Predicting failures with code flow tensors

Patent metadata

Field	Value
Publication number	US-11631003-B2
Application number	US-202016916735-A
Country	US
Kind code	B2
Filing date	Jun 30, 2020
Priority date	Jun 30, 2020
Publication date	Apr 18, 2023
Grant date	Apr 18, 2023

Abstract

Techniques for predicting states may include: receiving data sets of counter values, wherein each counter values denotes a number of times a particular code flow point associated with the counter value is executed at runtime during a specified time period; receiving images generated from the data sets; labeling each of the images with state information, wherein first state information associated with a first image indicates that the first image is associated with a first error state of a system or an application; training a neural network using the images to recognize the first state; receiving a next image generated from another data set; and predicting, by the neural network and in accordance with the next image, whether the system or the application is expected to transition into the first state. In at least one embodiment, the foregoing processing may optionally use matrices generated from the data sets rather than images.

First claim

What is claimed is:

1. A method of predicting states comprising:
receiving a plurality of data sets, wherein each of the plurality of data sets includes a plurality of counter values, wherein each of the plurality of counter values in each of the plurality of data sets denotes a number of times a particular code flow point associated with said each counter value is executed at runtime during a specified time period;
receiving a plurality of images generated from the plurality of data sets, wherein each of the plurality of data sets is used to generate a different one of the plurality of images;
labeling each of the plurality of images with state information, wherein first state information associated with a first image of the plurality of images indicates that the first image is associated with a first error state of a system or an application;
training a neural network using the plurality of images, wherein said training includes training the neural network to recognize the first error state;
subsequent to said training, receiving a next image generated from another data set including a plurality of counter values each denoting a second number of times a particular code flow point associated with said each counter value is executed at runtime; and
predicting, by the neural network and in accordance with the next image, whether the system or the application is expected to transition into the first error state, and wherein the first state information, that is associated with a first image, includes a time interval denoting a first amount of time prior to the first error state, and wherein the first image is generated from a first data set acquired the first amount of time prior to an occurrence of the first error state in the system or the application.

2. The method of claim 1 , further comprising:
providing the next image as an input to the neural network; and
responsive to providing the next image as an input to the neural network, generating by the neural network, a first output value corresponding to a probability indicating a likelihood that the system or the application subsequently transitions into the first error state.

3. The method of claim 2 , further comprising:
determining whether the first output value is greater than a threshold; and
responsive to determining the first output value is greater than the threshold, predicting that the system or the application is expected to transition into the first error state, and otherwise predicting that the system or the application is not expected to transition into the first error state.

4. The method of claim 1 , wherein the first image is generated from a first of the plurality of data sets and the method further comprises:
receiving the first data set;
waiting a specified amount of time for an occurrence of one of a plurality of defined error states, wherein the first error state is included in the plurality of defined error states;
receiving notification regarding a first occurrence of the first error state at a first point in time, wherein the first data set is acquired the first amount of time prior to the first occurrence of the first error state in the system or the application; and
responsive to said notification, labeling the first data set and the first image with the first state information.

5. The method of claim 1 , wherein each of the plurality of images is a gray scale image.

6. The method of claim 1 , wherein each of the plurality of images is a color image denoting a heat map of counter values included in a particular one of the plurality of data sets used to generate said each image.

7. The method of claim 1 , wherein the system is a data storage system.

8. The method of claim 1 , wherein each of the plurality of data sets is acquired at a different point in time.

9. The method of claim 1 , wherein the first image is generated from a first data set of the plurality of data sets, wherein the first image is correlated with the first error state and wherein the first image includes pixels representing the plurality of counter values of the first data set.

10. The method of claim 9 , wherein the first image is included in a first time sequence of images corresponding to states of the system or the application at different time intervals prior to the system or the application transitioning into the first error state.

11. A method of predicting states comprising:
receiving a plurality of data sets, wherein each of the plurality of data sets includes a plurality of counter values, wherein each of the plurality of counter values in each of the plurality of data sets denotes a number of times a particular code flow point associated with said each counter value is executed at runtime during a specified time period;
receiving a plurality of images generated from the plurality of data sets, wherein each of the plurality of data sets is used to generate a different one of the plurality of images;
labeling each of the plurality of images with state information, wherein first state information associated with a first image of the plurality of images indicates that the first image is associated with a first error state of a system or an application;
training a neural network using the plurality of images, wherein said training includes training the neural network to recognize the first error state;
subsequent to said training, receiving a next image generated from another data set including a plurality of counter values each denoting a second number of times a particular code flow point associated with said each counter value is executed at runtime; and
predicting, by the neural network and in accordance with the next image, whether the system or the application is expected to transition into the first error state, and wherein the neural network is a first neural network that is assigned an active role at a first point in time, and wherein a second neural network is assigned an idle role at the first point in time, and wherein the active role assigned to the first neural network indicates the assigned first neural network is in a non- learning mode and the first neural network is used to predict a subsequent state of the system or the application based on newly acquired data sets, and wherein the idle role assigned to the second neural network indicates the second neural network is in a learning mode and the newly acquired data sets are used to generate first images used to train the second neural network.

12. The method of claim 11 , wherein the second neural network that is assigned the idle role and that is in the learning mode has one or more internal weights adjusted responsive to receiving at least some of the first images as input.

13. The method of claim 12 , wherein, at a second point in time subsequent to the first point in time, the first neural network transitions from the active role to the idle role indicating that the first neural network is in the learning mode, and wherein at the second point in time, the second neural network transitions from the idle role to the active role and indicates that the second neural network is in the non-learning mode.

14. The method of claim 13 , wherein, subsequent to the second point in time, the second neural network is used to predict a subsequent state of the system or the application based on second newly acquired data sets, and wherein subsequent to the second point in time, the second newly acquired data sets are used to generate second images used to train the first neural network.

15. The method of claim 14 , wherein while the first neural network is assigned the idle role and is in the lear

Assignees

• Emc Ip Holding Co Llc

Inventors

• Martin Owen
• Randolph Benjamin A
• Romano Scott J
• Zaitoun Hani
• Sanap Abhilash

Classifications

• G06T11/26

  Drawing of charts or graphs · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06F8/433

  Dependency analysis; Data or control flow analysis · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title