Charm++ parallel programming system

[Junchao Zhang <junchao.zhang AT gmail.com>]

I avoided the problem by building charm on Ranger with ./build charm++ mpi-linux-x86_64 mpicxx --no-build-shared -O3,

and substituted mpirun with ibrun in the generated charmrun script under ChaNGa directory

Thanks
--Junchao Zhang

On Wed, Aug 10, 2011 at 10:06 AM, Junchao Zhang <junchao.zhang AT gmail.com> wrote:

Hi,

  When I run Charm on Ranger of teragrid, I met some problems. I built charm with  ./build charm++ net-linux-x86_64  ibverbs icc

  I need 432 cores to run an application, ChaNGa.  Each node has 16 cores, so 27 nodes are enough, but charm says running on 28 nodes.

charmnodes >nodelist

./charmrun ++scalable-start +p 432 ./ChaNGa $HOME/input/lambb.param  +balancer OrbLB

Charmrun> scalable start enabled.

Charmrun> IBVERBS version of charmrun

Warning> Randomization of stack pointer is turned on in kernel, thread migration may not work! 

Charm++: scheduler running in netpoll mode.

Charm++> Running on 28 unique compute nodes (16-way SMP).

...

Chip #22: 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 

Chip #23: 368 369 370 371 372 373 374 375 

Chip #24: 376 377 378 379 380 381 382 383

 

Also, I find many messages like

OrbLB Info> PE 6 with 52.525892 background load will have 0 object. 

OrbLB Info> PE 8 with 55.647333 background load will have 0 object. 

OrbLB Info> PE 10 with 85.024853 background load will have 0 object.

 I also find the execution time is unreasonably long.

 Does it imply something wrong?  Could someone help me?

Thanks a lot

-- Junchao Zhang

--
-- Junchao Zhang

--
-- Junchao Zhang