WO2012093469A1 - 性能評価装置及び性能評価方法 - Google Patents
性能評価装置及び性能評価方法 Download PDFInfo
- Publication number
- WO2012093469A1 WO2012093469A1 PCT/JP2011/007343 JP2011007343W WO2012093469A1 WO 2012093469 A1 WO2012093469 A1 WO 2012093469A1 JP 2011007343 W JP2011007343 W JP 2011007343W WO 2012093469 A1 WO2012093469 A1 WO 2012093469A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- workload
- composite
- performance
- model
- data
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
- G06F11/3414—Workload generation, e.g. scripts, playback
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3447—Performance evaluation by modeling
Definitions
- the present invention relates to a computer performance evaluation technique.
- resources can be used efficiently by operating virtual machines running on different hardware (computers) on the same hardware.
- the number of virtual machines executed on the same hardware is excessively limited, so that resources that can be originally used cannot be sufficiently utilized.
- Patent Document 1 proposes a technique for automatically and continuously executing a performance test of an information processing system under a plurality of different measurement conditions. Specifically, it has been proposed to perform a performance test of an information processing system based on performance test data generated by arbitrarily combining setting values of items having different measurement conditions.
- a performance test is only performed for each item to be measured (for example, for each computer resource), and a performance test combining a plurality of items is not considered.
- a performance test is only performed for each item to be measured (for example, for each computer resource), and a performance test combining a plurality of items is not considered.
- only one process using one computer resource is rarely executed on a computer (system), and a plurality of processes are executed in parallel. There are many cases.
- the performance data obtained by the above-described method may have low accuracy when actual operation is considered.
- the configuration of systems and virtual machines can be changed flexibly, and multiple virtual machines with various properties coexist on the same hardware, so the capacity (performance) of a computer can be estimated. very difficult.
- An object of the present invention is to provide a technique for evaluating the performance of a computer with high accuracy.
- the first aspect relates to a performance evaluation apparatus for evaluating the performance of an evaluation target computer.
- the performance evaluation apparatus sequentially generates a plurality of workloads having different sizes on each evaluation target computer for each of a plurality of single workload types, and at least two different single workload types
- the workload control means that sequentially generates multiple composite workloads combining different types of workloads corresponding to the composite workload types on the evaluation target computer, and the evaluation targets during each workload generation
- a data analyzer that generates each primary performance model for a workload type Comprising the, on the basis of the primary performance model generated by the obtained composite performance data and the data analysis means by the data acquisition means, and model generation means for generating a composite performance model for the complex workload species, a.
- the second aspect relates to a performance evaluation method for evaluating the performance of an evaluation target computer executed by a computer.
- a plurality of workloads having different sizes with respect to each of a plurality of single workload types are sequentially generated on the evaluation target computer.
- the next performance data is acquired, and multiple composite workloads combining different workloads corresponding to the composite workload type are sequentially placed on the evaluation target computer for the composite workload type combining at least two different single workload types
- Generating the primary performance model for each single workload type by acquiring the composite performance data of the evaluation target computer during the generation of the composite workload and modeling the acquired primary performance data Based on the acquired composite performance data and the generated primary performance model And generating a composite performance model for over de species.
- the third aspect is a performance evaluation program that causes a computer to realize each configuration included in the second aspect.
- a computer-readable recording medium recording the performance evaluation program according to the third aspect may be used.
- This recording medium includes a non-transitory tangible medium.
- FIG. 1 is a diagram conceptually illustrating a configuration example of a performance evaluation system in the first embodiment.
- FIG. 2 is a diagram conceptually illustrating a processing configuration example of the performance evaluation device and the server device in the first embodiment.
- FIG. 3 is a diagram conceptually illustrating a processing configuration example of the performance evaluation device and the server device in the second embodiment.
- FIG. 4 is a diagram illustrating an example of a workload plan file of a single workload type related to the CPU.
- FIG. 5 is a diagram illustrating an example of a workload plan file of a single workload type related to network transmission.
- FIG. 6 is a diagram illustrating an example of a workload plan file of the secondary composite workload type.
- FIG. 7 is a flowchart showing an operation example of the performance evaluation system in the second embodiment.
- FIG. 8 is a diagram conceptually illustrating a processing configuration example of the server apparatus in the third embodiment.
- FIG. 1 is a diagram conceptually illustrating a configuration example of a performance evaluation system in the first embodiment.
- the performance evaluation system 1 according to the first embodiment includes a performance evaluation apparatus 10 and a server apparatus 20 that is a target of performance evaluation.
- the performance evaluation device 10 evaluates the performance of the server device 20 regarding a predetermined computer resource.
- the performance evaluation device 10 evaluates performance related to a predetermined computer resource by generating a performance model.
- the performance evaluation apparatus 10 includes a RAM (Random Access Memory) 12, a ROM (Read Only Memory, not shown), a memory such as a hard disk (HDD) 13, a CPU (Central Processing Unit) 11, An output interface 14 and the like are included. These hardware elements are connected by a bus 15, for example.
- the input / output interface 14 includes a network interface that enables communication of a predetermined communication method between the performance evaluation device 10 and the server device 20.
- the hardware configuration of the server device 20 is the same as that of the performance evaluation device 10.
- the present embodiment does not limit the hardware configurations of the performance evaluation device 10 and the server device 20.
- the computer that is the target of performance evaluation is not limited to the server device, but may be a computer.
- the performance evaluation apparatus 10 and the server apparatus 20 each implement the processing units illustrated in FIG. 2 by executing a program stored in a memory by the CPU 11.
- FIG. 2 is a diagram conceptually illustrating a processing configuration example of the performance evaluation device 10 and the server device 20 in the first embodiment.
- the performance evaluation apparatus 10 includes a workload control unit 101, a data acquisition unit 102, a data analysis unit 103, a model generation unit 104, and the like.
- the workload control unit 101 sequentially generates a plurality of workloads having different sizes on each evaluation target computer for each of a plurality of single workload types, and combines at least two different single workload types.
- a plurality of composite workloads which are combinations of different workloads corresponding to the composite workload types, are sequentially generated on the evaluation target computer.
