On Tue, 10 Feb 2004, Little, Chris wrote:

> at one time, was IBM recommending differently?  i went to an introduction to
> z/VM several years ago and the instructor was recommending 4 virtual CPU's
> no matter how many real CPU's existed.  He said that VM's multiprocessing
> capability was superior to linux so it was better to queue processes to VM.

Don't know about how z/VM virtualises things, but in terms of access to
real CPs the MVS guys I work with will often prefer two smaller engines to
one big one.  Turns out that if you have one task get away (say a failing
process starts an SMF dump), *everything* stops because the runaway task
consumes all the CPU resource.  If you have two or more engines, at least
some other workload can take place on the other engine(s).

This of course is a fairly specific example, generalities may be
different. ;-)  I must say though that we are having some real issues in a
resource-constrained single-engine LPAR, where spreading the workload
across two half-engines might at least give the illusion of more headroom.

Cheers,
Vic Cross

Reply via email to