[mpich-discuss] Hydra binding questions

[Pavan Balaji]

Hi Rohit,

On 08/04/2011 12:16 PM, Jain, Rohit wrote:
> 1. We used taskset with Hydra. We are seeing about 20% slowdown with
> (2-1.4 Hydra + taskset) Vs (2-1.0.6 mpd + taskset) on same machine
> with same taskset binding. We were expecting Hydra to perform same or
> faster than mpd. Is that not true?

What kind of performance degradation are you seeing? If this is 
application performance degradation, then mpd vs. hydra might have 
little to nothing to do with it. This might be some totally separate 
issue (e.g., something in the changes to MPICH2 from v1.0.6 to v1.4).

Also, you can use hydra's mpiexec from 1.4 to execute your application 
compiled with mpich2-1.0.6. If you want to do a more head-to-head 
comparison.

> 2. We are seeing relatively high system activity seen in 'top' using
> new mpich (with or without hydra binding). Also, user time on the cpu
> cores hovers around 98-99% compared to old mpich (with mpd) where is
> always nearly 100%. Is that expected due to new Hydra?

98-99% vs. 100% is probably within the noise to really say anything 
reasonable. I can't really say anything about this.

> 3) Do you have any performance benchmarks with mpd vs hydra? Above
> issues are hindering our plan to upgrade to Hydra.

We have seen significant performance improvements for Hydra compared to 
mpd (more than 10x in some cases). But this is for "launching" 
processes. But application execution times should not change much. 
Initial MPI calls can accelerate a bit, but for long running 
applications, it shouldn't matter.

  -- Pavan

-- 
Pavan Balaji
http://www.mcs.anl.gov/~balaji