Re: suspend processes at load

Re: suspend processes at load

[happz]

What about this: give to process way how to tell kernel "it is not
good to suspend me, because there are process' that depend on me and
wouldn't be blocked." Syscall or /proc filesystem could be used.

It is not the way how to say which process should be suspended but a
way how to say which could NOT - usefull for example for X server, may
be some daemons, aso.

Milos Prchlik


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: suspend processes at load

[James A. Sutherland]

On Mon, 19 Mar 2001 21:06:55 +0100, you wrote:

>What about this: give to process way how to tell kernel "it is not
>good to suspend me, because there are process' that depend on me and
>wouldn't be blocked." Syscall or /proc filesystem could be used.
>
>It is not the way how to say which process should be suspended but a
>way how to say which could NOT - usefull for example for X server, may
>be some daemons, aso.

Possibly; TBH, I don't think it's worth it. Remember, "suspending" X
would just stop your mouse moving etc. for (e.g.) 5 seconds; in fact,
that should block most graphical processes, which may well resolve the
thrashing in itself!


James.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: suspend processes at load

[Eric W. Biederman]

"James A. Sutherland" <jas88@cam.ac.uk> writes:

> On Mon, 19 Mar 2001 21:06:55 +0100, you wrote:
> 
> >What about this: give to process way how to tell kernel "it is not
> >good to suspend me, because there are process' that depend on me and
> >wouldn't be blocked." Syscall or /proc filesystem could be used.
> >
> >It is not the way how to say which process should be suspended but a
> >way how to say which could NOT - usefull for example for X server, may
> >be some daemons, aso.
> 
> Possibly; TBH, I don't think it's worth it. Remember, "suspending" X
> would just stop your mouse moving etc. for (e.g.) 5 seconds; in fact,
> that should block most graphical processes, which may well resolve the
> thrashing in itself!

Actually we should only apply suspension and the like to SCHED_OTHER.
The realtime scheduling classes should be left as is.  If an
application is safe to run realtime, it should be o.k. in the
thrashing situation. 

Also actually suspending a realtime process would be a violation of
the realtime scheduling guarantees, where with SCHED_OTHER you can be
expected to be suspended at any time.

Eric
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: suspend processes at load

[James A. Sutherland]

On 19 Apr 2001 22:11:28 -0600, you wrote:

>"James A. Sutherland" <jas88@cam.ac.uk> writes:
>
>> On Mon, 19 Mar 2001 21:06:55 +0100, you wrote:
>> 
>> >What about this: give to process way how to tell kernel "it is not
>> >good to suspend me, because there are process' that depend on me and
>> >wouldn't be blocked." Syscall or /proc filesystem could be used.
>> >
>> >It is not the way how to say which process should be suspended but a
>> >way how to say which could NOT - usefull for example for X server, may
>> >be some daemons, aso.
>> 
>> Possibly; TBH, I don't think it's worth it. Remember, "suspending" X
>> would just stop your mouse moving etc. for (e.g.) 5 seconds; in fact,
>> that should block most graphical processes, which may well resolve the
>> thrashing in itself!
>
>Actually we should only apply suspension and the like to SCHED_OTHER.
>The realtime scheduling classes should be left as is.  If an
>application is safe to run realtime, it should be o.k. in the
>thrashing situation. 
>
>Also actually suspending a realtime process would be a violation of
>the realtime scheduling guarantees, where with SCHED_OTHER you can be
>expected to be suspended at any time.

Yes, I was taking that for granted; apart from anything else, realtime
processes are "supposed to" (according to the manpages, anyway!) be
mlock()ed, which makes suspending them pointless: it won't free any
memory anyway.


James.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/