Systems Manager Corner

Converting To Continuum – VOS Scheduler Issues

Article 11/97

There was a lot of discussion in one of the Tucson sessions about converting to the Continuum platform. Several people expressed concerns about response time problems while batch or background jobs are running.

As sites upgrade to a Continuum or another RISC machine from an older system, they are frequently converting from a box with 4 or more processors to a box with one or two processors. On those older boxes, batch jobs could tie up a processor or two, but there were usually enough processors to give reasonable response to the online system. However, with a one or two-processor RISC machine, a batch job could tie up a significant portion of the entire module. The online system will suffer, since it has to compete for CPU power.

A site experiencing these problems have a couple of options:

1. Examine the time of day that the batch jobs run. If they can be rescheduled away from the peak online time, it will help.
2. Check to see if multiple batch jobs are running at the same time. It's not unusual to see 3 or 4 batch jobs start up at the same time, with a heavy impact on CPU and disk. Fix these, if possible, so that they run serially. This will also help alleviate disk and disk cache contention.
3. Investigate the current setup of the VOS scheduler by running display_scheduler_info. The default VOS scheduler is more appropriate for a development shop than for a shop who online work is accomplished in the background. Note that the "batch" portion of the scheduler applies to all started processes , as opposed to login processes. "Batch" is misleading in this context. A previous System Manager's Corner article discussed the VOS scheduler in depth. Check out the article or drop me a note and I'll send you a copy.