Re: [moose-devel] Parallel execution of Models with HSolve

[Saeed S. <sd....@gm...>]

Dear Upi,

Thank you. I'll have a look at the Vivek's code.
But so far, everything working properly. I did develop a Master Solver to
get a bunch of model, put some parts in GPU and keep CPU parts intact.
Therefore, the GPU version works beside of CPU and can extend gradually.
The documentation is almost ready and I will send it soon.

But still my main problem is to set benchmark and compare the performance.
It would be my great pleasure if any of the moose developers guide me, how
to simulate models with hsolve in parallel, efficiently.


Best Regards,
Saeed



> Dear Saeed,
>      This is very good news! Vivek had gotten GPUs to run from MOOSE but
> not the full implementation. Perhaps you may wish to look at his code to
> see how to fold into MOOSE.
> Best,
>      Upi


> On Wednesday 24 September 2014 06:47 PM, Saeed Shariati wrote:
> Dear all,
>
> The GPU version of HSolve is almost ready and I'll put codes in svn as
soon
> as solving some issues.
>
> But, to test the performance, I couldn't utilize Moose to run models with
> hsolve objects in multicore systems.
>
> As an alternative, I could manage use MPI version available in _test_
> directory and assign an instance to each core. But it doesn't support
> hsolves.
>
> Would you please guide me what is the best way to run moose on multicore
> systems and afterwards cluster?
>
> Thank you in advance,
> Best Regards,
> Saeed
>
------------------------------------------------------------------------------
> Meet PCI DSS 3.0 Compliance Requirements with EventLog Analyzer
> Achieve PCI DSS 3.0 Compliant Status with Out-of-the-box PCI DSS Reports
> Are you Audit-Ready for PCI DSS 3.0 Compliance? Download White paper
> Comply to PCI DSS 3.0 Requirement 10 and 11.5 with EventLog Analyzer
>
> _______________________________________________
> moose-devel mailing list
> moose-devel@...
> https://lists.sourceforge.net/lists/listinfo/moose-devel​

On Wed, Sep 24, 2014 at 10:17 AM, Saeed Shariati <sd....@gm...>
wrote:

> Dear all,
>
> The GPU version of HSolve is almost ready and I'll put codes in svn as
> soon as solving some issues.
>
> But, to test the performance, I couldn't utilize Moose to run models with
> hsolve objects in multicore systems.
>
> As an alternative, I could manage use MPI version available in _test_
> directory and assign an instance to each core. But it doesn't support
> hsolves.
>
> Would you please guide me what is the best way to run moose on multicore
> systems and afterwards cluster?
>
> Thank you in advance,
> Best Regards,
> Saeed
>