Slow Pmode on HPC

Currently I'm working on a particle simulation code on Matlab/pmode for my class project. Running the code on my laptop with 4e5 particles, which is working fine. To use more particles (such as 1e6) and finish the simulation faster, I'm connecting to university's High Performance Computer. But I'm in trouble with the slowness of the HPC.

First of all these are the things I do:

• 1)Connecting to university's HPC server with ssh -X command. (it has Matlab 2008a).
• 2)Validating the configuration "mytorque".
• 3)Starting pmode by typing: pmode start mytorque 32
• 4)Pmode pops up, copy pasting the code and running it (same as I do on my laptop)

The code consist of a few thousands of lines and many functions. Working with the composite arrays as you may guess. So I'm summarizing the code:

...
%initialization (creating composites) and some calculations 
...
while t<1e7
...
%some calculations etc
...
labBarrier;
t_n=gplus(n); %communication
labBarrier;
...
%some calculations etc
...
t=t+1;

end

As you may see there is only 1 command (gplus) for communication, other parts are independent of each other. It is working much slower than my laptop on the HPC server (even with the 32 workers). If I cancel the gplus (communication) part of the code, it gets very fast. Seems like slowness is due to slow communication between the workers and the client. What can be the reason of this problem, any ideas?

By the way this parallel code works perfectly on my laptop (using 4 workers / i7 processor).

Thank you

[Amita Giri]

Have you found the problem?