- the workload means a processing load using a certain computer resource, such as a CPU load, a network transmission load, a network reception load, a hard disk reading load, a hard disk writing load, an image processing load, and the like.
- the single workload type means the type of processing load (workload).
- this type of floating-point operation corresponds to a single workload type.
- a test image display process is used as a graphic-related workload, the type of test image display process corresponds to a single workload type.
- a single workload type indicating a floating-point operation
- a plurality of floating-point operations with different calculation amounts or calculation loads are prepared, and each of these floating-point operations corresponds to each workload.
- the composite workload type means a type in which the workload types of two different single workloads executed in parallel on the server device 20 to be evaluated are combined.
- a combination of two different single workload types of floating point arithmetic and network transmission processing corresponds to one composite workload type.
- a composite workload type composed of floating point arithmetic and network transmission processing a combination of a floating point arithmetic having a certain amount of computation and a network transmission processing having a certain amount of transmission is each composite workload. It corresponds to.
- the data acquisition unit 102 acquires the primary performance data of the server device 20 in which each workload is generated and the composite performance data of the server device 20 in which the composite workload is generated.
- the primary performance data means data indicating the operating status of a predetermined computer resource. For example, CPU usage rate, CPU usage amount, network transmission amount, network reception amount, hard disk read amount, hard disk write amount, etc. Any one or a combination of any of the above.
- the data analysis unit 103 generates the primary performance model for each single workload type by modeling the primary performance data acquired by the data acquisition unit 102.
- This modeling for example, a well-known statistical analysis method or the like is used.
- the present embodiment does not limit this modeling method.
- the model generation unit 104 generates a composite performance model for the composite workload type based on the composite performance data acquired by the data acquisition unit 102 and each primary performance model generated by the data analysis unit 103.
- This composite performance model indicates characteristics of performance data of the server device 20 when a combination of at least two workloads corresponding to the composite workload type is executed.
- the server device 20 includes a workload generation unit 201, a data measurement unit 202, and the like.
- the workload generation unit 201 executes each workload and composite workload at different timings under the control of the workload control unit 101 of the performance evaluation apparatus 10. In the execution of a composite workload, each workload that constitutes the composite workload is executed in parallel.
- the workload generation unit 201 may execute a program (task, process, etc.) corresponding to each workload, which is stored in the server device 20 in advance. Alternatively, the program may be held for each single workload type, and the workload generation unit 201 may apply the parameter sent from the workload control unit 101 to the program and execute the parameter.
- a program task, process, etc.
- the data measuring unit 202 measures each primary performance data during the generation of each workload and composite performance data during the generation of the composite workload.
- the measured primary performance data and composite performance data are sent to the performance evaluation apparatus 10.
- each workload is executed in the server device 20 according to the control of the performance evaluation device 10, and the primary performance data of the server device 20 in which each workload is generated is measured and measured. Each primary performance data is sent to the performance evaluation device 10.
- the composite workload is executed, the composite performance data of the server device 20 in which the composite workload is generated is measured, and the composite performance data is sent to the performance evaluation device 10.
- each primary performance model for each single workload type is generated, and based on each generated primary performance model and composite performance data.
- a composite performance model is generated for the composite workload type.
- the performance model composite performance model
- the performance under actual operation of the server device 20 can be accurately evaluated.
- the performance evaluation system 1 in the second embodiment has the same configuration as that of the first embodiment shown in FIG.
- the performance evaluation system 1 according to the second embodiment will be described focusing on the content different from the first embodiment.
- FIG. 3 is a diagram conceptually illustrating a processing configuration example of the performance evaluation device 10 and the server device 20 in the second embodiment.
- the server apparatus 20 is the same as that of 1st Embodiment, description of the server apparatus 20 is abbreviate
- the performance evaluation apparatus 10 in the second embodiment further includes a model estimation unit 105, a performance data storage unit 106, and a performance model storage unit 107 in addition to the configuration of the first embodiment.
- Each of these new processing units is also realized, for example, by the CPU 11 executing a program stored in the memory.
- the processing units other than the new processing units will be described below mainly on the contents different from the first embodiment.
- the workload control unit 101 acquires information about the generated workload when generating the workload and the composite workload. This information is obtained from, for example, a workload plan file as shown in FIGS. These workload plan files may be stored in the performance evaluation device 10 or may be acquired from other devices. Each workload plan file stores information necessary to generate a single workload type workload on the server device 20.
- FIG. 4 is a diagram showing an example of a workload plan file of a single workload type related to the CPU.
- FIG. 5 is a diagram illustrating an example of a workload plan file of a single workload type related to network transmission.
- a workload plan file of a single workload type information for identifying a single workload type, the number of divisions, an occurrence time, and a workload upper limit value (W_Thres_CPU, W_Thres_NWSEND) The communication destination address and the like are stored.
- the workload upper limit value is the minimum workload value that maximizes the performance of computer resources corresponding to a single workload type.
- the workload upper limit value of a single workload type related to the CPU is the minimum workload value at which the CPU usage rate is maximized (eg, 100%).
- the workload upper limit value of a single workload type related to network transmission is, for example, the minimum workload value at which the bit rate is maximized.
- the workload value indicates the size of the workload. For example, if the workload is a floating point operation, the workload value may be set to the number of floating point operations per unit time. If the workload is network transmission processing, the workload value may be set to the amount of transmission data per unit time.
- each workload value (V_CPU (i), Each workload having V_NWSEND (k)) is generated on the server device 20.
- V_CPU (i) i ⁇ (W_Thres_CPU / D) (Formula 1)
- V_NWSEND (k) k ⁇ (W_Thres_NWSEND / D) (Formula 2)
- the workload control unit 101 is a server that processes 21 (0 ⁇ i ⁇ 20) workloads each having a different workload value for each single workload type for the CPU for 60 seconds. Generated on the device 20. Thereafter, the workload control unit 101 generates 21 (0 ⁇ k ⁇ 20) workloads having different workload values for the single workload type related to network transmission on the server device 20 for 60 seconds.
- the workload control unit 101 generates a desired workload by sending a workload execution instruction including information for identifying a single workload type, the number of divisions, an occurrence time, and a workload upper limit value to the server device 20. Also good. 4 and 5, each workload value is obtained by dividing the workload upper limit value by a predetermined number of divisions. However, a desired number of workload values are set in advance. May be.
