Re: [Lse-tech] more on scheduler benchmarks

[Gerrit H. <ge...@us...>]

FWIW, DYNIX/ptx did process pairs originally for a BAAN application and
later for Oracle.  The first lesson we learned is that pairs of
processes was nice but often processes came in little clusters.  As a
result, we implemented a system call which allowed any process to bind
to any other process.  End result was that a set of processes could
bind to one process.  Scheduling decisions then needed to take this
into account, moving an entire set into a run queue when moving
processes and making sure to factor in the "weight" of the set of
processes when making a "migration" decision.

Of course, long chains of relationships and the full generalized cases
that resulted would have been a pain to handle (not impossible but time
consuming).  The general usage model was that spawned children would 
through an explicit call, typically bind to their parent.  This would
also be possible for pipes.  We avoided having these bindings take place
automatically.  New run queue homes were typically selected as part of
the exec() process, rather than the fork() process, so parents and children
communicating via pipes would typically be on the same CPU, at least until
the child did an exec().  Of course, I'd bet that the general case with
pipes is that the child exec()'s a binary afterwards, so our strategy
(which didn't focus on pipes) might not work as well in this case.

gerrit

> Andi Wrote :
> 
> > On Mon, Jan 22, 2001 at 02:23:05PM -0500, Bill Hartner wrote:
> > > Mike K, wrote :
> > >
> > > >
> > > > If the above is accurate, then I am wondering what would be a
> > > > good scheduler benchmark for these low task count situations.
> > > > I could undo the optimizations in sys_sched_yield() (for testing
> > > > purposes only!), and run the existing benchmarks.  Can anyone
> > > > suggest a better solution?
> > >
> > > Hacking sys_sched_yield is one way around it.
> >
> > How about process pairs that bounce around tokens in pipes ?
> >
> > -Andi
> 
> Process pairs - good idea, should be easy to hack sched_test_yield
> to do this.
> 
> Bill
> 
> 
> _______________________________________________
> Lse-tech mailing list
> Lse...@li...
> http://lists.sourceforge.net/lists/listinfo/lse-tech
>