[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: thread ids, task ids and subsystems
From: |
Volkmar Uhlig |
Subject: |
RE: thread ids, task ids and subsystems |
Date: |
Wed, 2 Apr 2003 21:35:54 +0200 |
> -----Original Message-----
> From: Peter De Schrijver [mailto:address@hidden
> Sent: Wednesday, April 02, 2003 9:29 PM
>
> > > Additionally, you have significant cache-line migration
> > > costs to get the request onto the other CPU. When deleting
> > > and re-creating the task, it will show up on the wrong CPU
> > > and the load balancer has to migrate it afterwards adding
> > > another X-CPU scheduling request (per thread!). So
> > > better do it on the right CPU in the first place.
> >
> > Ouch :) That makes sense.
> >
>
> But this does not apply to P4 hyperthread instances, does it ?
Depends. I plan to experiment with different locking schemes on HT
processors which may lead to completely different scheduling
abstractions for HT and real SMP, but this is still open. Therefore, yes
the cache line migration costs are lower, and no, you still may have to
migrate the thread and therefore potentially trigger an IPI.
- Volkmar
- Re: thread ids, task ids and subsystems, (continued)
- Re: thread ids, task ids and subsystems, Marcus Brinkmann, 2003/04/14
- Re: quotas [was: thread ids, task ids and subsystems], Michal 'hramrach' Suchanek, 2003/04/16
- Re: quotas [was: thread ids, task ids and subsystems], Marcus Brinkmann, 2003/04/16
- Re: quotas [was: thread ids, task ids and subsystems], Michal 'hramrach' Suchanek, 2003/04/16
- Re: quotas [was: thread ids, task ids and subsystems], Marcus Brinkmann, 2003/04/16
RE: thread ids, task ids and subsystems, Volkmar Uhlig, 2003/04/02
RE: thread ids, task ids and subsystems, Volkmar Uhlig, 2003/04/02
RE: thread ids, task ids and subsystems,
Volkmar Uhlig <=