From: Herbert Poetzl (herbert_at_13thfloor.at)
Date: Wed 24 Nov 2004 - 21:16:28 GMT
On Wed, Nov 24, 2004 at 11:00:40AM -0500, Gregory (Grisha) Trubetskoy wrote:
>
>
> On Wed, 24 Nov 2004, Herbert Poetzl wrote:
>
> >>Then, if you are a fat jabba, maybe you might end up getting rescheduled
> >>instead of getting more memory whenever you want it!
> >
> >thought about a simpler approach, with a TB for the
> >actual page-ins, so that every page-in will consume
> >a token, and you get a number per interval, as usual ...
>
> There probably still needs to be a "target" size, which if exceeded, your
> bucket is refilled slower. This way small contexts would not be suffering
> because of a large and very active context. The sysadmins would need to
> make sure that the sum of all targets does not exceed physical RAM.
>
> So you'd have two additional parameters - target size and fill-interval
> multiplier.
well, actually I was more thinking of using something
similar to the scheduler TB (see the vserver paper for
details) and just do the check/dec for every 'page'
> if (is_exceeded(target)) {
> interval *= multplier;
> }
>
> Also - at which point does a malloc actually fail? It seems like context 0
> should have a priority over other contexts - a non-0 context should under
> no circumstances be able to exhaust the system memory.
well, malloc is something which will fail if RSS or
VM limits are reached (actually it's brk() and mmap())
but in the typical case the page TB would just cause
the context to be not scheduled (i.e. a page in would
take a _lot_ longer than usual ;)
best,
Herbert
> May be there should be an additional level in the bucket - "reschedule
> level". If I actually empty the bucket, the malloc fails?
>
> my $0.02
>
> Grisha
> _______________________________________________
> Vserver mailing list
> Vserver_at_list.linux-vserver.org
> http://list.linux-vserver.org/mailman/listinfo/vserver
_______________________________________________
Vserver mailing list
Vserver_at_list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver