15. WORKLOAD CHARACTERIZATION › 15.6 Case Studies › 15.6.1 Batch Job Class Study › 15.6.1.1 Problem Description

15.6.1.1 Problem Description

This problem asks you to determine a new set of resource-
limited job classes for an installation.  Job class
structures are generally established to meet two objectives
(FER78, ART79a):

o  Provide operational controls
o  Facilitate scheduling and implement service policies

You usually define job classes to meet the first objective
for reasons other than resource requirements.  For example,
online systems like IMS and CICS are often run in unique job
classes to avoid interference with other workloads.  Also,
many computer centers use one or two job classes to segregate
production jobs from the remainder of their workload.

Job classes established to meet the second objective provide
users with alternative classes for job assignments based on
resource requirements.  Typically, the service objective of a
job class decreases (that is, jobs have longer turnaround
times) as its resource requirements increase.  Using a job
class structure that focuses on scheduling and service has a
number of potential benefits:

o  Many jobs can be completed in a short time (assuming
   demand) since the job classes representing smaller jobs
   may be expedited.

o  The availability of better turnaround time for the smaller
   classes may encourage users to conserve resources.

o  Job scheduling is simplified since each job class
   represents a homogeneous set of jobs.

Clustering is an ideal approach to this problem since we want
to exploit the natural characteristics of the workload.  If
we arbitrarily select job classes, we might select classes
that split normal resource consumption patterns or choose
some that are too large or small to be useful.

For this sample installation, we investigated job classes
based on CPU time and tape drives usage.  We selected these
resources to quantify the system, output, and operator
requirements presented by the jobs.  Additionally, several
other reporting metrics were included to quantify the
resources represented by each cluster.  Although a number of
other features could have been selected for analysis, only
these were chosen to keep the resulting job class structure
as simple as possible.