Charm++ parallel programming system

[Michael Robson <mprobson AT illinois.edu>]

Hello Pedro and Mauricio,

I just downloaded and ran the no barrier code on my laptop using the same parameters but built on a different communication layer without a problem. However, when I build on my desktop using the mpi layer I see the same hang you mentioned. So we can debug this further, what version (or commit) of charm are you running? Can you also try running on the netlrts layer (built with netlrts-linux-x86_64 instead of mpi-linux-x86_64)? Please let me know the results of this so we can help to resolve the problem.

Thanks,

Michael

On Tue, Mar 7, 2017 at 11:52 AM Pedro Esequiel Tarazi <pedrotarazi AT gmail.com> wrote:

Hi! We are students of Computer Engineer of Facultad Ciencias Exactas, Fisicas y Naturales, Cordoba National University, Argentina. 

We are doing a research about Charm++ Framework. We analyze several codes given in Tutorial, and we founded an error. During the execution of the Parallel Prefix No Barrier program, this program it stop of print in the console, but the processors still running in background. Probably is starvation. This no ocurr always (approximately 2 times of 5). We want know if is a problem in the code, or a problem in the our execution. 

This program was executed in two notebooks and one cluster, both running Linux. Below, we show an example of the result of an execution. 

We await your reply. Thank you. Regards.

Aguilar, Mauricio - Tarazi, Pedro

Students of Computer Engineer

Facultad de Ciencias Exactas, Físicas y Naturales - UNC - Argentina

****************************************************************************************************************************

[14:36:36] pedrotarazi 5.ParallelPrefix_NoBarrier $ ./charmrun +p4 prefix 10000 

Running on 4 processors:  prefix 10000

charmrun>  /usr/bin/setarch x86_64 -R  mpirun -np 4  prefix 10000

Charm++> Running on MPI version: 3.0

Charm++> level of thread support used: MPI_THREAD_SINGLE (desired: MPI_THREAD_SINGLE)

Charm++> Running in non-SMP mode: numPes 4

Converse/Charm++ Commit ID: 

Charm++: Tracemode Projections enabled.

Trace: traceroot: prefix

CharmLB> Load balancer assumes all CPUs are same.

Charm++> Running on 1 unique compute nodes (8-way SMP).

Charm++> cpu topology info is gathered in 0.007 seconds.

Running "Parallel Prefix" with 10000 elements using 4 processors.

Before: Prefix[0].value = 1

Before: Prefix[1].value = 1

Before: Prefix[2].value = 1

Before: =�Prefix[3].value = 1

Before: =�Prefix[4].value = 1

Before: =�Prefix[5].value = 1

Before: =�Prefix[6].value = 1

Before: =�Prefix[7].value = 1

Before: =�Prefix[8].value = 1

Before: =�Prefix[9].value = 1

Before: =�Prefix[10].value = 1

...

...

...

Before: =��� Prefix[9987].value = 1

Before: =��� Prefix[9988].value = 1

Before: =��� Prefix[9989].value = 1

Before: =��� Prefix[9990].value = 1

Before: =��� Prefix[9991].value = 1

Before: =��� Prefix[9992].value = 1

Before: =��� Prefix[9993].value = 1

Before: =��� Prefix[9994].value = 1

Before: =��� Prefix[9995].value = 1

Before: =��� Prefix[9996].value = 1

Before: =��� Prefix[9997].value = 1

Before: =��� Prefix[9998].value = 1

Before: =��� Prefix[9999].value = 1

^C

**********************************************************************************************************************

P/D: We compile Charm++ with:

  ./build charm++ mpi-linux-x86_64 --with-production -j8

and, also: 

  ./build charm++ mpi-linux-x86_64 smp --with-production -j8

--

Pedro Esequiel Tarazi

Estudiante de Ingeniería en Computación

Facultad de Ciencias Exactas, Físicas y Naturales - Universidad Nacional de Córdoba

    

--

Pedro Esequiel Tarazi

Estudiante de Ingeniería en Computación

Facultad de Ciencias Exactas, Físicas y Naturales - Universidad Nacional de Córdoba

    

--

Michael Robson <mprobson AT illinois.edu>

PhD

 Candidate in Computer Science

University

 of Illinois at Urbana-Champaign

http://charm.cs.illinois.edu/people/mprobson