[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] Solution for HPC



Ok I understand. Anyway thanks for explanation. My software is Blender. I dont know why for such a thing like fluid dynamics somebody somebody coded software without multithreading. I will ask on Blender forum maybe there is any good version with support for multithreading. Thanks for fast and profi support :) I appriaciate.

Konrad



On Wed, 20 Apr 2011 09:05:52 +0000, "Jankowski, Chris" <Chris Jankowski hp com> wrote:
Konrad,

The first thing to do is to recompile your application using a
parallelizing compiler with proper parameter equal to the number of
cores on your server. This of course assumes that you have the source
code for your application.

For a properly written Fortran and C application a modern
parallelizing compiler would do a great job.

Note that today you may have easily 48 real physical cores i.e. 96
independent parallel threads of execution with hyperthreading turned
on a modern Intel x86_64 server such as HP DL980 G7.

Then the next step is to tune the application on the source code
level to increase its parallelism such that it can actually use the 96
threads.

Only then, if the elapsed time of your processing is still
unacceptably long (weeks), you would move to a HPTC cluster. This is
very expensive - the Infiniband interconnects do not come cheap and
you still need to put in a few man years of work to tune your code for
the HPTC cluster.

I hope this helps.

Regards,

Chris Jankowski

-----Original Message-----
From: linux-cluster-bounces redhat com
[mailto:linux-cluster-bounces redhat com] On Behalf Of dlugi
Sent: Wednesday, 20 April 2011 17:50
To: linux-cluster redhat com
Subject: [Linux-cluster] Solution for HPC

 Hi Gurus,

    I would like to ask You about something. Since few days I`m
 preparing 3D fluid simulation. The problem is that my simulation is
rendered only on one core. CPU usage provading information that only 1 core is 100% used by process. In my opinion this software doesnt support
 multithreading thats why everything is calculated on one core.

 Is it possible to build some kind of HPC cluster where this single
 process could be distributed for several machines ?
I`m not thinking about dividing this job for several small peaces and
 distributing them. I`m thinking about infrastructure where single
 process could use CPU power from several machines at the same time.

 Is it possible to do this on RH or Fedora ?

 cheers

 Konrad

--
Linux-cluster mailing list
Linux-cluster redhat com
https://www.redhat.com/mailman/listinfo/linux-cluster

--
Linux-cluster mailing list
Linux-cluster redhat com
https://www.redhat.com/mailman/listinfo/linux-cluster


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]