Re: cluster of PW433au's
Am Mon, 2003-02-10 um 15.52 schrieb terry.bowling@verizon.com:
> With all of our layoffs, be have an abundance of spare Digital PW433au's.
> I have convinced my boss to loan a pair to the local university, Indiana
> Institute of Technology (IIT) where I teach. It will be a student led
> project to get a single Master & Slave cluster going. If they can do this,
> they get more. Probably a maximum of 10 nodes with primary and secondary
> masters.
>
> Unfortunatley, I am extremely interested in this but have not had time to
> participate or help with research. They basically need 2 things.
> Suggestions/advice on configuration and management tools, and
> suggestions/advice on what to do with it when it's finished. Right now,
> they're thinking of letting it be used by SETI.
>
> The other issue - which distro. I gave them copies of RH7.2 when I gave
> them the hardware. But the student who is leading it wants to use Debian.
> Does anyone have experience to say which would be better interms of the
> clustering software available? I've read great things about Scyld's
> cluster software, but it looks like it's only x86 and I couldn't find the
> source to download. Right now I'm reading the references on www.lcic.org.
>
> Thanks in advance.
> ----
> Terry Bowling
> Verizon NOTD Systems Support
> 260.461.3772
OpenPBS and LAM-MPI are two things they should investigate. I'd suggest
using Debian, too. At university I build a 10 Machine cluster with SuSE
(I *had* to use it), which sucks... apt just rocks if you have several
headless machines to keep up to date.
A good parallel program which needs a lot of CPU power is tree-puzzle
(www.tree-puzzle.de), a program to calculate evolutionary trees from
nucelotide or protein sequences. You can actually run one computation on
several nodes using lam-mpi or mpich, it also scales very good. The code
is GPL'ed, so your students can study the parallisation code and maybe
help improve it. Heiko Schmidt of Max-Planck-Institute for Molecular
Genetics, the maintainer, is a nice guy to deal with. Maybe there is a
Institute for Biology at your university which could get involved, too.
Bioinformatics is the THE science with the biggest need for number
crunching for the next few years, I think.
hth,
Jan Lentfer
--
Jan Lentfer
System Administrator
Molecular Cell Biology / AG Holstein, Darmstadt University of
Technology, Schnittspahnstr. 10, 64287 Darmstadt, Germany
Tel: +49 6151 16 5563 / Tel private: +49 6155 899393 / mobile: +49 163
4712037
Reply to: