Using worklist parallelism
The
Change Manager
component uses the
Cross-System Image Manager
(
XIM
) technology to provide sysplex performance improvements by enabling the distribution and management of discrete units of work (UOW) across one or more IBM z/OS images in a data sharing environment. By doing so, the
Change Manager
component can divide single, long-running tasks into multiple parallel tasks that can be run across multiple computers.
By distributing the work to multiple address spaces, Change Manager avoids any memory constraints for a single address space and allocates the work to images that have adequate CPU capacity. As a result, work is distributed dynamically to processors that are under utilized, workloads are balanced, and elapsed time for processing is improved. For example, in a PeopleSoft environment, where thousands of tables exist, worklist parallelism can be used to unload, load, and copy structures and data to migrate one subsystem to another quickly and efficiently. Tasks can also be run on a single z/OS image to achieve the same results that are realized in a data sharing environment.
Two functions in the Change Manager component control the worklist parallelism feature:
- Analysis—creates the appropriate worklist commands to run a worklist in parallel
- Execution—executes the worklist in parallel
You can include the appropriate worklist parallelism commands in a worklist without actually running the worklist in parallel. The performance of the worklist is the same, regardless of whether the commands are included and the worklist is not run in parallel, or the commands are not included.
The following figure illustrates the parallel processing of a worklist. Each UOW is processed by an XIM initiator on a z/OS image on which XIM is installed. When processing is complete on an XIM initiator, the next UOW is assigned to that initiator. For example, after XIMINIT 2 completes the processing of UOW 2, it begins processing UOW 4.
In a UOW, commands for unload, load, check, copy, and statistics utilities can be run in parallel. Data definition language (DDL) statements are run sequentially.
This section contains the following topics: