Automated identification of performance crisis
AssigneeMicrosoft Corporation
InventorsMoises Goldszmidt, Peter Bodik
Life of this patent
Estimating the term…
Abstract
Methods for automatically identifying and classifying a crisis state occurring in a system having a plurality of computer resources. Signals are received from a device that collects the signals from each computer resource in the system. For each epoch, an epoch fingerprint is generated. Upon detecting a performance crisis within the system, a crisis fingerprint is generated consisting of at least one epoch fingerprint. The technology is able to identify that a performance crisis has previously occurred within the datacenter if a generated crisis fingerprint favorably matches any of the model crisis fingerprints stored in a database. The technology may also predict that a crisis is about to occur.
Claims20
A method for automatically identifying a crisis type occurring in a system having a plurality of computer resources, comprising
- summarizing a plurality of signals collected from each computer resource in the system during an epoch;
- generating an epoch fingerprint based on summarizing the plurality of signals collected from each computer resource in the system during an epoch,
- upon detecting a performance crisis within the system, generating a crisis fingerprint consisting of at least one epoch fingerprint;
- comparing the generated crisis fingerprint against one or more crisis fingerprints stored in a database;
- and
- identifying a crisis type based on a favorable match between the generated crisis fingerprint and one of the crisis fingerprints stored in the database.
A method for forecasting a performance crisis occurring in a system having a plurality of computer resources, and identifying its type comprising:
- generating an epoch fingerprint for each epoch that signals are collected from each computer resource in the system, each epoch fingerprint representing a summary of the state of the computer resources during the epoch across a subset of the signals;
- upon identifying a first performance crisis in the system, generating a first crisis fingerprint including at least one epoch fingerprint associated with an epoch prior to the occurrence of the first performance crisis;
- storing the first crisis fingerprint in a database;
- upon identifying a second performance crisis in the system, generating a second crisis fingerprint including at least one epoch fingerprint associated with an epoch prior to the occurrence of the second performance crisis;
- determining that the second performance crisis is a recurrence of the first performance crisis based on a favorable comparison between the second crisis fingerprint and the first crisis fingerprint;
- and
- predicting that the first performance crisis may recur again in the system upon identifying at least one newly generated epoch fingerprint that favorably matches the the first crisis fingerprint that was generated prior to the occurrence of the first performance crisis.
A method for automatically identifying a performance crisis type occurring in a system having a plurality of computer resources, comprising:
- generating an epoch fingerprint for each period of time that signals are collected from the computer resources in the system;
- upon detecting a performance crisis in the system, generating a crisis fingerprint consisting of at least one epoch fingerprint generated prior to the performance crisis;
- comparing the generated crisis fingerprint against one or more model crisis fingerprints stored in a database;
- augmenting the generated crisis fingerprint with at least one epoch fingerprint generated after the detection of the performance crisis;
- comparing the augmented crisis fingerprint against one or more model crisis fingerprints stored in a database;
- and
- identifying that the performance crisis associated with the augmented crisis fingerprint has previously occurred in the system based on favorably matching the augmented crisis fingerprint with a model crisis fingerprint stored in the database.
Dependencies are inferred from the claim text.