On Mon, 25 Apr 2011 09:34:25 -0400, "Dean, David (I/S)"
<[email protected]> wrote:
> Can some of you weigh in on the merits (demerits) of defining multiple
> CPU's to virtual Linux boxes?  We are heavy WebSphere and have gotten
> differing opinions.

For what it's worth we generally allocate vCPUs in environments based on
utilisation, more than any other factor.  We did experiment with multiple
CPUs for guests that didn't actually need it (per some SHARE advice), but
we couldn't see any measurable (or even anecdotal) improvement in
behaviour, either for guests or the LPAR.

That said, orthodox advice from IBM performance specialists we've had
consulting with us has been that the limit for some pools (such as the web
container thread pool) is a funtion of the number of CPUs available (e.g.
in the case of the web container, the recommended starting point is 50
threads per CPU).  Our own testing has generally borne this out.

The second consideration, which sometimes overrides the first, is harm
reduction: in our dev/test regions we tend to lock down the number of vCPUs
to stop badly-behaved apps shitting things up too much.  Bad enough when
the first cut of a piece of code eats a whole IFL.

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390
----------------------------------------------------------------------
For more information on Linux on System z, visit
http://wiki.linuxvm.org/

Reply via email to