I would say threading overhead as might be seen by an application. The tests

in www.mcs.anl.gov/~thakur/thread-tests also try to measure that. I will

point Sameer at those.

Rajeev

-----Original Message-----

From: owner-mpich2-core@mcs.anl.gov 

[mailto:owner-mpich2-core@mcs.anl.gov] On Behalf Of Pavan Balaji

Sent: Wednesday, January 30, 2008 2:57 PM

To: mpich2-core@mcs.anl.gov

Subject: Re: [mpich2-core] MPICH2 Threading Overhead

Yeah, I misunderstood the code. Darius pointed it out too. 

However, it 

again comes back to what exactly we want to measure -- threading 

overhead in MPICH2 or threading overhead in the overall communication 

and networking stack?

  -- Pavan

On 01/30/2008 01:57 PM, Rajeev Thakur wrote:

Well the original code IBM sent does not have this 1 

process limitation. See

attached. The threads of a process communicate with the 

threads of another

process. I think that is a better test. Why did we change it?

Rajeev

-----Original Message-----

From: owner-mpich2-core@mcs.anl.gov 

[mailto:owner-mpich2-core@mcs.anl.gov] On Behalf Of Darius Buntinas

Sent: Wednesday, January 30, 2008 1:07 PM

To: mpich2-core@mcs.anl.gov

Subject: Re: [mpich2-core] MPICH2 Threading Overhead

I just want to check what the benchmark should do.

My understanding from the conference calls and Sameer's 

slides is that 

they're interested in the case where multiple threads of a 

process are 

sending/receiving to/from different processes.  This allows 

the process 

to use multiple DMAs concurrently.

Right now, we've been testing the case where one process is 

sending/receiving with itself.  So moving the locks around VCs may 

decrease the size of the CS, but will still serialize.

So what's the test we want to run?

   I imagine something like P processes each with P-1 threads 

where each 

thread is communicating with a different process.  But 

then we need a 

machine with P(P-1) cores.

   On the 8-core nodes we have, we can run 3 processes 

with 2 threads 

each.  But I think we need more threads to really stress things.

It seems that what we can see from the one or two processes 

is the just 

the effect of the size of the CS.

Comments?

-d

-- 

Pavan Balaji

http://www.mcs.anl.gov/~balaji