yade-dev team mailing list archive

[Anton Gladky <724396@xxxxxxxxxxxxxxxxxx>]

I have done some performance tests on 6-core machine.

The simulation have been carried out with different thread numbers and
execution time has been measured. In the first case it was the latest
trunk revision (2781) clean compile, in the second case the
ForceContainer used only "non-parallel flavor" even in multy-threading
mode, so the "parallel flavor" have been deleted from ForceContainer.hpp
(patch provided).

Number of bodies was 47412, number of iterations was 5000.

Results are presented below (execution time).

-j     Parallel           Non-Parallel             %
------------------------------------------------------------
1       997                     984                      1.3
2       821                     800                      2.6
3       764                     736                      3.6
4       752                     709                      5,7
5       745                     699                      6,2
6       742                     693                      6,6

So using non-parallel ForceContainer makes the simulation sufficiently
faster (from 1 to 6% and more).

My proposal is to delete parallel implementation of ForceContainer. It
will increase the productivity and "fixes" this bug.

What ideas and opinions?

** Attachment added: "Patch deletes parallel implementation of ForceContainer"
   https://bugs.launchpad.net/yade/+bug/724396/+attachment/1895037/+files/p

-- 
You received this bug notification because you are a member of Yade
developers, which is the registrant for Yade.
https://bugs.launchpad.net/bugs/724396

Title:
  crash after add particles in parallel mode

Status in Yet Another Dynamic Engine:
  Confirmed

Bug description:
  Yade crash in parallel mode if to add particles after run of simulation.
  Attached script runs without errors in serial mode (-j 1 option),
  but in parallel mode (-j 2 or more) it crash at second O.run.