Performance analysis apparatus and performance analysis method

Patent metadata

Field	Value
Publication number	US-11762562-B2
Application number	US-202117194538-A
Country	US
Kind code	B2
Filing date	Mar 8, 2021
Priority date	May 25, 2020
Publication date	Sep 19, 2023
Grant date	Sep 19, 2023

Abstract

The present invention aims to specify a bottlenecked resource more accurately in an IT system.
In a performance analysis apparatus which includes a processing device, a storage device and an input/output interface and analyzes a performance of an IT system, the processing device creates an overall performance prediction model by using a nonlinear regression algorithm with a first overall performance value being set as an objective variable and with a first resource performance value being set as an explanatory variable, creates an explanatory model by inputting the overall performance prediction model and using an XAI technology and calculates a degree of per-resource influence on the overall performance value which is output by inputting a second resource performance value into the explanatory model.

First claim

What is claimed is:

1. A performance analysis apparatus for use in a performance analysis of an IT system comprising:
a processing device;
a storage device; and
an input/output interface, wherein
the processing device
creates an overall performance prediction model by using a nonlinear regression algorithm with an overall performance value which is input via the input/output interface and which is measured from the IT system being set as an objective variable and with a first resource performance value which is measured simultaneously with measurement of the overall performance value and which is a value of a resource which configures the IT system being set as an explanatory variable,
creates an explanatory model by inputting the overall performance prediction model and using an XAI (Explainable Artificial Intelligence) technology, and
calculates a degree of per-resource influence on the overall performance value which is output by inputting a second resource performance value which is measured at a time which is different from the time that the first resource performance value of the resource is measured into the explanatory model.

2. The performance analysis apparatus according to claim 1 , wherein
the overall performance value is a response time of the IT system.

3. The performance analysis apparatus according to claim 2 , wherein
the IT system is a storage system, and the overall performance value is an I/O response time of an I/O request to the storage system or the number of the I/O requests which are left unanswered.

4. The performance analysis apparatus according to claim 1 , wherein
the processing device decides resources as bottlenecked resources in descending order of degrees of influence of the resources on the overall performance value which is measured.

5. The performance analysis apparatus according to claim 4 , wherein
the processing device includes an evaluation period acceptance unit and totalizes the frequency that the resource concerned is decided as the bottlenecked resource in the evaluation period per resource and decides the resource which is the largest in the number of times as the bottlenecked resource in the evaluation period.

6. The performance analysis apparatus according to claim 5 , wherein
the processing device sets only a time at which the overall performance value is particularly decreased in the evaluation period as a target for totalization and thereby performs a process of totalizing the bottlenecking frequency per resource.

7. The performance analysis apparatus according to claim 4 , wherein
the processing device gives accuracy which is set higher as a predicted-actual difference which is a difference between a predictive value of the overall performance value which is calculated by inputting the second resource performance value of the resource into the overall performance prediction model and an actually measured value of the overall performance value which is measured simultaneously with measurement of the second resource performance value becomes smaller to the per-resource degree of influence on the overall performance value and gives the accuracy which is set lower as the predicted-actual difference becomes larger to the degree of influence.

8. The performance analysis apparatus according to claim 7 , wherein
the processing device performs the process of totalizing the bottlenecking frequency per resource by excluding a time at which the accuracy is low.

9. The performance analysis apparatus according to claim 5 , wherein
in a value of the per-resource degree of influence on the overall performance value, in a case where a positive value that an absolute value of the degree of influence exceeds a threshold value and a negative value that the threshold value exceeds the absolute value are present in a mixed state, the processing device gives a low accuracy to the per-resource degree of influence on the overall performance value.

10. The performance analysis apparatus according to claim 9 , wherein
the threshold value which is set for the absolute value is an average value of the overall performance values which are measured simultaneously with measurement of the second resource performance values.

11. The performance analysis apparatus according to claim 1 , wherein
the processing device further includes an overall performance linear prediction model which is created by using a linear regression algorithm with a first resource performance value which is measured simultaneously with measurement of the overall performance value and which is the value of the resource which configures the IT system being set as the explanatory variable, and
in a case where a difference between the degree of influence on the overall performance value which is calculated by using the overall performance linear prediction model and the degree of influence on the overall performance value which is calculated by using the explanatory model is small, adopts the overall performance linear prediction model as the model.

12. A performance analysis method comprising:
creating an overall performance prediction model by using a nonlinear regression algorithm with an overall performance value which is input via an input/output interface and which is measured from an IT system being set as an objective variable and with a first resource performance value which is measured simultaneously with measurement of the overall performance value and which is a value of a resource which configures the IT system being set as an explanatory variable;
creating an explanatory model by inputting the overall performance prediction model and using an XAI (Explainable Artificial Intelligence) technology; and
calculating a degree of per-resource influence on the overall performance value which is output by inputting a second resource performance value which is measured at a time which is different from the time that the first resource performance value of the resource is measured into the explanatory model.

Assignees

• Hitachi Ltd

Inventors

• Masuda Mineyoshi
• Himura Yosuke
• Murayama Kouichi

Classifications

• G06Q10/04 Primary

  Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202) · CPC title

• G06F3/0613 Primary

  in relation to throughput · CPC title

• G06F11/3419

  by assessing time · CPC title

• G06F11/3447

  Performance evaluation by modeling · CPC title

• G06F3/0635

  by changing the path, e.g. traffic rerouting, path reconfiguration · CPC title