Re: Gang Scheduling in linux

[Hubertus Franke <frankeh@watson.ibm.com>]

On Saturday 20 July 2002 12:59 pm, Ingo Molnar wrote:
> On Fri, 19 Jul 2002, Hubertus Franke wrote:
> > For this it seems sufficient to simply STOP apps on a larger granularity
> > and have that done through a user level daemon. The kernel scheduler
> > simply schedules the runnable threads that given the U-Sched would
> > always amount to a limited number of threads/tasks.
>
> yep, this is my suggestion as well. Any reason to do gang scheduling in
> the scheduler and not via a userspace daemon that stops/resumes (and
> binds) managed tasks explicitly?
>
> 	Ingo

Not from our experience from a cluster based application. Each node was 
a SMP in that case.

On a single SMP I could imagine for instance for parallel reendering
or similar tightly integrated parallel programs that need data 
synchronization.  Most of these apps assume a tightly coupled non-virtual
resource, i.e., scheduling of tasks is aligned.

SGI used to have that stuff in their base kernel. Read a paper about this some 
years ago.
Again, at the beginning I'd go with a user level scheduler approach that 
certainly would satisfy national labs etc. Most of the cluster schedulers, 
like PBS and LoadLeveler etc., already provide that functionality.


-- 
-- Hubertus Franke  (frankeh@watson.ibm.com)