- the workload control unit 101 When the workload control unit 101 needs to execute a program corresponding to the workload generated on the server device 20 on another computer, the workload control unit 101 instructs the other computer to execute the program. For example, when a workload related to network reception is generated on the server device 20, a program for transmitting data to the server device 20 on another computer is required. In such a case, the workload control unit 101 controls the data transmission program to be executed on another computer before or simultaneously with generating a workload related to network reception on the server device 20.
- the workload control unit 101 may instruct the data measurement unit 202 of the server device 20 to start measurement.
- the workload control unit 101 instructs the data measurement unit 202 to start measurement of predetermined performance data before or simultaneously with the workload generation unit 201 executing the workload or the composite workload.
- FIG. 6 is a diagram showing an example of a workload plan file of the secondary composite workload type.
- the composite workload type will be referred to as a K-th order composite workload type according to the number (K) of the single workload types to be configured.
- a composite workload in which K workloads are combined is referred to as a K-th order composite workload.
- a workload plan of a secondary composite workload type in which a single workload type related to the CPU and a single workload type related to network transmission are combined is set.
- the workload plan file for a composite workload type stores information that identifies the composite workload type, the number of divisions, the occurrence time, the communication destination address, and the workload upper limit value for each single workload type that constitutes the information. .
- the workload control unit 101 sets the number of composite workloads generated for a composite workload type to a number smaller than the total number of combinations of workloads generated for each single workload type constituting the composite workload type. Set.
- the number (22) subtracted here is a combination in which i is 0 or k is 0 in (Expression 1) and (Expression 2), and these are single workloads and not composite workloads.
- the number of composite workloads generated for a composite workload type is particularly limited as long as it is less than the total number of workloads generated for each single workload type that constitutes the composite workload type. Not. By reducing the number of complex workloads, the present embodiment can reduce the number of measurements and reduce the measurement time.
- the data acquisition unit 102 may acquire performance data corresponding to the type of performance to be evaluated. For example, when it is desired to finally evaluate the CPU performance, the data acquisition unit 102 may acquire performance data indicating the CPU performance. In addition, when it is desired to finally evaluate the network transmission performance, the data acquisition unit 102 may acquire performance data indicating the network transmission performance.
- the performance data storage unit 106 stores each performance data acquired by the data acquisition unit 102.
- the performance model storage unit 107 includes a primary performance model generated by the data analysis unit 103, a K-th order performance model generated by the model generation unit 104 (K is 2 or more), and an approximate composite estimated by the model estimation unit 105. Stores performance models, etc.
- Each data stored in the performance data storage unit 106 and the performance model storage unit 107 is appropriately extracted and used in each processing unit.
- each processing unit that uses each data stored in the performance data storage unit 106 and the performance model storage unit 107 from the performance data storage unit 106 and the performance model storage unit 107, It is not specified that it is extracted.
- the model estimation unit 105 uses each primary performance model corresponding to each single workload type that constitutes the secondary composite workload type among the primary performance models generated by the data analysis unit 103, Estimate an approximate composite performance model for the next composite workload type.
- the model estimation unit 105 estimates the approximate composite performance model, for example, by adding the primary performance models.
- M2CPU_CPU_NWSEND ⁇ K (M1CPU_CPU (WL_CPU) + M1CPU_NWSEND (WL_NWSEND))
- WL_CPU is a variable indicating the workload value of the CPU workload
- WL_NWSEND is a variable indicating the workload value of the network transmission workload.
- M1CPU_CPU is a CPU performance model for a single workload type of CPU.
- M1CPU_NWSEND is a CPU performance model for a single workload type of network transmission.
- K indicates a function that returns the maximum CPU performance value when the argument value exceeds the maximum CPU performance value.
- the approximate composite performance model estimated here is a combination of the primary performance models, and therefore, the competing factors due to the simultaneous generation of two types of workloads are not considered. Therefore, there is an error between the approximate composite performance model and the actual composite performance model.
- the model estimation unit 105 selects (K ⁇ 1) different workload types generated by the model generation unit 104 when K (K is an integer of 3 or more) single workload types is used. Using the (K-1) th order performance model for the combined (K-1) th order composite workload type, an approximate composite performance model for the Kth order composite workload type combining the K different workload types is estimated. To do. The model estimation unit 105 generates an approximate composite performance model for the K-th order composite workload type, for example, by adding the (K-1) th order performance models.
- M2CPU_CPU_NWSEND indicates a secondary performance model for a secondary composite workload type related to a combination of a single workload type of CPU and a single workload type of network transmission.
- M2CPU_CPU_HDDWT represents a secondary performance model for a secondary composite workload type related to a combination of a single workload type of CPU and a single workload type of hard disk writing.
- M2CPU_NWSEND_HDDWT indicates a secondary performance model for a secondary composite workload type related to a combination of a single workload type for network transmission and a single workload type for hard disk writing.
- K indicates a function that returns the maximum CPU performance value when the argument value exceeds the maximum CPU performance value.
- each workload value is divided by each of the two is because of reducing duplicate performance elements is there.
- each variable is divided by (M ⁇ 1).
- the data analysis unit 103 models the difference data between the composite performance data acquired by the data acquisition unit 102 and the approximate composite performance data obtained from the approximate composite performance model estimated by the model estimation unit 105, thereby combining the composite data. Generate a differential data model for the workload type.
- the difference data model generated here corresponds to an error between the approximate composite performance model and the actual composite performance model.
- the model generation unit 104 generates a composite performance model for a composite workload type using the approximate composite performance model estimated by the model estimation unit 105 and the differential data model generated by the data analysis unit 103.
- the model generation unit 104 generates a composite performance model for a composite workload type, for example, by adding together the approximate composite performance model and the difference data model.
- a composite performance model M2CPU_CPU_NWSEND () for a secondary composite workload type for evaluating CPU performance can be expressed by the following equation.
- a secondary composite workload type in which a single workload type of CPU and a single workload type of network transmission are combined is taken as an example.
