Memory leak analysis by usage trends correlation

Patent metadata

Field	Value
Publication number	US-9454454-B2
Application number	US-201414481687-A
Country	US
Kind code	B2
Filing date	Sep 9, 2014
Priority date	Sep 9, 2014
Publication date	Sep 27, 2016
Grant date	Sep 27, 2016

Abstract

Tools and techniques assist developers with the detection of memory leaks by using correlation of data type memory usage trends. In particular, investigations of memory leaks can be prioritized without always resorting to the use of bulky and performance-degrading memory dumps, by using these tools and techniques to identify leaky correlated data types. Data about a program's memory usage is processed to identify memory usage trends over time for respective data types, and the trends are searched for significant correlations. Correlated trends (and hence their corresponding data types) are grouped. Memory usage analysis information is displayed for grouped data types, such as the names of the most rapidly leaking data types, the names of correlated data types, leak rates, and leak amounts in terms of memory size and/or data object counts. Memory usage data may also be correlated with processing load requests to indicate which requests have associated memory leaks.

First claim

What is claimed is:

1. A computational process for improving the functioning of a computer by assisting detection of memory leaks in a software program which has a processing load and which uses memory, the process comprising:
(a) obtaining memory usage data which includes memory usage samples which collectively specify a plurality of data types, each memory usage sample specifying a value of at least one usage statistic for at least one of the data types at a specified sample time;
(b) computationally identifying respective memory usage trends over time for a plurality of the data types, by processing at least part of the memory usage data with a processor;
(c) computationally searching for correlations between data type memory usage trends, wherein the searching comprises computing statistical measures of distances between pairs or other tuples of the data type memory usage trends;
(d) computationally grouping data types into memory-usage-trend-correlated groups based on the computed statistical measures of distances between pairs or other tuples of the data type memory usage trends, such that all data types in a given memory-usage-trend-correlated group have memory usage trends that satisfy a predetermined trend correlation criterion, and data types whose trends do not satisfy the predetermined trend correlation criterion are not in the given memory-usage-trend-correlated group;
(e) utilizing a result of the grouping step;
(f) computationally searching for correlations between data type memory usage trends and one or more processing load request trends; and
(g) presenting to a user a report which includes information, based on said searching, about a correlation between data type memory usage and processing load requests.

2. The computational process of claim 1 , wherein the predetermined trend correlation criterion specifies that trends are correlated when, and only when, they have a correlation coefficient beyond a predetermined correlation threshold.

3. The computational process of claim 1 , wherein usage statistics for each of the data types include at least one of the following:
a count indicating how many items of the data type were allocated;
a count indicating how many items of the data type were deallocated;
a memory size indicating how much memory was designated as allocated to items of the data type;
a memory size indicating how much memory designated as allocated to items of the data type was deallocated.

4. The computational process of claim 1 , wherein computationally identifying respective memory usage trends includes at least one of the following:
performing simple linear regression using least squares to fit usage statistics to a linear trend;
performing a segmented regression to fit usage statistics to a piecewise linear trend;
performing a nonlinear regression to fit usage statistics to a nonlinear trend.

5. The computational process of claim 1 , wherein computationally searching for correlations between data type memory usage trends includes at least one of the following:
determining a Pearson product-moment correlation coefficient for data type memory usage trends;
determining a Spearman's rank correlation coefficient for data type memory usage trends;
determining a Kendall rank correlation coefficient for data type memory usage trends;
determining a distance correlation for data type memory usage trends;
determining a distance covariance for data type memory usage trends;
determining a Brownian distance covariance for data type memory usage trends.

6. The computational process of claim 1 , wherein computationally identifying respective memory usage trends includes normalizing usage statistics for the processing load.

7. The computational process of claim 1 , wherein computationally grouping data types into memory-usage-trend-correlated groups includes grouping two data types which have trends that have a correlation coefficient above a predetermined threshold even though the trends have different growth rates.

8. The computational process of claim 1 , wherein utilizing a result of the grouping step includes at least one of the following:
displaying data type names of data types which are in a group identified to a user as having the most memory leakage;
displaying data type names of user-defined data types which are in a group identified as having the most memory leakage of the user-defined data types;
displaying an estimated leak size for a particular data type;
displaying an estimated leak rate in memory size per unit of time for a particular data type;
displaying an estimated leaked item count for a particular data type;
displaying an estimated leak rate in item count per unit of time for a particular data type.

9. The computational process of claim 1 , further comprising identifying a group of data types which all showed little or no memory leakage, then showed a spurt of leaks, and then again showed little or no memory leakage.

10. A computer-readable storage medium configured with data and with instructions that when executed by at least one processor causes the processor(s) to perform a technical process for assisting detection of memory leaks in software which has a processing load and which uses memory, the technical process comprising the steps of:
(a) obtaining memory usage data which includes memory usage samples which collectively specify a plurality of data types, each memory usage sample specifying a value of at least one usage statistic for at least one of the data types at a specified sample time;
(b) computationally identifying respective memory usage trends over time for a plurality of the data types, by processing at least part of the memory usage data with a processor;
(c) computationally searching for correlations between data type memory usage trends and one or more processing load request trends; and
(d) presenting to a user a report which includes information, based on said searching, about a correlation between data type memory usage and processing load requests.

11. The configured storage medium of claim 10 , wherein the technical process further comprises restarting a software process in response to a result of said searching for correlations between data type memory usage trends and processing load request trends.

12. The configured storage medium of claim 10 , wherein usage statistics for each of the data types include at least one of the following:
a count indicating how many items of the data type were allocated;
a memory size indicating how much memory was designated as allocated to items of the data type.

13. The configured storage medium of claim 10 , wherein computationally searching for correlations between data type memory usage trends and processing load request trends includes at least one of the following:
determining a Pearson product-moment correlation coefficient;
determining a Spearman's rank correlation coefficient;
determining a Kendall rank correlation coefficient;
determining a distance correlation.

14. The configured storage medium of claim 10 , wherein the technical process further comprises at least two of the following:
displaying an estimated leak size for a particular data type;
displaying an estimated leak rate in memory size per unit of time for a particular data type;
displaying an estimated leaked item count for a particular data type;
displaying an estimated leak rate in item count per unit of time for a particular data type;
restarting a software process in response to an estimated leaked item count;
restarting a software process in response to an estimated leak rate.

15.

Assignees

• Microsoft Corp
• Microsoft Technology Licensing Llc

Inventors

• Abraham Arun Mathew
• Crawford Brian Robert
• Vann Daniel
• Fan Jing
• Rosen Douglas Jay

Classifications

• G06F11/3037

  where the computing system component is a memory, e.g. virtual memory, cache (accessing, addressing or allocating within memory systems or architectures G06F12/00; checking stores for correct operation G11C29/00) · CPC title

• G06F11/3452 Primary

  Performance evaluation by statistical analysis · CPC title

• G06F9/5016

  the resource being the memory · CPC title

• G06F11/3612 Primary

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

• G06F12/0253

  Garbage collection, i.e. reclamation of unreferenced memory · CPC title