3. PERFORMANCE ANALYSIS TOOLS › 3.3 Data Clustering Analysis › 3.3.2 Usage Guidelines

3.3.2 Usage Guidelines

Before the introduction of cluster-based workload
characterization, a number of other workload characterization
techniques were utilized.  These approaches provide valuable
insight into potential errors in the workload
characterization process.

The equivalent job approach fails to provide an adequate
characterization, since it lacks the important property of
robustness.  Relatively small changes in the workloads can
cause large changes in the definition of an equivalent job.
This means that the results of the equivalent job approach
are too closely associated with the data used in the analysis
to provide general insight into the workloads on the system.
Cluster-based workload characterization can replace commonly
practiced ad hoc techniques.

The most important issues regarding implementation concern
the type of data elements that are used to characterize the
clusters and the treatment of unusual points that cannot be
represented by clusters.  These points, known as outliers,
can have a profound effect on the clustering study.

These topics and the use of certain CA MICS database files
for clustering analysis are discussed in the following
sections:

 1 - Historical Approaches
 2 - Basic Hypotheses
 3 - Application of Workload Characterization
 4 - Data Selection and Outliers
 5 - Suggested CA MICS Input Files