|
WLM batch initiator management was introduced with OS/390 R4 and OS/390 JES2 R4. Later, in OS/390 R8, JES3 started to provide similar functionality for JES3-environments.
Beginning in OS/390 R4, WLM has the capability of controlling the rate at which queued jobs are initiated. Moreover, the WLM will dynamically change the number of WLM initiators and/or their work selection criteria in an attempt to meet installation defined goals.
The installation can choose between JES-managed initiators and WLM-managed initiators by job class. Both types of initiators can coexist. With WLM-managed initiators, it is WLM who controls the number and placement of the initiator address spaces.
New initiators are started when service class goals are missed, when a system is underutilized and there are jobs waiting to be selected, or when jobs have an affinity to a system on which no initiator is available yet. Decision factors are the available CPU and memory resources on a system, the service class' importance, and the projected net value on overall goal achievement.
WLM stops initiators, when the number of started initiators is 1.5 times of the long term average queue length, when a system runs short of CPU or memory, or when the last initiator was inactive for 1 hour.
With z/OS V1R4, WLM improves the balancing of WLM managed batch initiators between the systems of a sysplex. On highly utilized systems the number of initiators can be reduced while more are started on low utilized systems. This enhancement improves the performance of the sysplex with better use of the processing capacity of each system.
WLM attempts to distribute the initiators across all members in the sysplex to optimize the throughput while taking care that jobs with affinities to specific systems are not hurt by WLM decisions.
When the available CPU capacity of a system decreases to less than 5%, WLM stops an initiator address space when the current system is observed as the system with the highest CPU demand and when there is another remote system, that has enough available resources to start a new initiator. This evaluation is done every 10 seconds. The order of decrease is to stop initiators serving lower importance service classes first.
WLM increases the number of initiators on lower utilized systems. To speed up job selection for a high volume of waiting jobs, up to 5 initiators can be started every 10 seconds on underutilized systems that have enough idle CPU and memory capacity. This value used to be 1 before z/OS V1R4.
Benefits
Comments
-
Reduction starts only when system is >= 95% utilized
-
Therefore, don’t expect balanced situation on completely underutilized systems (see build-up phase in both cases)
| 