- M2CPU_CPU_NWSEND (WL_CPU, WL_NWSEND) M2CPU_CPU_NWSEND to (WL_CPU, WL_NWSEND) + M2CPU_DIFF_CPU_NWSEND (WL_CPU, WL_NWSEND)
- M2CPU_DIFF_CPU_NWSEND (WL_CPU, WL_NWSEND) in the above formula is a difference data model for the secondary composite workload type generated by the data analysis unit 103 and combining a single workload type of CPU and a single workload type of network transmission. Indicates.
- FIG. 7 is a flowchart showing an operation example of the performance evaluation system 1 in the second embodiment.
- an operation when the CPU performance of the server device 20 is an evaluation target will be described as an example.
- the computer resources of the server device 20 to be evaluated may be set in advance in the performance evaluation device 10 or may be input by the user.
- the following operation example is applicable not only when the CPU performance is an evaluation target, but also when the performance related to other computer resources, that is, the network performance, the hard disk access performance, the graphic performance, or the like is an evaluation target.
- the workload control unit 101 acquires information on a single workload type to be used when generating a performance model. For example, the workload control unit 101 acquires the information from the workload plan file. When three single workload types relating to CPU, network transmission, and hard disk writing are used, the workload control unit 101 sets 3 to the number M of single workload types to be used, and the variable K is set to 1 (S1, S2).
- the workload control unit 101 selects one single workload type from among the M single workload types, and acquires the workload plan information for the selected single workload type (S3).
- the workload plan information is acquired from the workload plan file as shown in FIGS.
- the workload control unit 101 instructs the server device 20 to generate a plurality of workloads related to the selected single workload type based on the acquired workload plan information.
- This instruction may include the workload plan information.
- the workload generation unit 201 receives the instruction and executes the workload according to the instruction (S4).
- the CPU workload with the workload value (V_CPU (0)) is executed for 60 seconds, and then the CPU workload with the workload value (V_CPU (1)) is executed for 60 seconds.
- each workload corresponding to each of the 21 workload values is sequentially executed for 60 seconds.
- the data measuring unit 202 measures performance data (primary performance data) indicating the CPU performance of the server device 20.
- the CPU usage rate is measured as the primary performance data.
- the data measuring unit 202 sends the measured primary performance data to the performance evaluation apparatus 10.
- the data acquisition unit 102 acquires primary performance data sent from the data measurement unit 202 of the server device 20.
- the primary performance data acquired here indicates the performance during the generation of each workload related to the selected single workload type.
- the data analyzing unit 103 generates a primary performance model for the selected single workload type by modeling each primary performance data acquired by the data acquiring unit 102 (S5; YES). , S6).
- a polynomial regression method is used for this modeling.
- the primary performance model generated in this case is a polynomial, that is, a polynomial regression model expressed by a value indicating performance (objective variable) and a workload value (explanatory variable) affecting the performance model.
- the workload control unit 101 has completed generation of the primary performance model for a certain single workload type, but has not completed processing for all the single workload types (S9; NO).
- the single workload type is selected, and the workload plan information is acquired for the selected single workload type (S3).
- a second order approximate composite in which a primary performance model related to CPU and a primary performance model related to network transmission are added together.
- a secondary approximate composite performance model in which a performance model, a primary performance model related to a CPU, and a primary performance model related to hard disk writing are added together, and a primary performance model related to network transmission and a primary performance model related to hard disk writing The added second-order approximate composite performance model is estimated.
- the workload control unit 101 acquires the workload plan information related to the secondary composite workload type (S3).
- the workload plan information is acquired from the workload plan file as shown in FIG.
- the workload control unit 101 instructs the server device 20 to generate a secondary composite workload according to the acquired plan information.
- the workload generation unit 201 executes a secondary composite workload in response to the instruction (S4).
- the workload generation unit 201 executes two workloads corresponding to the composite workload type indicated by the instruction in parallel.
- the data measuring unit 202 measures performance data (composite performance data) during generation of the secondary composite workload in the server device 20.
- the measured composite performance data during the generation of the secondary composite workload is acquired by the data acquisition unit 102 of the performance evaluation apparatus 10.
- the data analysis unit 103 adds the workload value of each workload constituting the composite workload corresponding to the acquired composite performance data to the secondary approximate composite performance model already estimated by the model estimation unit 105.
- each approximate composite performance data is acquired.
- the data analysis unit 103 generates a differential data model by modeling the differential data between the composite performance data acquired by the data acquisition unit 102 and the acquired approximate composite performance data (S5; NO, S7). .
- a polynomial regression method is used as described above.
- the model generation unit 104 adds the approximate composite performance model already estimated by the model estimation unit 105 and the differential data model generated by the data analysis unit 103 to thereby add the composite performance for the secondary composite workload type.
- a model is generated (S8).
- the addition of the models is, for example, addition of polynomials.
- the model estimator 105 either 3 in M C 2 pieces of secondary performance model A third-order approximate composite performance model is estimated by adding the individual combinations (S11). When M is 3 and K is 2, one third-order approximate composite performance model is generated.
- the combined performance for a secondary composite workload type combining each single workload type for CPU and network transmission Model, a composite performance model for a secondary composite workload type combined with each CPU and hard disk write single workload type, and a combined secondary model with a network transmit and hard disk write single workload type
- a third order approximate composite performance model obtained by adding the composite performance model for the composite workload type is estimated.
- the model estimation unit 105 constructs a third-order approximate composite performance model
- 1 is added to the variable K.
- the workload control unit 101 acquires workload plan information related to a tertiary composite workload type (S3).
- processing is performed in the same manner as the processing of the secondary composite workload type (S4, S5; NO, S7), and the composite performance model for the tertiary composite workload type is used by using the above-mentioned tertiary approximate composite performance model Is generated (S8).
- each primary performance model for a plurality of single workload types is generated. Specifically, according to the control of the workload control unit 101, a plurality of workloads for each single workload type are sequentially executed on the server device 20 by the workload generation unit 201 for each predetermined time, and each workload is being generated. Performance data of the server device 20 is measured by the data measuring unit 202.
- the measured performance data is acquired by the data acquisition unit 102, and the data analysis unit 103 models a plurality of performance data for each single workload type, whereby a primary performance model for each single workload type is obtained. Is generated.
