NetNews Usenet Archive 1992 #26/NN_1992_26.iso/spool/comp/os/vms/17951

Internet Message Format | 1992-11-14 | 1.7 KB

Path: sparky!uunet!comp.vuw.ac.nz!zl2tnm!toyunix!don
Newsgroups: comp.os.vms
Subject: Re: TUNING - QUANTUM SYSGEM parameter
Message-ID: <[e-mail]>
From: [e-mail] (Don Stokes)
Date: 14 Nov 92 03:23:39 GMT
Sender: [e-mail] (GNEWS Version 2.0 news poster.)
Distribution: world
Organization: The Wolery
Lines: 28

[e-mail] writes:
> 2. What are some reasonable starting QUANTUM settings for the following
>    machines:

The calculation of QUANTUM isn't really tied to the hardware speed.  Rather,
it should be based on the number of interactive computable processes that
would use up their quantum and the acceptable response time.

Also, the calculation was a bit off for V5.2(& 5.3?), because under these
versions, a process would not preempt another unless it was 3 or more
priority levels above the current process.  This has now been fixed, but
with those versions, knocking QUANTUM right down would have helped a lot,
effectively providing a form of preemption, although not nearly as good
as real preemption as done in current and earlier versions.  Could this
be why you got such an improvement by diddling QUANTUM before?

Basically, QUANTUM should be calculated as acceptable response divided by
the number of long term computable interactive processes, ie those that are
unlikely to let the processor go before quantum end.  I'd put "acceptable
response" at around 100ms, and you should estimate a little on the high
side for the number of computable processes.



--
Don Stokes, ZL2TNM (DS555)                        [e-mail] (home)
Network Manager, Computing Services Centre            [e-mail] (work)
Victoria University of Wellington, New Zealand              [phone]