← Back to team overview

yade-users team mailing list archive

Re: [Question #240608]: How to increase the computing scale in YADE

 

Question #240608 on Yade changed:
https://answers.launchpad.net/yade/+question/240608

    Status: Answered => Open

Fu zuoguang is still having a problem:
Dear Bruno Chareyre and all users:
       Thanks for helping me these days and I should say sorry for my misunderstanding of large problem running by assigning N threads for them in Yade.

       (1). My original expectation is that large problem can
successfully run in an improved hardware by assigning them as many
threads as possible for running. I considered that Yade-batch-mode may
provide me this chance but it is not the case. The role of Yade-batch-
mode is to run as many jobs as possible for using effectively the
hardware’s resources. And now it rarely works for improved hardware
(more CPUs and Memories than ordinary PC) to run just a single task
since that the ‘InsertionSortCollider, which is one of the most
important key role in Yade simulation and may take the most of time in
the whole process of calculation, defaultly works in one thread no
matters how many threads can be used. Yades computational codes now
provide little help for running large problems effectively. This
conclusion is from what you said:

Worst than "no good parallelized", yade runs contact detection on one
thread whatever the number of available threads. For more than 100k
elements and j>1, this can be 70% of the computation time or more.

I hope that I am right here.

       (2). You suggested that a parallel version of the collider could
be used for getting an effective running process in large problem. I
have read the description about your testing records of it in [1]. One
of the most important results is that the total running time reduces
from 4.08 hours to 2.14 hours with 200813 particles when ‘yade-trunk' is
replaced by ‘yade- parallel’. And the percent of time taken by
'InsertionSortCollider’is from 64.02% to 33.42%. It seems like that that
is a good way for treating large problem in Yade. I want to know whether
or not I can do the same testing in my own hardware. If I can, what can
I do for this purpose ?

Seeking your help !

[1] https://lists.launchpad.net/yade-dev/msg10498.html

-- 
You received this question notification because you are a member of
yade-users, which is an answer contact for Yade.