FREQUENCY-DOMAIN ANALYSIS OF DATA-CENTER OPERATIONAL AND PERFORMANCE METRICS

• US 20170310556A1
• Filed: 04/25/2016
• Published: 10/26/2017
• Est. Priority Date: 04/25/2016
• Status: Active Grant

• Current Assignee: Vmware LLC (Broadcom, Inc.)
• Original Assignee: VMware, Inc. (Broadcom, Inc.)
• Inventors: Knowles, Christopher, Crukley, Jeffery
• Granted Patent: US 11,245,593 B2
• Time in Patent Office: Days
• CPC Class Codes: H04L 41/142 using statistical or mathematical methods
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]
• CPC Class Codes: H04L 67/131 Protocols for games, networked simulations or virtual reality

Abstract

The current document is directed to methods and systems for frequency-domain analysis of operational and performance metric values and other data generated and collected within computer systems, including large distributed computer systems and virtualized data centers. In one implementation, each set of time-ordered values for each metric in a set of metrics is partitioned into time intervals, transformed from the time domain to the frequency domain, and aligned to generate a metric surface in a frequency-time-amplitude space. The metric surfaces are then pairwise compared to identify related metrics. Transfer functions are generated for transforming metric surfaces into one another. The comparison values and transfer functions are used to produce graphs that encapsulate discovered relationships between metrics.

Claims

1. An automated subsystem of a computer system, the automated subsystem comprising:

one or more processors;

one or more memories;

one or more mass-storage devices; and

computer instructions stored in one or more of the one or more memories that, when retrieved from memory and executed by one or more of the one or more processors, control the automated subsystem to

identify time-ordered metric-value data stored within the computer system,

partition a total time spanning the earliest-in-time metric value to the latest-in-time metric value into multiple time intervals,

for each time interval, transform the metric-value data associated with times within the time interval from the time domain to the frequency domain to generate corresponding frequency-domain data for the time interval,

assemble the frequency-domain data for the multiple time intervals into a metric surface within a frequency-amplitude-time space, and

store, in one or more of the one or more memories and mass-storage devices, a representation of the metric surface in association with an indication of a metric with respect to which the time-ordered metric-value data was generated.

2. The automated subsystem of claim 1 wherein the time-ordered metric-value data is a set of representations of time/metric-value data points that represent sampled values of the metric selected from among a performance metric, an operational-characteristic metric, and an operational-behavior metric for the computer system over a time period of length equal to the total time.

3. The automated subsystem of claim 2 wherein the metric is related to the performance, operational characteristics, or operational behavior of one of:

the computer system;

a single component or subcomponent of the computer system; and

an aggregation of multiple components and/or subcomponents of the computer system.

4. The automated subsystem of claim 2 wherein the metric is related to values computed from one or more sets of time-ordered metric values related to the performance, operational characteristics, or operational behavior of one of:

the computer system;

a single component or subcomponent of the computer system; and

an aggregation of multiple components and/or subcomponents of the computer system.

5. The automated subsystem of claim 2 wherein the computer system is one of:

a personal computer;

a server;

a workstation;

a multi-computer-system data center;

a distributed computer system;

a virtual data center;

a geographically distributed computer system; and

and aggregation of geographically distributed data centers.

6. The automated subsystem of claim 1 wherein the metric-value data associated with times within the time intervals are transformed from the time domain to the frequency domain by application of a discrete Fourier transform.

7. The automated subsystem of claim 1

wherein the metric surface within the frequency-amplitude-time space is a discrete set of points or a continuous surface generated from a discrete set of points;

wherein the frequency-amplitude-time space is defined by frequency, time, and amplitude axes;

wherein the frequency-domain data for each of the multiple time intervals are arranged as amplitude-time slices at successive positions along the time coordinate axis, the amplitude-frequency data points for each amplitude-time slice representing a cross-section of the metric surface; and

wherein the amplitude axis represents amplitude or squared amplitude values.

8. An automated subsystem of a computer system, the automated subsystem comprising:

one or more processors;

one or more memories;

one or more mass-storage devices; and

computer instructions stored in one or more of the one or more memories that, when retrieved from memory and executed by one or more of the one or more processors, control the automated subsystem to

identify two metric surfaces stored within the computer system;

compare the two metric surfaces to generate a comparison-metric value and at least one transfer function that, when applied to a first metric surface of the two metric surfaces, transforms the first metric surface into the other of the two metric surfaces; and

store a representation of the metric surface in association with an indication of the metric in one or more of the one or more memories and mass-storage devices.

