112 Customizing IBM Tivoli Workload Scheduler for z/OS V8.2 to Improve Performance
5.3 Job streams in an end-to-end environment
The usual good scheduling practices apply equally in an end-to-end
environment. However, there is an additional constraint in that you should avoid
as much as possible references to other FTWs, and especially to the controller
on the mainframe. This has a significant impact on both performance and fault
tolerance. Two techniques for reducing this are:
Do not use mainframe dummy jobs as placeholders at the beginning and end
of distributed job streams. Use jobs issuing a simple distributed command (for
example, echo) instead.
Avoid having many jobs testing for a file or special resource. Instead, have
one “anchor job” dependent on the file or special resource, and make the
other jobs successors to it.
Keeping the Symphony file as small as possible is good for performance, not only
in terms of building and distributing it, but also in referencing it on the distributed
agent. In an end-to-end environment, the Symphony file contains:
Jobs to be executed on the distributed agents
Mainframe jobs that are predecessors to distributed jobs
Job streams that have at least one job in the Symphony file
Consequently, we can see that having many job streams, containing a mix of
both mainframe and distributed jobs, will have a significant impact on the
5.4 Moving JCL into the JS VSAM files
This section defines the best methods for improving the rate at which Tivoli
Workload Scheduler for z/OS is able to move the JCL into the JCL VSAM
repository (JS files).
The JCL is fetched from the JS VSAM files when a job is being submitted. If it is
not found in the VSAM file, it is either fetched by EQQUX002 (user-written exit) or
from the EQQJBLIB data set concatenation.
Normally, the JCL is moved into the VSAM files immediately prior to submission;
however, this means that any delays in fetching the JCL are imbedded into the
batch window. To avoid this delay, the JCL could be moved into the VSAM file
earlier in the day, ready for submission. To be able to do this, you do need to
know which jobs cannot be pre-staged, for example, any JCL that is built
dynamically by some external process. You also need a PIF program that can do
this selective pre-staging for you.