|
|
 |
|
|
|
|
Using Batch Initiator Management
|
 |
Q: It was announced that with z/OS 1.4, WLM Initiator Balancing is available to balance WLM managed initiators between the systems of a sysplex.
However on my sysplex with all systems running under z/OS 1.4 and JES2, I can see that one system has 80% processor utilization and runs more batch workload than another system with only 20% utilization. Why does WLM not rebalance WLM managed initiators in this situation?
A: The target of Batch Initiator Balancing is to optimize the performance and throughput of batch workload across the sysplex. Its intention is not to reach an equally balanced distribution of batch jobs over the LPARs of a sysplex.
This is why initiator balancing will only react when at least one of the systems in the sysplex has a processor utilization of more than 95%, while other systems have more idle capacity. If even the most loaded system has enough idle capacity to run batch jobs without CPU constraints, it would not improve the total batch throughput if initiators were moved away from that system to other systems, even if the other systems have significantly more idle CPU capacity.
Q: How do I recycle a WLM initiator that is getting ABEND822s?
A: There are two ways to do this:
- If you can determine the ASID of the initiator that is abending, you can stop it by issuing the command P INIT,A=asid. The initiator does not need to be idle at the time that you enter the command. If the initiator is busy processing a job, it will stop itself after the job finishes. WLM will automatically replace the initiator with a new one.
- If you cannot determine the ASID, or if you want to recycle all initiators as part of a regular cleanup process, you can enter the commands $P XEQ and $S XEQ. The $P XEQ command causes all WLM initiators on that system to be "flagged" to terminate. The $S XEQ command enables WLM to start new initiators (without needing to wait for the old initiators to end). Beware that the $P XEQ command purges WLM's history which tells it how many initiators are needed for each service class. It may take some time for WLM to build up the same number of initiators that existed before.
Q: After I switch some jobclasses to WLM management, how do I know how many initiators WLM has started?
A: There is no special display for the number of WLM initiators because this is not something you can control manually. The real test of whether WLM created "enough" initiators is seeing that the work is meeting its goals. The methods suggested below should be enough to assure yourself that the function is indeed working.
Note that for methods 1 and 2, the counts could include JES initiators if you recently transitioned from JES-managed jobclasses to WLM-managed. If a job is already executing at the time its jobclass is changed to MODE=WLM, it is allowed to complete normally. Once the job completes, the JES initiator is no longer used for that jobclass.
- By jobclass:
Use the SDSF JC display to find the jobclasses that are WLM mode and look at the count of executing jobs for each job class. The actual number of WLM initiators may be slightly higher than the sum of the executing jobs in WLM-managed classes. This is due to idle initiators that are kept around for a period of time in case additional jobs become eligible for selection.
- By service class period:
Look at the RMF workload activity report to find out the average number of address spaces for a service class period (AVG ADRSP). If you adhere to the recommendation of not mixing JES-managed batch work and WLM-managed batch work in the same service class, then AVG ADRSP indicates the average number of WLM initiators servicing the period.
- By system:
If you know how many traditional JES initiators you have, issue the D A,L command and subtract the JES inits out of the total initiator count on this display. The result is the number of WLM-created initiators for the system.
- Idle initiators:
To see the number of idle WLM batch initiators, use the SDSF DA OINIT command which shows only idle initiators. Those without JOBIDs are WLM initiators. WLM initiators don't have a JOBID because they are started under the master subsystem. You can set up an SDSF filter to show only the WLM idle initiators.
|
 |
|
|
|