Browse Prior Art Database

Algorithm for General Diagnosis Involving Historical Performance Data

IP.com Disclosure Number: IPCOM000116029D
Original Publication Date: 1995-Jul-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 2 page(s) / 56K

Publishing Venue

IBM

Related People

Berry, RF: AUTHOR

Abstract

Disclosed is a heuristic method for general performance problem diagnosis involving historical performance data.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 68% of the total text.

Algorithm for General Diagnosis Involving Historical Performance Data

      Disclosed is a heuristic method for general performance problem
diagnosis involving historical performance data.

      Measurement values for a key performance metric are collected
over time and recorded in a historical repository.  Periodically,
data
in the repository is examined for evidence of a system performance
problem.

      An important objective is to report a problem only when a
problem is likely to exist (avoid so-called false positives).
Another important objective is to present the analysis in a manner
that mirrors the approach a human analyst might take.  (The rationale
for this is to improve the credibility of any automatically generated
report resulting from this diagnosis).

      The heuristic employed is outlined in the flowchart shown in
the figure.  It applies to clusters of metrics and individual
metrics.  Metric clusters are relevant for assessments of a
load-balancing nature (e.g., disk activity, multi-processor cpu
utilization).  Individual metric analysis applies to concepts that
have associated thresholds (e.g., file system sizes, process limits),
as well as to those that have no explicit thresholds.

      The general philosophy embodied in the approach is to first
check for balance; if balanced, then determine whether or not a
threshold is being pushed or exceeded.  If not balanced, then report
this as an indication of a problem, and evaluate...