- each primary performance model is generated based on a plurality of performance data corresponding to a plurality of workloads of the same type and different sizes.
- the greater the number of workloads (number of samples) generated with respect to a single workload type the more detailed performance data is acquired, so the accuracy of the generated primary performance model is also increased.
- the composite workload for the composite performance model is a combination of at least two different workloads. Therefore, if the number of samples of a single workload type to be combined is increased, the number of samples of the composite workload will dramatically increase. To increase. That is, a long measurement time is usually spent to generate a highly accurate composite performance model.
- a composite performance model is efficiently generated without degrading accuracy using a plurality of primary performance models generated as described above.
- the number of samples for the primary performance model as described above may be set to a normally set value that does not deteriorate the accuracy of the primary performance model.
- a K-th order performance model is generated by adding the K-th order approximate composite performance model and the difference data model (model generation unit 104).
- the K-th order approximate composite performance model is estimated by addition of each of the K combinations in the plurality of (K-1) th order performance models (model estimation unit 105).
- the K-th order approximate composite performance model does not take into account the competing factors due to the simultaneous generation of K workloads, and therefore there is an error with the true K-th order performance model.
- the difference data model generated by the data analysis unit 103 is used to fill this error.
- the difference data model is a plurality of differences between the performance data on the server device 20 during the generation of the Kth order composite workload measured by the data measuring unit 202 and the approximate composite data obtained from the Kth order approximate composite performance model. Data is generated by being modeled by the data analysis unit 103.
- the substantial modeling process executed to generate the Kth-order performance model is a process for constructing a differential data model.
- the complexity of the differential data model is smaller than the true Kth order performance model. Therefore, according to the second embodiment, the number of samples for generating the differential data model, that is, the number of samples of the composite workload can be reduced. This is because in order to model data showing a more complicated distribution, measurement with more measurement points is necessary.
- a K-th order composite performance model can be efficiently generated by reducing the number of measurements.
- the accuracy of the performance model can be maintained even if the number of measurements is reduced. This is because the complexity of the modeling target is reduced by converting the composite performance modeling for the composite workload into a differential value modeling between the approximate performance model and the true composite performance model. .
- FIG. 8 is a diagram conceptually illustrating a processing configuration example of the server device 20 in the third embodiment.
- the server device 20 in the third embodiment newly includes a plurality of virtual machines 260 (# 1) and 260 (# 2) and a server control unit 250 in addition to the configurations of the first embodiment and the second embodiment.
- a general method may be used as a method for realizing the virtual machine.
- the server control unit 250 controls the virtual machines 260 (# 1) and 260 (# 2), and responds to the instruction content from the workload control unit 101 of the performance evaluation device 10.
- the workload generation units 201 (# 1) and 201 (# 2) are realized on the virtual machines 260 (# 1) and 260 (# 2).
- the workload generation units 201 (# 1) and 201 (# 2) are similar to the workloads on the virtual machines 260 (# 1) and 260 (# 2), respectively, as in the first and second embodiments. Generate complex workloads.
- the data measurement unit 202 measures the performance data of each virtual machine 260 (# 1) and 260 (# 2) and the performance data of the entire server device 20, respectively. Thereby, the data acquisition unit 102 of the performance evaluation apparatus 10 acquires the performance data of each virtual machine 260 (# 1) and 260 (# 2) and the entire performance data.
- the processing configuration of the performance evaluation apparatus 10 may be the same as the configuration of the first embodiment and the second embodiment.
- the third embodiment when different virtual machines are generated even for the same type of workload, they are handled as different workloads. That is, when two single workload types of CPU and network transmission are used, the CPU workload and network transmission workload generated on the virtual machine 260 (# 1), the virtual machine 260 (# 2) It is treated as if there were four single workload types: CPU workload generated above and network transmission workload.
- the performance data of each virtual machine 260 (# 1) and 260 (# 2) acquired by the data acquisition unit 102 and the entire performance data are used.
- the generation method of the primary performance model and the composite performance model is the same as that in the first embodiment and the second embodiment.
- a performance model for each virtual machine can be efficiently generated without degrading accuracy.
- the performance model is gradually generated in order from the composite workload type with a small number of combined workload types. This is to increase the efficiency without degrading the accuracy of the performance model by reducing the complexity of the differential data model for generating the next performance model.
- the following performance model generation method may be used.
- a method of directly generating a final third-order or higher performance model using the first-order performance model may be used. This method estimates the Kth order approximate composite performance model by adding the K primary performance models, measures the composite performance data during the generation of the Kth order composite workload, and generates a differential data model Then, the Kth order composite performance model is generated by adding the Kth order approximate composite performance model and the difference data model.
- the competing factors generated when generating a second-order composite workload and the competing factors generated when generating a third-order compound workload are considered. Therefore, the difference data model is complicated compared to the above embodiments. Therefore, it is preferable to increase the number of samples for obtaining the differential data model as compared with the above embodiments.
- the processing amount and the measurement time can be reduced as compared with the above embodiments.
- a third order approximate composite performance model may be generated by adding a second order performance model and a first order performance model.
- the differential data model is more complicated than in the above embodiments.
- a composite workload type in which a plurality of workloads having different sizes with respect to each of a plurality of single workload types are sequentially generated on the evaluation target computer, and at least two different single workload types are combined.
- a workload control means for sequentially generating a plurality of composite workloads combining different types of workloads corresponding to the composite workload types on the evaluation target computer, and primary performance data of the evaluation target computer during the generation of each workload, And, by modeling the data acquisition means for acquiring the composite performance data of the evaluation target computer during the generation of the composite workload, and the primary performance data acquired by the data acquisition means, the primary for each single workload type is modeled.
- Data analysis means for generating performance models and data acquisition means Based on the primary performance model generated by more acquired composite performance data and data analyzing means, performance evaluation apparatus characterized by comprising: a model generation means for generating a composite performance model for the complex workload species, a.
- the said workload control means is more than the total number of the combination of the workload each generate
- the model generation means includes model estimation means for estimating an approximate composite performance model for a composite workload type using at least one of the primary performance models generated by the data analysis means, and the data
- the analysis unit models a composite workload by modeling difference data between the composite performance data acquired by the data acquisition unit and the approximate composite performance data obtained from the approximate composite performance model estimated by the model estimation unit.
