Archives


- Beowulf
- Beowulf Announce
- Scyld-users
- Beowulf on Debian

[Beowulf] GPFS on Linux (x86)

Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.

Search

Toon Knapen toon.knapen at fft.be
Thu Sep 14 01:05:49 PDT 2006


We have tested GPFS and got some interesting results.

First some background data: we performed our tests at IBM's european HPC 
center. In collaboration with IBM, we tried to find the 'ideal' (in 
terms of minimizing wallclock-time) cluster-configuration to run our 
acoustics simulation software. Due to the size of the models we treat, 
we generally use a direct parallel out-of-core solver. Since it is an 
out-of-core solver, disk IO is pretty intensive.

So using our out-of-core solver, we tested our app. using local scratch 
disks and compared that to using a GPFS. The main advantage in the end 
to me is the flexibility of the system. If some node needs more 
scratch-space, this can be easily configured. Also the scalability was 
very good. Using 4-core nodes, we also tested the scalability by having 
2 or 4 processes using the GPFS simultaneously. The effect of having 
multiple processes performing IO simultaneously did however not have a 
big effect on the individual bandwidth.

toon

Brent Franks wrote:
> Hello,
> 
> We have been researching GPFS from IBM and are very interested in what
> looks like promising performance and scalability.
> 
> Does anyone use GPFS, we are looking at running it in a distributed
> mode, rather than using one central SAN.
> 
> Additionally, the sales information seems to be rather elusive.  We
> have contacted IBM, however the channels we contact have all been
> somewhat dissapointing as we have had to educate the sales people on
> what the product is.
> 
> We are very interested in setting up a trial of the software and then
> purchasing if it meets our needs.  Does anyone have a contact at IBM
> or solutions provider who is familiar with GPFS and could send out an
> eval kit?  30 or 60 day time bomb would be fine, we just need to
> justify to management that it will indeed work for our application.
> 
> Any help would be greatly appreciated.
> 
> Thanks and kindest regards,
> 
> - Brent
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit 
> http://www.beowulf.org/mailman/listinfo/beowulf
> 


-- 
Toon Knapen

------------------------------------------------
Check out our training program on acoustics
and register on-line at http://www.fft.be/?id=35



More information about the Beowulf mailing list