System and method for assigning performance defects to software processing components

Patent metadata

Field	Value
Publication number	US-12505377-B2
Application number	US-202117236892-A
Country	US
Kind code	B2
Filing date	Apr 21, 2021
Priority date	Apr 21, 2021
Publication date	Dec 23, 2025
Grant date	Dec 23, 2025

Abstract

Embodiments of the present disclosure relate to a method, a system, and a non-transitory machine-readable medium for assigning performance defects to software processing components. Provided is a method that receives performance data of a set of performance runs of a software application across a set of software processing components of a file software system, where the performance data comprises multiple labeled performance features that are associated with the set of software processing components; receives, from a software processing component expert, a user-selection of a subset of labeled performance features of the several labeled performance features; and trains a machine learning (ML) model to determine whether one or more of the set of software processing components is associated with performance regressions of the software application, using the subset of labeled performance features of the performance data as training data.

First claim

What is claimed is:

1 . A method comprising:
performing a set of performance runs of a software application across a set of software processing components of a file system, wherein the set of performance runs comprises a performance regression; receiving performance data of the set of performance runs, wherein the performance data comprises a plurality of labeled performance features that are associated with the set of software processing components; receiving, from a user device, a user-selection of a subset of labeled performance features of the plurality of labeled performance features via a user interface displayed on the user device, wherein the user interface includes the plurality of labeled performance features; and training a machine learning (ML) model to determine whether one or more of the set of software processing components of the file system is associated with performance defects of the software application, using the subset of labeled performance features of the performance data as training data, and assign the one or more of the set of software processing components of the file system to one or more users when the ML model determines that the one or more of the set of software processing components of the file system is associated with the performance defects.

2 . The method of claim 1 , wherein at least one of the set of performance runs of the software application is a performance defect in which one or more software processing components of the set of software processing components performed below a performance threshold.

3 . The method of claim 2 , wherein the subset of labeled performance features are associated with a subset of the one or more software processing components that are responsible for the performance defect.

4 . The method of claim 1 further comprising generating synthetic performance data using the subset of labeled performance features, wherein the training data further comprises the synthetic performance data.

5 . The method of claim 4 , wherein the user-selection of the subset of labeled performance features reduces the performance data used as training data by removing data associated with the unselected labeled performance features, wherein the synthetic performance data includes more data of the subset of labeled performance features than the reduced performance data.

6 . The method of claim 4 further comprising
identifying labeled performance features of the subset from a performance run of the set of performance runs that is not a performance defect of the software application,
wherein generating the synthetic performance data comprises varying each of the identified labeled performance features from the performance run within a predefined range.

7 . The method of claim 1 , wherein the ML model is one of a decision tree, a random forest, a recurrent neural network (RNN), a long short term memory (LSTM) neural network, and a multi-head attention.

8 . The method of claim 1 , wherein at least one labeled performance feature indicates a usage of a resource of an electronic device by a software processing component during a performance run of the software application.

9 . A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations, the operations comprising:
performing a set of performance runs of a software application across a set of software processing components of a file system, wherein the set of performance runs comprises a performance regression; receiving performance data of the set of performance runs, wherein the performance data comprises a plurality of labeled performance features that are associated with the set of software processing components; receiving, from a user device, a user-selection of a subset of the labeled performance features of the plurality of labeled performance features via a user interface displayed on the user device, wherein the user interface includes the plurality of labeled performance features; and training a machine learning (ML) model to determine whether one or more of the set of software processing components of the file system is associated with performance defects of the software application, using the subset of labeled performance features of the performance data as training data, and assign the one or more of the set of software processing components of the file system to one or more users when the ML model determines that the one or more of the set of software processing components of the file system is associated with the performance defects.

10 . The non-transitory machine-readable medium of claim 9 , wherein at least one of the set of performance runs of the software application is a performance defect in which one or more software processing components of the set of software processing components performed below a performance threshold.

11 . The non-transitory machine-readable medium of claim 10 , wherein the subset of labeled performance features are associated with a subset of the one or more software processing components that are responsible for the performance defect.

12 . The non-transitory machine-readable medium of claim 9 , wherein the operations further comprise generating synthetic performance data using the subset of labeled performance features, wherein the training data further comprises the synthetic performance data.

13 . The non-transitory machine-readable medium of claim 12 , wherein the user-selection of the subset of labeled performance features reduces the performance data used as training data by removing data associated with the unselected labeled performance features, wherein the synthetic performance data includes more data of the subset of labeled performance features than the reduced performance data.

14 . The non-transitory machine-readable medium of claim 12 , wherein the operations further comprise:
identifying labeled performance features of the subset from a performance run of the set of performance runs that is not a performance defect of the software application, wherein generating the synthetic performance data comprises varying each of the identified labeled performance features from the performance run within a predefined range.

15 . The non-transitory machine-readable medium of claim 9 , wherein the ML model is one of a decision tree, a random forest, a recurrent neural network (RNN), a long short term memory (LSTM) neural network, and a multi-head attention.

16 . The non-transitory machine-readable medium of claim 9 , wherein at least one labeled performance feature indicates a usage of a resource of an electronic device by a software processing component during a performance run of the software application.

17 . A method comprising:
receiving performance data of a performance defect of a software application across a set of software processing components of a file system; and determining, using a machine learning (ML) model that has an input based on the performance data of the performance defect, a software processing component of the set of software processing components that has a performance defect, wherein the performance data includes a plurality of labeled performance features that are associated with the set of software processing components, wherein the ML model has been trained, using a subset of the labeled performance features that have been user-selected, via a user interface displayed on a user device, from the plurality of labeled performance features as training data, to determine whether one or

Assignees

• Emc Ip Holding Co Llc

Inventors

• Nguyen Tuan Anh
• Lakshmikantha Shashikanth
• Shilane Philip

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06F11/3612

  by runtime analysis (performance monitoring G06F11/3466) · CPC title

• G06F16/1734

  Details of monitoring file system events, e.g. by the use of hooks, filter drivers, logs · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F11/3447

  Performance evaluation by modeling · CPC title