Re: CPU affinity & IPI latency

[Mike Kravetz]

On Thu, Jul 12, 2001 at 05:36:41PM -0700, Larry McVoy wrote:
> Be careful tuning for LMbench (says the author :-)
> 
> Especially this benchmark.  It's certainly possible to get dramatically better
> SMP numbers by pinning all the lat_ctx processes to a single CPU, because 
> the benchmark is single threaded.  In other words, if we have 5 processes,
> call them A, B, C, D, and E, then the benchmark is passing a token from
> A to B to C to D to E and around again.  
> 
> If the amount of data/instructions needed by all 5 processes fits in the 
> cache and you pin all the processes to the same CPU you'll get much 
> better performance than simply letting them float.
> 
> But making the system do that naively is a bad idea.

I agree, and can't imagine the system ever attempting to take this
into account and leave these 5 tasks on the same CPU.

At the other extreme is my observation that 2 tasks on an 8 CPU
system are 'round robined' among all 8 CPUs.  I think having the
2 tasks stay on 2 of the 8 CPUs would be an improvement with respect
to CPU affinity.  Actually, the scheduler does 'try' to do this.

It is clear that the behavior of lat_ctx bypasses almost all of
the scheduler's attempts at CPU affinity.  The real question is,
"How often in running 'real workloads' are the schduler's attempts
at CPU affinity bypassed?".

-- 
Mike Kravetz                                 [e-mail]
IBM Linux Technology Center