Rob van der Heij wrote:
> So what options does z/VM have then when you overcommit resources and
> suddenly everyone comes to get his share?
Limit the working set size of the guest, and thereby throttle it.

> Should CP randomly force
> virtual machines off the system so that people not just *think* it has
> crashed? (just like Linux kill processes when you run out of swap
> space).
In this situation, z/VM does not run out of paging space as far as I see.

> If it were just for CPU resources, it might help to guarantee a
> minimum amount of cycles to keep the guest operating system alive (but
> can we be sure it would be used for that). But when the virtual
> machine also needs a lot of pages resident to consume those cycles,
> then CP would spend more time rushing pages in and out.
But similar to cpu, CP could set a minimal set of pages resident.

> Maybe it would be nice if CP could put the guest in some "power saving
> mode" where we know that it will only use resources required to avoid
> bigger damage. How about when Linux would run those emergency tasks
> only on a specific virtual CPU so that CP can tell that work is
> important?
By using sched_setaffinity [see man page] you can bind processes to CPUs if
you want to. How would Linux tell what work is "important" for you? Usually
load-balancing over CPUs is what people want.
--

Carsten Otte
IBM Linux technology center
ARCH=s390

----------------------------------------------------------------------
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

Reply via email to