- a differential data model for the seed, and the model generation means uses the approximate composite performance model estimated by the model estimation means and the differential data model generated by the data analysis means to The performance evaluation according to appendix 1 or 2, characterized by generating a model Location.
- the model estimation means is a (K-1) -order composite work that combines (K-1) (K is an integer of 3 or more) different workload types generated by the model generation means.
- Appendix 3 characterized by estimating an approximate composite performance model for a K-th order composite workload type by combining K different workload types using a (K-1) th order performance model for the load type Performance evaluation equipment.
- the workload control means generates all the workloads for all the single workload types, and then executes the composite workloads in order from the composite workload type with the smallest number of combined workload types.
- the performance evaluation apparatus according to any one of appendices 1 to 4, wherein the performance evaluation apparatus is generated.
- the computer sequentially generates a plurality of workloads having different sizes on each of the plurality of single workload types on the evaluation target computer, and the primary performance data of the evaluation target computer during the generation of each workload is obtained. Acquire multiple composite workloads that combine different types of workloads corresponding to the composite workload type on the target computer, and then generate multiple composite workloads that are combined with at least two different single workload types.
- a primary performance model for each single workload type is generated and acquired.
- composite workflow Generating a composite performance model for the species, the performance evaluation method comprising.
- the number of composite workloads generated for a composite workload type should be set to be smaller than the total number of combinations of workloads generated for each single workload type constituting the composite workload type.
- the composite performance model is generated by estimating an approximate composite performance model for a composite workload type using at least one of the generated primary performance models, and the obtained composite performance data By modeling the difference data with the approximate composite performance data obtained from the above approximate composite performance model, a differential data model for the composite workload type is generated, and the estimated approximate composite performance model and the generated differential data are generated.
- the above approximate composite performance model is estimated by combining (K-1) generated different workload types (K-1) and the following composite workload types
- the method for evaluating performance according to appendix 8 further comprising: estimating an approximate composite performance model for a K-th order composite workload type by combining K different workload types using a (K-1) th order performance model for .
- a workload control means for sequentially generating a plurality of composite workloads combining different types of workloads corresponding to the composite workload types on the evaluation target computer with respect to the workload type, and a primary of the evaluation target computer that is generating each workload
- Data acquisition means for acquiring performance data and composite performance data of a computer under evaluation during generation of the composite workload, and primary performance data acquired by the data acquisition means are modeled to obtain each single workload.
- a model generation means for generating a composite performance model for the composite workload type based on the composite performance data acquired by the data acquisition means and the primary performance model generated by the data analysis means.
- a performance evaluation program characterized by that.
- the said workload control means is more than the total number of the combination of the workload each generate
- the model generation means includes model estimation means for estimating an approximate composite performance model for a composite workload type using at least one of the primary performance models generated by the data analysis means,
- the data analysis means models the difference data between the composite performance data acquired by the data acquisition means and the approximate composite performance data obtained from the approximate composite performance model estimated by the model estimation means.
- a differential data model for the load type is generated, and the model generation unit uses the approximate composite performance model estimated by the model estimation unit and the differential data model generated by the data analysis unit to generate a composite data for the composite workload type.
- a performance model is generated, according to appendix 11 or 12, Performance evaluation program.
