In addition to improving sequential performance and logging, you
can tune several things in the DFSMStvs environment to influence the
performance. What you tune depends on the information you gained from
performance monitoring, as
Monitoring performance describes. The following
list describes some things that you might want to tune:
- Applications
If you detect that there is excessive lock contention
or that the system logger is forced to spill active log records to
disk, you might want to tune the commit frequency that you have implemented
in an application program.
- Coupling-facility storage
- SMSVSAM use
If you see structure-full events for the SMSVSAM
structures, you might want to change the amount of storage in the
coupling facility available for SMSVSAM.
- System logger use
If you see structure-full events for the logger
structures, you might want to change the amount of storage within
the coupling facility available to the system logger.
When
a log becomes full and the system logger has to offload log data,
the system logger starts surfacing temporary errors. While this occurs,
it is impossible to write anything to the log.
- DFSMStvs SMS settings
If you see many occurrences of a coupling
facility log structure filling and spilling to disk, you might want
to reduce the activity keypoint frequency. Setting the activity keypoint
frequency too low, however, would increase the amount of processor
time needed to trim logs.
- Application parallelism
When you run multiple batch jobs against
the same shared VSAM data sets, you can obtain benefits by rescheduling
existing jobs. You can take advantage of application parallelism still
further by taking existing jobs and splitting them into multiple parallel
jobs. This reduces the overall run time substantially, depending on
how many ways you split a single job. A shorter run time does, however,
mean that the total amount of resources consumed by the job is now
consumed in a shorter period of time, so creating more parallel jobs
can cause a peak in processor use and I/O demand.
Recommendation: Improve performance by setting
up your z/OS® system optimally in these ways:
- Run DFSMStvs batch and CICS® from separate z/OS images
rather than combining the two within one z/OS image.
- Place a couple data sets and JES2 checkpoint on different volumes.
- Place the primary sysplex CDS and the coupling facility resource
management (CFRM) CDS on different volumes.
- Ideally, spread the primary and alternate couple data sets and
the CFRM data sets across four volumes.
- Give SMSVSAM a higher dispatching priority than VTAM® and CICSPlex® System
Manager, which in turn should have a higher dispatching priority than CICS.
When you run in goal mode, however, you should allow these system
address spaces to default to SYSTEM/SYSSTC.
- As a starting point, set the activity keypoint to 5000.
- Use GRS star mode.
- Define only the number of systems that will actually join the
sysplex in a couple data set MAXSYSTEM value.
The RLS lock structure,
IGWLOCK00, bases the size of each lock entry on the number of systems
permitted to join the sysplex. Each lock entry increases in size as
more systems are defined. If you define more systems in the couple
data set MAXSYSTEM value than will actually join, each record is larger
than necessary and you can fit fewer records in a given amount of
coupling facility space.
- Increase the size of your lock structure (IGWLOCK00).
DFSMStvs introduces additional locking because batch jobs do not hold
locks in today's processing environment.