[Beowulf] Digital Image Processing via HPC/Cluster/Beowulf - Basics

Reuti reuti at staff.uni-marburg.de
Sun Nov 11 09:32:33 PST 2012


Am 05.11.2012 um 22:52 schrieb Rayson Ho:

> IIRC, ScaleMP is another distributed shared memory implementation, and
> both software and hardware (SGI NUMALink machines?) implementations
> are more expensive and less fault tolerant than shared nothing nodes
> managed by some sort of cluster software.

I wouldn't say that this is disjunct - you can also use a queuing system in a good way on such a machine IMO. But the idea was more: you have an application which is only Open MP parallel but scales in a nice way. Hence you could run larger jobs on such a machine. Is it worth to try it (and regarding this setup I was asking, whether anyone is using it and has experience - it's like running OpenMP across nodes).

-- Reuti


> Since rendering is mostly an embarrassingly parallel workload, it is
> way easier to install & run something like Rocks cluster (or even
> Platform-HPC), and it is not going to run any slower than on a ScaleMP
> "cluster".
> 
> Rayson
> 
> ==================================================
> Open Grid Scheduler - The Official Open Source Grid Engine
> http://gridscheduler.sourceforge.net/
> 
> 
> On Sun, Nov 4, 2012 at 12:48 PM, Reuti <reuti at staff.uni-marburg.de> wrote:
>> What about ScaleMP / Kerrighed to make a single machine out of many - has anyone experience with it and uses it in a cluster?
>> 
>> http://www.scalemp.com/ http://www.kerrighed.org/wiki/index.php/Main_Page
>> 
>> -- Reuti
>> _______________________________________________
>> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
>> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf




More information about the Beowulf mailing list