* sedf internals question
@ 2006-03-12 17:33 Tim Freeman
2006-04-05 13:34 ` Stephan Diestelhorst
0 siblings, 1 reply; 3+ messages in thread
From: Tim Freeman @ 2006-03-12 17:33 UTC (permalink / raw)
To: xen-devel
With SEDF, let us say a domain has a certain timeslice allocated to it (say, 500
units) for a certain period (say 1000). It runs for 200, then blocks. When it
has become unblocked, the domain is put on the runqueue -- but say we are at the
end of the period, maybe 900 units in. Even if it immediately runs for 100
units, what happens to the extra 200 units it is guaranteed?
Thanks,
Tim
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: sedf internals question
2006-03-12 17:33 sedf internals question Tim Freeman
@ 2006-04-05 13:34 ` Stephan Diestelhorst
2009-06-17 13:05 ` Gaurav Somani
0 siblings, 1 reply; 3+ messages in thread
From: Stephan Diestelhorst @ 2006-04-05 13:34 UTC (permalink / raw)
To: Tim Freeman, xen-devel
Hi Tim,
>With SEDF, let us say a domain has a certain timeslice allocated to it (say, 500 units) for a certain period (say 1000). It runs for 200, then blocks. When it has become unblocked, the domain is put on the runqueue -- but say we are at the end of the period, maybe 900 units in. Even if it immediately runs for 100 units, what happens to the extra 200 units it is guaranteed?
>
>
Those guaranteed 500 units get binned once a domain blocks, because, as
you already pointed out, the scheduler can't make anymore guarantees at
that point. It can however give some of the suddenly available idle-time
to the domain. This is done by a priority scheme that supports those
domains most, that lost most of their time due to blocking. There are
(or used to be?) some different approaches in sched_sedf.c, which are
selectable at compile-time by exchanging macros. Play with those if you
want to tweak your system, although my measurements suggested the one
that is curently used to be the most useful. (And please send me some of
your results if you do ;-) )
Cheers,
Stephan
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: sedf internals question
2006-04-05 13:34 ` Stephan Diestelhorst
@ 2009-06-17 13:05 ` Gaurav Somani
0 siblings, 0 replies; 3+ messages in thread
From: Gaurav Somani @ 2009-06-17 13:05 UTC (permalink / raw)
To: xen-devel
Dear Stephen,
I want to know the importance of the extra command line field in sedf
scheduler. which is l (latency). It only says about when domain does high
I/O work. What does it mean by "high"?
please give me the way. It will elp me in my research.
Thanks
Gaurav
onlineengineerATgmailDOTcom
Stephan Diestelhorst-2 wrote:
>
> Hi Tim,
>
>>With SEDF, let us say a domain has a certain timeslice allocated to it
(say, 500 units) for a certain period (say 1000). It runs for 200, then
blocks. When it has become unblocked, the domain is put on the runqueue --
but say we are at the end of the period, maybe 900 units in. Even if it
immediately runs for 100 units, what happens to the extra 200 units it is
guaranteed?
>>
>>
> Those guaranteed 500 units get binned once a domain blocks, because, as
> you already pointed out, the scheduler can't make anymore guarantees at
> that point. It can however give some of the suddenly available idle-time
> to the domain. This is done by a priority scheme that supports those
> domains most, that lost most of their time due to blocking. There are
> (or used to be?) some different approaches in sched_sedf.c, which are
> selectable at compile-time by exchanging macros. Play with those if you
> want to tweak your system, although my measurements suggested the one
> that is curently used to be the most useful. (And please send me some of
> your results if you do ;-) )
>
> Cheers,
> Stephan
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>
>
--
View this message in context: http://www.nabble.com/sedf-internals-question-tp3366114p24073256.html
Sent from the Xen - Dev mailing list archive at Nabble.com.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2009-06-17 13:05 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2006-03-12 17:33 sedf internals question Tim Freeman
2006-04-05 13:34 ` Stephan Diestelhorst
2009-06-17 13:05 ` Gaurav Somani
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.