleneck, the output unit 514 outputs a message to that effect to the output device 106. If the HW resource is not a bottleneck, the HW unused state is observed as in the first embodiment. Based on whether or not, the number of waiting processes for each SW resource is output to the output device 106.

  As described above, in this embodiment, the time required for measuring the process state is measured a plurality of times, and the average value can be calculated. Therefore, the performance bottleneck is on the HW resource side. More accurate judgment can be made.

(Sixth embodiment)
Next, the configuration of a performance bottleneck analysis system for a computer system according to a sixth embodiment of the present invention will be described with reference to the drawings. FIG. 11 is a block diagram showing the configuration of a performance bottleneck analysis system 600 for a computer system according to the sixth embodiment of the present invention.

  Referring to FIG. 11, the performance bottleneck analysis system 600 according to the sixth embodiment of the present invention has a HW configuration changing unit 640 as compared with the performance bottleneck analysis system 400 according to the fourth embodiment shown in FIG. 7. It is different in point.

  The HW configuration changing means 640 inputs whether the bottleneck is in the HW resource or the SW resource from the HW bottleneck determination unit 620 provided in the processing device 602, and changes the number of HW resources that the external system 120 has. This change is reflected in the HW resource number storage unit 116 provided in the storage device 604.

  Next, the overall operation of the performance bottleneck analysis system 600 of this embodiment will be described in detail with reference to FIGS. 11 and 12. FIG. 12 is a flowchart showing the operation of the performance bottleneck analysis system 600 of the computer system according to this embodiment.

  In the performance bottleneck analysis system 600 of the computer system according to the present embodiment shown in steps A61 to A67 and steps C61 to C63 in FIG. 12, the process state measuring means 108, the HW resource utilization process number comparison unit 110, and the SW resource waiting process count unit The operations of the measurement required time acquisition unit 422 and the HW bottleneck determination unit 420 are the same as the operations of the corresponding parts in the fourth embodiment, and thus description of these operations is omitted.

  In the fourth embodiment, the processing is terminated when it is determined in step C43 that the bottleneck is on the HW resource side. However, in this embodiment, the HW configuration is changed (step E61 in FIG. 12).

  Specifically, when the bottleneck is on the HW resource side, it is considered that the HW resource is insufficient, and the number of HW resources is increased. Such addition of HW resources can be realized by executing processing such as restarting after changing the parameters of the OS or BIOS in the external system 120. Further, in a system configured with a set of a plurality of calculation nodes and capable of dynamically changing the number of calculation nodes to be used, it is only necessary to issue a calculation node addition command.

  Whether the configuration of the HW can be changed is input from the outside according to the status of the external system 120 or the like. If the HW configuration has been changed, the number of HW resources is reset (step A61), and the processing after step A63 is repeated. If the HW configuration is not changed, the process ends.

  Also, although not described in the flowchart of FIG. 12, if it is determined that the bottleneck is not on the HW resource side, it is considered that there is a surplus in the HW resource in the system, and in any process after step C62 HW resources can also be reduced.

  Finally, if the location of the bottleneck or the configuration of the HW has been changed by the HW configuration change means 640, the output unit 614 outputs the change result to the output device 106. If the bottleneck is not on the HW resource side, the number of waiting processes related to each SW resource is output to the output device 106 based on whether or not an HW underutilization situation has been observed.

  Thus, in this embodiment, since it is comprised so that the number of HW resources can be increased / decreased based on the location of a bottleneck, a system can be operated with the optimal number of HW resources.

(Seventh embodiment)
Next, the configuration of a performance bottleneck analysis system for a computer system according to a seventh embodiment of the present invention will be described with reference to the drawings. FIG. 13 is a block diagram showing a configuration of a performance bottleneck analysis system 700 for a computer system according to the seventh embodiment of the present invention.

  Referring to FIG. 13, the performance bottleneck analysis system 700 according to the seventh embodiment of the present invention has a HW configuration changing unit 740, as compared with the performance bottleneck analysis system 500 according to the fifth embodiment shown in FIG. The difference is that the processing device 702 includes a storage initialization unit 730.

  Similarly to the HW configuration change unit 640 in the sixth embodiment, the HW configuration change unit 740 inputs whether the bottleneck is in the HW resource or the SW resource from the HW bottleneck determination unit 724 provided in the processing device 702. Then, the number of HW resources included in the external system 120 is changed, and this change is reflected in the HW resource number storage unit 116 provided in the storage device 704.

  The storage initialization unit 730 includes the measurement required time and measurement count storage unit 726, the HW resource number storage unit 116, the unutilized event occurrence count storage unit 722, and the SW resource queue length storage, which are the constituent elements of the storage device 704. The contents stored in the unit 718 are initialized to the initial state.

  Next, the overall operation of the performance bottleneck analysis system 700 of this embodiment will be described in detail with reference to FIGS. 13 and 14. FIG. 14 is a flowchart showing the operation of the performance bottleneck analysis system 700 of the computer system according to this embodiment.

  The process state measuring means 108 and the HW resource utilization process comparison unit 110 in the performance bottleneck analysis system 700 of the computer system according to the present embodiment indicated by steps A71 to A77, B71 to B75, C72 to C73, and D71 to D75 in FIG. The operations of the SW resource waiting process count unit 112, the average value calculation unit 320, the required measurement time acquisition unit 522, and the HW bottleneck determination unit 724 are the same as the operations of the corresponding parts in the fifth embodiment. The description of is omitted.

  In the fifth embodiment, the processing is terminated when it is determined in step C53 that the bottleneck is on the HW resource side, but in this embodiment, as in the sixth embodiment, an attempt is made to change the HW configuration ( Step E71 in FIG. 14).

  When the HW configuration is changed, the storage initialization unit 730 provided in the processing device 702 re-executes the initialization processing of Step A71, Step A72, Step B71, and Step D71 in order to perform analysis again from the beginning. The storage contents of the HW resource number storage unit 116, the SW resource queue length storage unit 718, the unutilized event occurrence count storage unit 722, and the measurement required time and measurement count storage unit 726 of the storage device 704 are initialized. After the initialization, the processing after step B72 is further repeated.

  Although not described in the flowchart of FIG. 14, if it is determined in step C72 that the bottleneck is not on the HW resource side, it is considered that there is a surplus in the HW resource in the system, and any of the steps after step C72 is performed. In this step, HW resources can be reduced.

  Finally, if the location of the bottleneck or the configuration of the HW has been changed by the HW configuration changing means 740, the output unit 714 outputs the change result to the output device 106. If the bottleneck is not on the HW resource side, the number of waiting processes related to each SW resource is output to the output device 106 based on whether or not an HW underutilization situation has been observed.

  Thus, in this embodiment, since it is comprised so that the number of HW resources can be increased / decreased based on the location of a bottleneck, a system can be operated with the optimal number of HW resources. In addition, since the time required for measuring the process state is measured a plurality of times and the average value can be calculated, when the HW resource is a bottleneck, the performance bottleneck is the HW resource. It is possible to more accurately determine that it is on the side.

(Eighth embodiment)
Next, the configuration of a performance bottleneck analysis system for a computer system according to an eighth embodiment of the present invention will be described with reference to the drawings. FIG. 15 is a block diagram showing a configuration of a performance bottleneck analysis system 800 for a computer system according to an eighth embodiment of the present invention.

  As shown in FIG. 15, the performance bottleneck analysis system 800 of this embodiment includes a processing device 802, a storage device 804, and an output device 106.

  The processing device 802 in this embodiment includes an SW resource identifier acquisition unit 806, an SW resource wait time measurement unit 808, and an output unit 810. The storage device 804 includes an SW resource identifier storage unit 812, and an SW resource wait time. A storage unit 814.

  The SW resource identifier acquisition unit 806 acquires identifier information of each SW resource used by a process operating on the external system 120, and stores the identifier information in the SW resource identifier storage unit 812.

  The SW resource wait time measurement unit 808 inputs the SW resource identifier from the SW resource identifier storage unit 812, newly inputs a process waiting for the acquisition of the SW resource to the external system 120, and sets the time required for the acquisition of each SW resource. It is measured and stored in the SW resource waiting time storage unit 814 provided in the storage device 804.

  The output unit 810 inputs the time required to acquire each SW resource from the SW resource wait time storage unit 814 provided in the storage device 804, and outputs this to the output device 106.