- the model estimation means includes a (K-1) -order composite work that combines (K-1) (K is an integer of 3 or more) different workload types generated by the model generation means. Item 14. The approximate composite performance model for the Kth order composite workload type obtained by combining the K different workload types is estimated using the (K-1) th order performance model for the load type. Performance evaluation program.
- the workload control means generates all the workloads for all of the single workload types, and then generates the composite workloads in order from the composite workload type having a small number of combined workload types. 15.
- the performance evaluation program according to any one of appendices 11 to 14, wherein the performance evaluation program is generated.
- a performance evaluation system comprising: the evaluation target computer including: primary performance data during generation of each workload; and data measurement means for measuring composite performance data during generation of the composite workload.
Landscapes
- Engineering & Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Debugging And Monitoring (AREA)
Abstract
Description
図1は、第1実施形態における性能評価システムの構成例を概念的に示す図である。第1実施形態における性能評価システム1は、図1に示すように、性能評価装置10と、性能評価の対象となるサーバ装置20とを有する。
第1実施形態では、性能評価装置10の制御に応じて、サーバ装置20において、各ワークロードが実行され、各ワークロード発生中のサーバ装置20の1次性能データがそれぞれ計測され、計測された各1次性能データがそれぞれ性能評価装置10に送られる。同様に、複合ワークロードが実行された場合には、その複合ワークロード発生中のサーバ装置20の複合性能データが計測され、その複合性能データが性能評価装置10に送られる。
第2実施形態における性能評価システム1は、図1に示す上述の第1形態と同様の構成を有する。以下、第2実施形態における性能評価システム1について第1実施形態と異なる内容を中心に説明する。
V_CPU(i)=i×(W_Thres_CPU/D) (式1)
V_NWSEND(k)=k×(W_Thres_NWSEND/D) (式2)
M2CPU_CPU_NWSEND~(WL_CPU、WL_NWSEND)
=K(M1CPU_CPU(WL_CPU)+M1CPU_NWSEND(WL_NWSEND))
M3CPU_CPU_NWSEND_HDDWT~(WL_CPU、WL_NWSEND、WL_HDDWT)
=K(M2CPU_CPU_NWSEND(WL_CPU/2、WL_NWSEND/2)+M2CPU_CPU_HDDWT(WL_CPU/2、WL_HDDWT/2)+M2CPU_NWSEND_HDDWT(WL_NWSEND/2、WL_HDDWT/2))
M2CPU_CPU_NWSEND(WL_CPU、WL_NWSEND)
=M2CPU_CPU_NWSEND~(WL_CPU、WL_NWSEND)+M2CPU_DIFF_CPU_NWSEND(WL_CPU、WL_NWSEND)
図7は、第2実施形態における性能評価システム1の動作例を示すフローチャートである。以下の説明では、サーバ装置20のCPU性能を評価対象とする場合の動作を例に挙げる。この評価対象とすべきサーバ装置20のコンピュータ資源については、予め性能評価装置10に設定されていてもよいし、ユーザにより入力されるようにしてもよい。以下の動作例は、CPU性能を評価対象とする場合のみならず、他のコンピュータ資源に関する性能、即ち、ネットワーク性能、ハードディスクアクセス性能、グラフィック性能等を評価対象とする場合にも適用可能である。
上述のように、第2実施形態における性能評価システム1では、まず、複数の単一ワークロード種に対する各1次性能モデルが生成される。具体的には、ワークロード制御部101の制御に応じて、単一ワークロード種毎の複数ワークロードがワークロード生成部201によってサーバ装置20上に所定時間ずつ順次実行され、各ワークロード発生中のサーバ装置20の性能データがデータ計測部202によってそれぞれ計測される。
第3実施形態では、上述の各実施形態における評価対象であるサーバ装置20が仮想マシン環境を提供可能である場合の例を挙げる。以下、第3実施形態について、上述の各実施形態と異なる内容を中心に説明する。
上述の各実施形態は、組み合わされるワークロード種の数が少ない複合ワークロード種から順に、漸次的に性能モデルを生成した。これは、各次の性能モデルを生成するための差分データモデルの複雑さを低減することにより、性能モデルの精度を劣化させることなく効率を上げるためであった。
Claims (7)
- 複数の単一ワークロード種の各々に関し大きさの異なる複数のワークロードを評価対象コンピュータ上に順次発生させ、かつ、少なくとも2つの異なる単一ワークロード種を組み合わせた複合ワークロード種に関し該複合ワークロード種に対応する異種のワークロードを組み合わせた複数の複合ワークロードを該評価対象コンピュータ上に順次発生させるワークロード制御手段と、
前記各ワークロード発生中の前記評価対象コンピュータの1次性能データ、及び、前記複合ワークロード発生中の前記評価対象コンピュータの複合性能データを取得するデータ取得手段と、
前記データ取得手段により取得された前記1次性能データをモデル化することにより、前記各単一ワークロード種に対する1次性能モデルをそれぞれ生成するデータ解析手段と、
前記データ取得手段により取得された前記複合性能データ及び前記データ解析手段により生成された前記1次性能モデルに基づいて、前記複合ワークロード種に対する複合性能モデルを生成するモデル生成手段と、
を備えることを特徴とする性能評価装置。 - 前記ワークロード制御手段は、前記複合ワークロード種に関し発生させる複合ワークロードの数を、前記複合ワークロード種を構成する各単一ワークロード種に関しそれぞれ発生させるワークロードの全組み合わせ数よりも少ない数に設定する、
ことを特徴とする請求項1に記載の性能評価装置。 - 前記モデル生成手段は、
前記データ解析手段により生成された前記1次性能モデルの少なくとも1つを用いて、前記複合ワークロード種に対する近似複合性能モデルを推定するモデル推定手段、
を含み、
前記データ解析手段は、前記データ取得手段により取得された前記複合性能データと、前記モデル推定手段により推定された前記近似複合性能モデルから得られる近似複合性能データとの差分データをモデル化することにより、前記複合ワークロード種に対する差分データモデルを生成し、
前記モデル生成手段は、前記モデル推定手段により推定された前記近似複合性能モデル及び前記データ解析手段により生成された前記差分データモデルを用いて、前記複合ワークロード種に対する複合性能モデルを生成する、
ことを特徴とする請求項1又は2に記載の性能評価装置。 - 前記モデル推定手段は、前記モデル生成手段により生成された、(K-1)個(Kは3以上の整数)の異なるワークロード種を組み合わせた(K-1)次の複合ワークロード種に対する(K-1)次性能モデルを用いて、K個の異なるワークロード種を組み合わせたK次の複合ワークロード種に対する近似複合性能モデルを推定する、
ことを特徴とする請求項3に記載の性能評価装置。 - 前記ワークロード制御手段は、前記各単一ワークロード種の全てに関しそれぞれ全てのワークロードを発生させた後、組み合わされるワークロード種の数が少ない複合ワークロード種から順に前記複合ワークロードを発生させることを特徴とする請求項1から4のいずれか1項に記載の性能評価装置。
- コンピュータが、
複数の単一ワークロード種の各々に関し大きさの異なる複数のワークロードを評価対象コンピュータ上に順次発生させ、
前記各ワークロード発生中の前記評価対象コンピュータの1次性能データを取得し、
少なくとも2つの異なる単一ワークロード種を組み合わせた複合ワークロード種に関し該複合ワークロード種に対応する異種のワークロードを組み合わせた複数の複合ワークロードを該評価対象コンピュータ上に順次発生させ、
前記複合ワークロード発生中の前記評価対象コンピュータの複合性能データを取得し、
取得された前記1次性能データをモデル化することにより、前記各単一ワークロード種に対する1次性能モデルをそれぞれ生成し、
取得された前記複合性能データ及び生成された前記1次性能モデルに基づいて、前記複合ワークロード種に対する複合性能モデルを生成する、
ことを含む性能評価方法。 - コンピュータに、
複数の単一ワークロード種の各々に関し大きさの異なる複数のワークロードを評価対象コンピュータ上に順次発生させ、かつ、少なくとも2つの異なる単一ワークロード種を組み合わせた複合ワークロード種に関し該複合ワークロード種に対応する異種のワークロードを組み合わせた複数の複合ワークロードを該評価対象コンピュータ上に順次発生させるワークロード制御手段と、
前記各ワークロード発生中の前記評価対象コンピュータの1次性能データ、及び、前記複合ワークロード発生中の前記評価対象コンピュータの複合性能データを取得するデータ取得手段と、
前記データ取得手段により取得された前記1次性能データをモデル化することにより、前記各単一ワークロード種に対する1次性能モデルをそれぞれ生成するデータ解析手段と、
前記データ取得手段により取得された前記複合性能データ及び前記データ解析手段により生成された前記1次性能モデルに基づいて、前記複合ワークロード種に対する複合性能モデルを生成するモデル生成手段と、
を実現させることを特徴とする性能評価プログラム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/978,412 US9529688B2 (en) | 2011-01-06 | 2011-12-28 | Performance evaluation device and performance evaluation method |
JP2012551765A JPWO2012093469A1 (ja) | 2011-01-06 | 2011-12-28 | 性能評価装置及び性能評価方法 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2011001467 | 2011-01-06 | ||
JP2011-001467 | 2011-01-06 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2012093469A1 true WO2012093469A1 (ja) | 2012-07-12 |
