Re: RUNPTY and TIMESLICE in *CLS objects

Hello Charles,

Am 16.12.2021 um 17:29 schrieb Charles Wilt <charles.wilt@xxxxxxxxx>:

Time slice on the IBM i really only comes into play in a CPU intensive workload...

Or even more, when there are CPU competing processes.

If your job is waiting for screen or disk I/O, the system will trade it out
for something that is ready to use the CPU now.

Having said that, last I heard the default timeslice assigned to IBM
defined classes (ex: 2000 interactive / 5000 batch) haven't changed since
the first AS-400 was shipped.

CPU's where much slower then. :)

Yap.

A much more modern CPU appropriate timeslice is 200 / 500 I believe.

Now that you say that…

I'm curious the difference between an internal and an external time slice is still there. According to (somewhat dated) Michael Catalani's Book "Performance Tuning Simplified", there is a hard, non-configurable 500ms real-clock time slice where the scheduler is taking the CPU to compute the next eligible process to take the CPU.

As far as I've understood, the "external" time slice is what can be set in *CLS objects and does *not* refer to wall clock seconds, but accumulated CPU time. This means, a process is my default interrupted by the internal time slice end by default (if there are other processes needing CPU cycles) and almost never by what is configured in the *CLS objects.

I'm not sure if this means that one of several competing processes can hog the CPU until the internal 500ms time slice ends, or a page fault forces a rescheduling of eligibility.

(Thanks to Mark Waterbury for mentioning that Book to me in a private message! He mentions that the internal time slice timer is not mentioned in "Work Management".)

On my 150, almost all tasks are almost always CPU bound when the machine has been running for a while and settled on which frequently accessed disk pages need to be kept in RAM.

:wq! PoC