  Next, the overall operation of the performance bottleneck analysis system 800 of this embodiment will be described in detail with reference to FIGS. 15 and 16. FIG. 16 is a flowchart showing the operation of the performance bottleneck analysis system 800 of the computer system according to this embodiment.

  First, the SW resource identifier acquisition unit 806 checks a system call issued to the OS by a process operating on the external system 120, and acquires identifier information of the SW resource specified as an argument of the system call. Then, the data is stored in the SW resource identifier storage unit 812 included in the storage device 804 (step F81 in FIG. 16).

  The argument information of the above system call can be obtained from the source code of the program when the identifier is directly described on the program. In addition, for example, for a process that runs on a UNIX OS such as Linux, a system call issued by a running process and an argument at the time of the call are issued in real time by a command such as trace or a system call such as ptrace. Can be obtained. When the time available for measurement is limited, only the SW resource identifier that can be acquired within a certain time is stored in the SW resource identifier storage unit 812.

  Next, the SW resource waiting time measuring unit 808 inputs one or more SW resource identifiers from the SW resource identifier storage unit 812, and creates a program waiting for the SW resources for each SW resource identifier, or a program that has already been created Is executed on the external system 120 by giving the SW resource identifier as an argument, and the time taken to actually acquire the SW resource is measured.

  The program to be executed at this time first acquires the current time before the process of waiting for the SW resource, then acquires the SW resource using the SW resource identifier, acquires the current time again after acquisition, and acquires the current time. This program calculates the time required for acquisition from the previous and subsequent times. The acquired SW resource identifier is released as soon as possible.

  SW resource waiting time is measured for all SW resource identifiers acquired by the SW resource identifier acquiring unit 806 (step F82 and step F83), and the measured required time is stored in the SW resource waiting time storage unit 814.

  Finally, the output unit 810 inputs the SW resource wait time measured for each SW resource identifier from the SW resource wait time storage unit 814, and outputs the wait time to the output device 106. At this time, the SW resource type and SW resource identifier may be output together with the waiting time. At this time, the SW resource identifiers may be sorted using the waiting time, or only the SW resource identifiers having a certain waiting time or more may be extracted and output.

  As described above, in the present embodiment, the SW resource identifier is configured so that the time required to acquire the SW resource can be measured. Can be identified.

(Ninth embodiment)
Next, the configuration of a performance bottleneck analysis system for a computer system according to a ninth embodiment of the present invention will be described with reference to the drawings. FIG. 17 is a block diagram showing a configuration of a performance bottleneck analysis system 900 for a computer system according to the ninth embodiment of the present invention.

  As shown in FIG. 17, the performance bottleneck analysis system 900 of this embodiment is different from the performance bottleneck analysis system 800 according to the eighth embodiment shown in FIG. The HW resource utilization process number comparison unit 918 is different from the storage device 904 in that the storage device 904 includes an HW resource number storage unit 920.

  The operations of the process state measurement unit 916 and the HW resource utilization process number comparison unit 918 are the same as the corresponding units and processing units described in the first embodiment.

  Next, the overall operation of the performance bottleneck analysis system 900 of this embodiment will be described in detail with reference to FIGS. 17 and 18. FIG. 18 is a flowchart showing the operation of the performance bottleneck analysis system 900 of the computer system according to this embodiment.

  The operations of the process state measuring unit 916 and the HW resource utilization process number comparison unit 918 in the present embodiment shown in Step A91 and Steps A93 to A95 in FIG. 18 are the same as the operations of the corresponding parts in the first embodiment. Description is omitted.

  In the first embodiment, when the HW underutilization state is observed in step A15, the number of dormant processes waiting for each SW resource is counted. However, in this embodiment, the SW resource resource identifier acquisition unit 906 For each process in the SW resource waiting state, the system call being issued is checked, SW resource identifier information given as an argument is acquired, and stored in the SW resource identifier storage unit 812 (step A96 and FIG. 18). Step G91).

  At this time, the system call and argument issued when the SW resource waiting process transitions to the SW resource waiting state are, for example, in the case of a UNIX-based OS such as Linux, by a command such as trace or a system call such as ptrace. It can be acquired.