Family
ID=46457337
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2011/007343 WO2012093469A1 (ja) | 2011-01-06 | 2011-12-28 | 性能評価装置及び性能評価方法 |
Country Status (3)
Country | Link |
---|---|
US (1) | US9529688B2 (ja) |
JP (1) | JPWO2012093469A1 (ja) |
WO (1) | WO2012093469A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016115065A (ja) * | 2014-12-12 | 2016-06-23 | 日本電気株式会社 | 情報処理装置、情報処理システム、タスク処理方法、及び、プログラム |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2015184818A (ja) * | 2014-03-20 | 2015-10-22 | 株式会社東芝 | サーバ、モデル適用可否判定方法およびコンピュータプログラム |
US10198307B2 (en) * | 2016-03-31 | 2019-02-05 | Netapp, Inc. | Techniques for dynamic selection of solutions to storage cluster system trouble events |
US10552413B2 (en) * | 2016-05-09 | 2020-02-04 | Sap Se | Database workload capture and replay |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007323245A (ja) * | 2006-05-31 | 2007-12-13 | Nec Corp | 計算機システムおよび性能計測方法ならびに管理サーバ装置 |
JP2009123174A (ja) * | 2007-11-19 | 2009-06-04 | Mitsubishi Electric Corp | 仮想マシンサーバサイジング装置及び仮想マシンサーバサイジング方法及び仮想マシンサーバサイジングプログラム |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3987333B2 (ja) | 2001-12-06 | 2007-10-10 | 新日鉄ソリューションズ株式会社 | 性能測定装置、性能測定システム、記録媒体およびプログラム |
CA2697965C (en) * | 2007-08-31 | 2018-06-12 | Cirba Inc. | Method and system for evaluating virtualized environments |
US7650259B2 (en) * | 2007-10-01 | 2010-01-19 | International Business Machines Corporation | Method for tuning chipset parameters to achieve optimal performance under varying workload types |
-
2011
- 2011-12-28 WO PCT/JP2011/007343 patent/WO2012093469A1/ja active Application Filing
- 2011-12-28 JP JP2012551765A patent/JPWO2012093469A1/ja active Pending
- 2011-12-28 US US13/978,412 patent/US9529688B2/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007323245A (ja) * | 2006-05-31 | 2007-12-13 | Nec Corp | 計算機システムおよび性能計測方法ならびに管理サーバ装置 |
JP2009123174A (ja) * | 2007-11-19 | 2009-06-04 | Mitsubishi Electric Corp | 仮想マシンサーバサイジング装置及び仮想マシンサーバサイジング方法及び仮想マシンサーバサイジングプログラム |
Non-Patent Citations (1)
Title |
---|
MITSURU YANAGISAWA ET AL.: "Proposal of method of estimating load of virtual server using performance characteristic information", DAI 72 KAI (HEISEI 22 NEN) ZENKOKU TAIKAI KOEN RONBUNSHU (1), INFORMATION PROCESSING SOCIETY OF JAPAN, 8 March 2010 (2010-03-08), pages 1-291 - 1-292 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016115065A (ja) * | 2014-12-12 | 2016-06-23 | 日本電気株式会社 | 情報処理装置、情報処理システム、タスク処理方法、及び、プログラム |
Also Published As
Publication number | Publication date |
---|---|
US9529688B2 (en) | 2016-12-27 |
JPWO2012093469A1 (ja) | 2014-06-09 |
US20130282337A1 (en) | 2013-10-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9003416B2 (en) | Predicting resource requirements for a computer application | |
US20160321037A1 (en) | Performance verification device for verifying performance of program, method, and recording medium having program recorded thereon for causing computer to perform the method | |
US9645909B2 (en) | Operation management apparatus and operation management method | |
US20090055834A1 (en) | Virtualization planning system | |
DE102019003851A1 (de) | Systeme und Verfahren zum automatischen Realisieren von Modellen zu Co-Simulation | |
US20210152424A1 (en) | Selecting and using a cloud-based hardware accelerator | |
US10282272B2 (en) | Operation management apparatus and operation management method | |
US8606905B1 (en) | Automated determination of system scalability and scalability constraint factors | |
US20170279697A1 (en) | Control device for estimation of power consumption and energy efficiency of application containers | |
JPWO2011105091A1 (ja) | 制御装置、管理装置、制御装置のデータ処理方法、およびプログラム | |
US20200151082A1 (en) | Managing cloud-based hardware accelerators | |
Willnecker et al. | Comparing the accuracy of resource demand measurement and estimation techniques | |
JP2010225133A (ja) | 消費電力評価装置、電力係数作成システム、消費電力評価方法及び電力係数作成方法。 | |
WO2012093469A1 (ja) | 性能評価装置及び性能評価方法 | |
US20190361746A1 (en) | Selecting hardware accelerators based on score | |
Westermann et al. | Efficient experiment selection in automated software performance evaluations | |
WO2015146100A1 (ja) | 負荷推定システム、情報処理装置、負荷推定方法、及び、プログラムを記憶する記憶媒体 | |
Pozin et al. | Models in performance testing | |
US20110191094A1 (en) | System and method to evaluate and size relative system performance | |
Brebner | Automatic performance modelling from application performance management (APM) data: an experience report | |
CN113056725B (zh) | 信息处理系统以及信息处理方法 | |
JP5871018B2 (ja) | 性能予測装置、性能モデル生成方法およびプログラム | |
US20120198063A1 (en) | Virtual server system, autonomous control server thereof, and data processing method and computer program thereof | |
JP2015022322A (ja) | 仮想マシン管理装置、仮想マシン管理方法、及びプログラム | |
CN115098405B (zh) | 软件产品的测评方法、装置、电子设备及存储介质 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 11855028 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2012551765 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13978412 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 11855028 Country of ref document: EP Kind code of ref document: A1 |