Charm++ parallel programming system

[Pavanakumar Mohanamuraly <m.pavanakumar AT gmail.com>]

Dear Prof. Kale,

We use Infiniband on the cluster for interconnect. But I could not compile Charm++ with ibverbs library as the dev rpm's of libibverbs is not installed on the cluster.

I ran the megatest in tests/charm++ on 3 nodes (Each cluster node has 12 cores (2 Intel Xeon) ) and to my surprise I found the following results

1)    for mpi- (./build charm++ mpi-linux-x86_64 icc mpt smp ifort)

       test 53: completed (541.85 sec)

2)    for net- (./build charm++ net-linux-x86_64 icc smp ifort)

       host config ==> ++shell ssh ++cpus 12

       test 53: completed (0.42 sec)

But this seems strange because I thought SGI MPT is much faster than the native linux sockets. I am petty sure I messed up something. 

Best,

On Thu, Dec 6, 2012 at 7:53 PM, Kale, Laxmikant V <kale AT illinois.edu> wrote:

What is your interconnect? 

-- 

Laxmikant (Sanjay) Kale         http://charm.cs.uiuc.edu

Professor, Computer Science     kale AT illinois.edu

201 N. Goodwin Avenue           Ph:  (217) 244-0094

Urbana, IL  61801-2302          FAX: (217) 265-6582

On 12/6/12 2:23 AM, "Pritish Jetley" <pjetley2 AT uiuc.edu> wrote:

There is a pingpong test in tests/charm++/pingpong. This program calculates round-trip latency in various scenarios using small messages. You could change the payload size therein to get communication times for different message sizes.

Pritish

On Thu, Dec 6, 2012 at 1:34 PM, Pavanakumar Mohanamuraly <m.pavanakumar AT gmail.com> wrote:

Hi

I have successfully installed Charm++ on our cluster (SGI Altix ICE). I have both the net- and mpi- versions of Charm++ built and want to compare the communication performance on both versions. The MPI version is compiled against SGI's MPT library MPI implementation which is supposed to be fast on the SGI cluster.

How can one compare the communication performance ? Is there a ready made program in Charm++ distribution ?

Regards,

-- 
Pavanakumar Mohanamuraly

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm

-- 
Pritish Jetley
Doctoral Candidate, Computer Science
University of Illinois at Urbana-Champaign

-- 
Pavanakumar Mohanamuraly

On Thu, Dec 6, 2012 at 7:53 PM, Kale, Laxmikant V <kale AT illinois.edu> wrote:

What is your interconnect? 

-- 

Laxmikant (Sanjay) Kale         http://charm.cs.uiuc.edu

Professor, Computer Science     kale AT illinois.edu

201 N. Goodwin Avenue           Ph:  (217) 244-0094

Urbana, IL  61801-2302          FAX: (217) 265-6582

On 12/6/12 2:23 AM, "Pritish Jetley" <pjetley2 AT uiuc.edu> wrote:

There is a pingpong test in tests/charm++/pingpong. This program calculates round-trip latency in various scenarios using small messages. You could change the payload size therein to get communication times for different message sizes.

Pritish

On Thu, Dec 6, 2012 at 1:34 PM, Pavanakumar Mohanamuraly <m.pavanakumar AT gmail.com> wrote:

Hi

I have successfully installed Charm++ on our cluster (SGI Altix ICE). I have both the net- and mpi- versions of Charm++ built and want to compare the communication performance on both versions. The MPI version is compiled against SGI's MPT library MPI implementation which is supposed to be fast on the SGI cluster.

How can one compare the communication performance ? Is there a ready made program in Charm++ distribution ?

Regards,

--
Pavanakumar Mohanamuraly

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm

--
Pritish Jetley
Doctoral Candidate, Computer Science
University of Illinois at Urbana-Champaign

-- 
Pavanakumar Mohanamuraly