[petsc-users] Software for load balancing to improve parallel performance to be used with PETSc

[TAY wee-beng]

Hi Jed,


On 5/1/2012 4:59 PM, Jed Brown wrote:
> On Thu, Jan 5, 2012 at 09:41, TAY wee-beng <zonexo at gmail.com 
> <mailto:zonexo at gmail.com>> wrote:
>
>     I just did a -log_summary and attach the text file, running across
>     8 and 16 processors. My most important concern is whether the load
>     is balanced across the processors.
>
>     In 16 processors case, for the time, it seems that the ratio for
>     many events are higher than 1, reaching up to 6.8 for VecScatterEnd
>
>
> This takes about 1% of the run time and it's scaling well, so don't 
> worry about it.
>
>     and 132.1 (?) for MatAssemblyBegin.
>
>
> This is about 2% of run time, but it's not scaling. Do you compute a 
> lot of matrix entries on processes that don't own the rows?

I only compute rows which the processor own. Can it be the memory 
allocation? I'll check on that.
>
> Most of your solve time is going into PCSetUp() and PCApply, both of 
> which are getting more expensive as you add processes. These are more 
> than 10x more than spent in MatMult() and MatMult() takes slightly 
> less time on more processes, so the increase isn't entirely due to 
> memory issues.
>
> What methods are you using?

What do you mean methods? I am doing Cartesian grid 3d CFD, using 
fractional mtd which solves the momentum and Poisson eqns. I construct 
the linear eqn matrix and insert them in PETSc matrix/vectors. Then I 
solve using Bicsstab and hypre AMG respectively. Why is PCSetUp() and 
PCApply using more time?
>
>     However, for the flops, ratios are 1 and 1.1. so which is more
>     important to look at? time or flops?
>
>
> If you would rather do a lot of flops than solve the problem in a 
> reasonable amount of time, you might as well use dense methods. ;-)

Thanks again!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120109/bee728cb/attachment.htm>