9. The automated subsystem of claim 8

wherein the metric surface within the frequency-amplitude-time space is a discrete set of points or a continuous surface generated from a discrete set of points;

wherein the frequency-amplitude-time space is defined by frequency, time, and amplitude axes;

wherein the frequency-domain data for each of the multiple time intervals are arranged as amplitude-time slices at successive positions along the time coordinate axis, the amplitude-frequency data points for each amplitude-time slice representing a cross-section of the metric surface; and

wherein the amplitude axis represents amplitude or squared amplitude values.

10. The automated subsystem of claim 9 wherein two metric surfaces are compared to generate a comparison-metric value by:

aligning the two metric surfaces in space by applying one or more operations to one or both of the metric surfaces; and

computing the comparison metric from the aligned metric surfaces.

11. The automated subsystem of claim 10 wherein the operations applied to one or both of the metric surfaces are selected from among:

a translation operation that translates a metric surface in a direction of the frequency axis;

a translation operation that translates a metric surface in a direction of the time axis;

a translation operation that translates a metric surface in a direction of the amplitude axis;

a resealing operation that rescales a metric surface, and the frequency-time-amplitude space that includes it, with respect to the frequency axis;

a resealing operation that rescales a metric surface, and the frequency-time-amplitude space that includes it, with respect to the time axis;

a resealing operation that rescales a metric surface, and the frequency-time-amplitude space that includes it, with respect to the amplitude axis;

a rotation operation that rotates a metric surface with respect to the frequency axis;

a rotation operation that rotates a metric surface with respect to the time axis;

a rotation operation that rotates a metric surface with respect to the amplitude axis; and

trimming and cutting operations.

12. The automated subsystem of claim 10 wherein each metric surface is generated from time-ordered metric-value data by

partitioning a total time spanning the earliest-in-time metric value to the latest-in-time metric value into multiple time intervals;

for each time interval, transforming the metric-value data associated with times within the time interval from the time domain to the frequency domain to generate corresponding frequency-domain data for the time interval,

assembling the frequency-domain data for the multiple time intervals into a metric surface within a frequency-amplitude-time space, and

storing, in one or more of the one or more memories and mass-storage devices, a representation of the metric surface in association with an indication of a metric with respect to which the time-ordered metric-value data was generated.

13. The automated subsystem of claim 10 wherein computing the comparison metric from the aligned metric surfaces further comprises:

identifying features within each of the aligned metric surfaces;

identifying one or more pairs of related features, a first member of each pair occurring within a first metric surface of the aligned metric surfaces and a second member of each pair occurring within a second metric surface of the aligned metric surfaces;

computing differences between the features of one or more of the one or more pairs of related features; and

generating the comparison metric to be inversely proportional to the computed differences.

14. The automated subsystem of claim 13 wherein the features are selected from among peaks and valleys of the metric surfaces.

15. The automated subsystem of claim 10 wherein computing the comparison metric from the aligned metric surfaces further comprises:

determining the volume lying between the two aligned metric surfaces; and

generating the comparison metric to be inversely proportional to the determined volume.

16. The automated subsystem of claim 10 wherein the at least one transfer function is generated from the operations applied to one or both of the two aligned metric surfaces to align the two metric surfaces in space.

17. The automated subsystem of claim 9 further comprising:

18. The automated subsystem of claim 8 further including using comparison metrics and transfer functions generated from multiple sets of time-ordered metric-value data collected for multiple components of the computer system to construct a component graph in which nodes represent the components and each edge is labeled with one or both transfer functions for the metric surfaces associated with the two components represented by the two nodes connected by the edge.

19. The automated subsystem of claim 18 further including combining multiple component graphs into a composite component graph.

20. Computer instructions stored in a physical data-storage device within a computer system having one or more processors, one or more memories, and one or more mass-storage devices that, when retrieved from the physical data-storage device and executed by one or more of the one or more processors, control the computer system to

identify time-ordered metric-value data stored within the computer system,

partition a total time spanning the earliest-in-time metric value to the latest-in-time metric value into multiple time intervals,

for each time interval, transform the metric-value data associated with times within the time interval from the time domain to the frequency domain to generate corresponding frequency-domain data for the time interval,

assemble the frequency-domain data for the multiple time intervals into a metric surface within a frequency-amplitude-time space, and

store, in one or more of the one or more memories and mass-storage devices, a representation of the metric surface in association with an indication of a metric with respect to which the time-ordered metric-value data was generated.