[Paradiseo-help] Trying to use ParadisEO-PEO

Clément Perrod clement.perrod at gmail.com
Mar 12 Mai 16:26:05 CEST 2009


I still working with the (C)VRP-TW contribution to do some statistics
about parallelisation. I have a grid (16 nodes with 16 processors quad
cores each) and I wanted to run the application with the evaluation
function parallelized.

I launched the application:
- without any parallelisation
- with the lesson 3 applied with 5, 10, 15 and 20 processors

As expected, the parallelisation with 5 processors gives me better
results than no parallelisation. But what a surprise when I found the
algorithm lasts the same time using 5, 10, 15 or 20 processors! (I
still wait my results for a test with 100 processors)

I am looking for more information about the parallel evaluation.
Assuming I run my algorithm on 10 processors, what is really gonna
happen on my application? If I understand what you wrote earlier the
main node will send evaluations one by one to each other nodes as soon
as a node is free. But if that is true why do I have the same results
with any number of processors?

It is also possible that I do not launch it correctly. I mean, maybe
my `schema.xml' is wrong but there is nowhere I could find explanation
(and I am not sure about what I could understand on the "Technical
Introduction"). If I want to launch a application with 20 processors
what should my `schema.xml' look like?

I tried something like that:

<?xml version="1.0"?>

 <group scheduler="0">
          <node name="0" num_workers="0">
          <node name="1" num_workers="0">
          <node name="2" num_workers="1">
          <node name="19" num_workers="1">

As it is not easy to understand your wrapper over MPI I am asking my
questions here.

Thanks for your time,

Best regards,

Clément Perrod
EPITA 2009
Operational Research - Resources Optimisation

Plus d'informations sur la liste de diffusion Paradiseo-help