Processor storage

Some amount of paging and swapping is normal for systems that serve a lot of users, or have a significant amount of data in virtual storage. The impact on your system performance from this activity is not normally severe. The few more I/O operations per transaction for paging and swapping are not usually a significant increase on the I/O that would be required without paging and swapping.

If performance is important, define the IBM Tivoli Workload Scheduler for z/OS address spaces as non-swappable.

The virtual storage of a processor may far exceed the size of central storage available in the configuration. Any excess must be maintained in auxiliary storage (DASD), or in expanded storage. This virtual storage occurs in blocks of addresses called pages. Only the most recently referenced pages of virtual storage are assigned to occupy blocks of physical central storage. When reference is made to a page of virtual storage that is not currently in central storage, the page is brought in from DASD or expanded storage to replace a page in central storage that is not in use and is least recently used. The newly referenced page is said to have been paged in. The displaced page may need to be paged out if it has been changed.

IBM Tivoli Workload Scheduler for z/OS uses a significant amount of virtual storage, both in the controller address space and by the batch programs that create the current plans.

The local shared resources (LSR) buffering technique is used by IBM Tivoli Workload Scheduler for z/OS for the current plan data set; a percentage of the current plan determined by the CPDTLIM keyword on the OPCOPTS statement is kept in LSR buffers above the 16-megabyte line. These LSR buffers are deleted and rebuilt at every current plan backup to ensure that the percentage of the plan that you want is always in storage.

Two data spaces, one for calendars and one for special resources, are also created and maintained by the controller address space. The size of calendar data space is dependent on the number of calendars defined. The special resource data space contains all special resource referenced in the current plan. The data spaces are automatically extended by IBM Tivoli Workload Scheduler for z/OS to accommodate additional data.

Paging problems

The page-in rate is of primary concern, because page-in activity occurs synchronously (that is, a z/OS task stops until the page fault is resolved). Page-out activity overlaps with IBM Tivoli Workload Scheduler for z/OS processing, so it does not appreciably affect throughput.

A page-in from expanded storage incurs only a small processor usage cost but a page-in from DASD incurs a time cost for the physical I/O and a more significant increase in processor usage.

Thus, extra DASD page-in activity slows down the rate at which transactions are processed by the controller. If you suspect that a performance problem is related to excessive paging, you can use RMF™ to obtain the paging rates.