  When the SW resource identifier information has been acquired for all the SW resource waiting processes, the SW resource waiting time measurement unit 808 then performs the SW resource identifier storage unit as in the corresponding processing unit in the eighth embodiment. One or more SW resource identifiers are input from 812, SW resource waiting time is measured for each SW resource identifier, and the measured waiting time is stored in the SW resource waiting time storage unit 814 (step F92 and step F93). .

  Finally, as in the eighth embodiment, the output unit 810 inputs the SW resource wait time measured for each SW resource identifier from the SW resource wait time storage unit 814 and outputs the SW resource wait time to the output device 106. At this time, the SW resource type and SW resource identifier may be output together with the waiting time.

  As described above, the present embodiment is configured to collect the SW resource identifier information only for the process waiting for the SW resource and measure the time required to acquire each SW resource. It is possible to identify the SW resource that is the bottleneck and the identifier thereof earlier.

  Next, the operation of the best mode for carrying out the present invention will be described using specific embodiments described below.

  Example 1 corresponds to the first and second embodiments of the present invention described above.

  Here, it is assumed that the HW resource is a CPU, and the number of HW resources included in the external system 120 is four, that is, four CPUs. Further, it is assumed that Linux, which is a UNIX OS, is installed in the external system 120.

  First, an integer value 4 that is the number of HW resources provided in the external system 120 is set in the storage device 104 such as a memory.

  Next, the process state measuring unit 108 issues a ps command on the external system 120 and acquires information on the execution state of the process. Here, an output example of the ps command is shown in FIG. FIG. 19 is a diagram illustrating an output example in which the ps command is issued with the l option.

  As shown in FIG. 19, each of the third and subsequent rows corresponds to each process, the PID column represents the process ID (identifier), and the COMMAND column represents the name of the process. The STAT column represents an outline of the process state. The R display represents that the CPU is being used, and the S display represents a sleep state waiting for SW resources. Further, the WCHAN column represents a function in the OS that is executed when a dormant process waits for acquisition of SW resources. For example, semop is displayed when waiting for acquisition of a semaphore, msggrcv is displayed when waiting for arrival of a message in inter-process communication, and wait4 is displayed when waiting for termination of a child process. That is, the type of SW resource that a sleeping process waits for can be known from the WCHAN column.

  Next, from the output of the ps command, the HW resource utilization process number comparison unit 110 counts the number of R processes indicating that the CPU is being used, that is, the STAT representing the outline of the process state is using the CPU. At this time, all processes may be targeted, but in general, there are processes that are not directly related to the process whose performance is to be analyzed on the external system 120, and therefore, a series of processes named abc here. Let's consider the performance.

  The number of CPUs provided in the external system 120 is four. However, for the process abc, since the number of processes using the CPU is only three, the HW resource use process number comparison unit 110 is used by the external system 120. It is detected that there is an unutilized HW in which a CPU that has not been used exists. At this time, the fact that it has been detected may be given to the output unit 114 and output to the output device 106.

  Next, the SW resource waiting process count unit 112 receives the detection of the HW unused state and counts how many dormant processes wait for each SW resource. In the example of FIG. 19, wait 4, that is, one process waiting for the end of a child process, three semops, that is, three processes waiting for a semaphore, and two msgrc, that is, two processes waiting for a message are calculated.

  Finally, the output unit 114 outputs the calculation result to the output device 106 such as a display. SW resources with many waiting processes are a more important bottleneck. At this time, the filtering unit 220 in the second embodiment may be used to sort in the order of the number of waiting processes and output such as semop: 3, msgrcv: 2, wait4: 1. Alternatively, only SW resources having a certain number of waiting processes may be output.

  Next, the operation of the best mode for carrying out the present invention will be described using the second embodiment. Example 2 corresponds to the third embodiment of the present invention.

  Similar to the first embodiment, the number of dormant processes waiting for each SW resource in the HW unused state is counted. In this embodiment, the average value of the number of dormant processes is obtained by performing a plurality of measurements.

  For example, it is assumed that the measurement is performed 5 times, and the HW unused state is detected about 3 times. When the first HW underutilization state is detected, wait4: 1, semop: 3, msggrcv: 2, that is, one process waiting for child process termination, three processes waiting for semaphores, and two processes waiting for messages are observed. In the second and third detections, it is assumed that observations are made as wait4: 1, semop: 3, msgrcv: 3 and wait4: 1, semop: 4, msggrcv: 1, respectively.

  The average value calculation unit 320 uses the total number of waiting processes for each SW resource stored in the SW resource queue length storage unit 318, that is, wait4: 3, semop: 10, msgrcv: 6 as the number of occurrences of unused events. By dividing by 3 stored in the storage unit 322, the average number of waiting processes wait4: 1.00, semop: 3.33, msgrcv: 2.00 is calculated. At this time, the average number of waiting processes may be a decimal value. Then, the output unit 314 outputs the obtained average number of waiting processes for each SW resource to the output device 106.

  Next, the operation of the best mode for carrying out the present invention will be described using the third embodiment. Example 3 corresponds to the fourth and fifth embodiments of the present invention.

  To explain using a specific example similar to the first embodiment, in this embodiment, the time required from the process state measuring means 108 issuing the ps command to acquiring the result of the ps command is measured. The unit 422 performs measurement.

  For the measurement at this time, a system call for obtaining the current time such as gettimeofday is used to obtain the time immediately before and after the execution of the ps command to calculate the response time of the command. There is a method of using a command such as time for outputting time.

  Next, the HW bottleneck determination unit 420 checks whether the required measurement time is within a certain range. When the measurement required time is 2.3 seconds and the allowable value is 1.5 seconds, the measurement required time exceeds the allowable value, so the HW bottleneck determination unit 420 determines that the bottleneck is on the HW resource side. That is, it is determined that it is in the CPU. In this case, the output unit 414 outputs to the output device 106 that the bottleneck is a CPU.

  If the required measurement time does not exceed the allowable value, the HW unused status is detected as in the first embodiment, and the average number of waiting processes for each SW resource is output.

  In the fifth embodiment, the HW bottleneck is determined based on the average value of the required measurement times based on a plurality of measurements. For example, it is assumed that the average value of the required time is 1.3 seconds as a result of measuring the process state 100 times per minute. If the allowable value is 1.5 seconds, it is determined that there is no bottleneck on the HW resource side during this period.

  Next, the operation of the best mode for carrying out the present invention will be described using a fourth embodiment. Example 4 corresponds to the sixth and seventh embodiments of the present invention.

  In the present embodiment, when the HW bottleneck determination unit 620 determines that the bottleneck is in the CPU, the CPU of the external system is augmented.

  For example, when there are unused CPUs or CPUs assigned to other services in the external system 120, the CPUs assigned to the processes subject to performance analysis by means such as OS parameter and BIOS rewriting and restarting. You can increase the number. The external system 120 may be replaced or switched to a computer system with higher performance. For example, a computer system connected to a network can be automatically switched to another computer system by means such as dynamic change of an IP address.

  When the external system 120 is switched or the configuration is changed, the output unit 614 outputs a message to that effect to the output device 106.

  When the processing is further continued, the number of HW resources held in the storage device 604 is reset to the changed configuration, and the processing after the measurement of the process state is repeated.

  In the seventh embodiment, whether the bottleneck is on the HW resource side is determined based on the average value of the measurement results as in the fifth embodiment in the third embodiment.

  Next, the operation of the best mode for carrying out the present invention will be described using a fifth embodiment. Example 5 corresponds to the eighth embodiment of the present invention.

  First, the SW resource identifier acquisition unit 806 acquires identifier information of each SW resource used by a process operating on the external system 120. For example, in the case of a semaphore, the SW resource identifier is a semaphore ID for identifying each semaphore. In the case of a UNIX OS or the like, the semaphore ID is a simple integer value.

  In some cases, the SW resource identifier can be obtained directly from the source code of the program, and the SW resource identifier can be obtained for a running process by using the trace command. For example, in the case of a process that waits for a semaphore, if the process ID is 246, for example, an output semop (8,.

  Next, the SW resource wait time measurement unit 808 creates a program that actually acquires the semaphore using the acquired semaphore ID and the function semaphore that acquires the semaphore, and executes it on the external system 120.

  The measurement of the time taken for acquisition includes a method of calculating from the time immediately before and immediately after calling the function semop using a system call gettimeofday or the like, and a method of using a time command.

  The same measurement is performed for different semaphore identifiers and SW resources other than semaphores.

  Finally, the output unit 810 outputs the measurement result to the output device 106 such as a display.

  Next, the operation of the best mode for carrying out the present invention will be described using Example 6. Example 6 corresponds to the ninth embodiment of the present invention.

  In the present embodiment, the SW resource identifier acquisition unit 906 is used for a dormant process acquired by the process state measurement unit 916 from the output of the ps command. Uses a command such as “trace” to obtain the identifier of the SW resource that the sleeping process waits for.

  The SW resource wait time measurement unit 808 uses the SW resource identifier to determine the actual wait time for the SW resource when the HW resource utilization process number comparison unit 918 detects the HW unused state, as in the fifth embodiment. measure.

  Finally, the output unit 810 outputs the measurement result to the output device 106 such as a display.

  As mentioned above, although preferred embodiment of this invention was described referring an accompanying drawing, it cannot be overemphasized that this invention is not limited to the example which concerns. It will be apparent to those skilled in the art that various changes and modifications can be made within the scope of the claims, and these are naturally within the technical scope of the present invention. Understood.

  The present invention is applicable to a performance bottleneck analysis system in a computer system, and in particular, a performance evaluation system for analyzing the cause of performance degradation by monitoring the state of a server, and for identifying a performance bottleneck at the time of system development It is applicable to uses such as development tools. In addition, in a system in which a plurality of computing nodes such as parallel computers and cluster type servers are prepared and the number of computing nodes used by a user can be adjusted, the computing nodes used by the user for reasons such as usage efficiency and charging for usage It is also applicable to uses such as optimizing the number of

It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 1st Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 3rd Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 4th Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 5th Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 6th Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 7th Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 8th Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a block diagram which shows the structure of the performance bottleneck analysis system of the computer system by 9th Embodiment of this invention. It is a flowchart which shows operation | movement of the performance bottleneck analysis system of the computer system by the embodiment. It is a figure which shows the example output of a ps command. It is a block diagram which shows the structure of the performance bottleneck analysis system of the conventional computer system.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Performance bottleneck analysis system 102 Processing apparatus 104 Storage apparatus 106 Output apparatus 108 Process state measurement means 110 HW resource utilization process number comparison part 112 SW resource waiting process count part 114 Output part 116 HW resource number storage part 118 SW resource queue length Storage unit 120 System 220 Filtering unit 320 Average value calculation unit 322 Unutilized event occurrence number storage unit 420 Bottleneck determination unit 422 Measurement required time acquisition unit 526 Measurement required time and measurement number storage unit 640 HW configuration change means 730 Storage initialization unit 806 SW resource identifier acquisition means 808 SW resource waiting time measurement unit 812 SW resource identifier storage unit 814 SW resource waiting time storage unit

Claims (12)

In a performance bottleneck analysis system for a computer system in which a plurality of processes use at least one hardware resource (hereinafter referred to as HW resource),
For each of the plurality of processes, waiting for acquisition of one of a plurality of software resources (hereinafter referred to as SW resources) that are in the HW resource use state in which the HW resource is being used or are mutually exclusive. A process state measuring means for measuring a process state indicating each of the SW resource waiting states and the types of SW resources waiting for the SW resource waiting state processes;
An underutilization status detecting means for detecting an unutilized status of an HW resource in which the number of processes in the HW resource utilization status is less than the number of the HW resources;
Counting means for counting the number of processes waiting for SW resources (hereinafter referred to as the number of waiting processes) for each of the plurality of SW resources when the HW resource unused state is detected;
An output means for outputting the number of waiting processes obtained for each SW resource as a performance bottleneck analysis result;
A performance bottleneck analysis system comprising: The process state measuring means measures the process state a plurality of times,
2. The performance bottleneck analysis system according to claim 1, further comprising an average value calculating unit that calculates an average number of processes as the number of waiting processes from the number of processes counted by the counting unit. And measuring the required time obtaining means for calculating a time during which the process state measuring means required for the measurement,
HW bottleneck determination means for comparing the time required for the measurement with a predetermined threshold to determine whether the bottleneck is on the HW resource side ;
Further comprising
The HW bottleneck determining means determines that the bottleneck is on the HW resource side if the time required for the measurement exceeds a predetermined threshold, and in that case, the output means is that the HW resource is a bottleneck. Output that
If the time required for the measurement is within a predetermined threshold , the HW bottleneck determination means determines that the bottleneck is not on the HW resource side, and in that case, the counting means counts the number of waiting processes.
The performance bottleneck analysis system according to claim 1. And measuring the required time obtaining means for calculating a time required for a plurality of times of measurement by said process condition measurement means,
First average value calculating means for calculating an average value of the measurement required time over the plurality of measurements;
Second average value calculating means for calculating the average number of processes as the number of waiting processes from the number of processes counted by the counting means;
HW bottleneck determination means for comparing the average measurement required time with a predetermined threshold value to determine whether the bottleneck is on the HW resource side;
Further comprising
The HW bottleneck determination means determines that the bottleneck is on the HW resource side if the average measurement time exceeds the predetermined threshold value, and in this case, the output means indicates that the HW resource is a bottleneck. Output that
The HW bottleneck determination means determines that the bottleneck is not on the HW resource side if the average measurement required time is within the predetermined threshold value , and in this case, the second average value calculation means determines the average number of processes. the calculated as the number of the waiting process,
The performance bottleneck analysis system according to claim 1.   5. The HW configuration change means for changing the number of HW resources of the computer system when the HW bottleneck determination means determines that the bottleneck is on the HW resource side. Performance bottleneck analysis system.   6. The performance bottleneck analysis system according to claim 1, further comprising a filtering unit that sorts and / or extracts the number of waiting processes based on a predetermined criterion. In a performance bottleneck analysis method for a computer system in which a plurality of processes use at least one hardware resource (hereinafter referred to as HW resource),
For each of the plurality of processes, waiting for acquisition of one of a plurality of software resources (hereinafter referred to as SW resources) that are in the HW resource use state in which the HW resource is being used or are mutually exclusive. A process state measurement step for measuring a process state indicating the SW resource waiting state and the type of SW resource waiting for the SW resource waiting process;
An underutilization status detection step of detecting an underutilization status of HW resources in which the number of processes in the HW resource utilization status is less than the number of HW resources;
A counting step of counting the number of SW resource waiting processes (hereinafter referred to as the number of waiting processes) for each of the plurality of SW resources when the HW resource unused state is detected;
An output step of outputting the number of waiting processes obtained for each SW resource as a performance bottleneck analysis result;
A performance bottleneck analysis method comprising: The process state measuring step measures the process state a plurality of times,
8. The performance bottleneck analysis method according to claim 7, further comprising an average value calculation step of calculating an average process number as the number of waiting processes from the number of processes counted in the counting step. And measuring the required time obtaining step of calculating the time required for measurement of the process conditions,
HW bottleneck determination step for comparing the time required for the measurement with a predetermined threshold value to determine whether the bottleneck is on the HW resource side ;
Further including
If the time required for the measurement exceeds a predetermined threshold value, the HW bottleneck determination step determines that the bottleneck is on the HW resource side, and in that case , the HW resource has a bottleneck in the output step. Is output,
If the time required for the measurement is within a predetermined threshold , the HW bottleneck determination step determines that the bottleneck is not on the HW resource side, and in that case , the number of waiting processes is counted in the counting step. ,
The performance bottleneck analysis method according to claim 7. A plurality of measurements required time obtaining step of calculating the time required for measurement in the process state measuring step,
A first average value calculating step for calculating an average value of the measurement required time over the plurality of measurements;
A second average value calculating step of calculating an average number of processes as the number of waiting processes from the number of processes counted in the counting step;
An HW bottleneck determination step for comparing the average measurement required time with a predetermined threshold to determine whether the bottleneck is on the HW resource side;
Further comprising
The HW bottleneck determination step determines that the bottleneck is on the HW resource side if the average required time exceeds the predetermined threshold value, and in that case , the output step causes a bottleneck in the HW resource. Is output,
The HW bottleneck determination step determines that the bottleneck is not on the HW resource side if the average measurement required time is within the predetermined threshold value . In this case , the second average value calculation step includes the average number of processes. the calculated as the number of the waiting process,
The performance bottleneck analysis method according to claim 7.   11. The HW configuration change step of changing the number of HW resources of the computer system when the HW bottleneck determination step determines that the bottleneck is on the HW resource side. Performance bottleneck analysis method.   The performance bottleneck analysis method according to any one of claims 7 to 11, further comprising a filtering step of sorting and / or extracting the number of waiting processes according to a predetermined criterion.

Images