[Bioclusters] RE: sge and mosix [was: Journalist request: linux cluster user feedback]

Hunter Matthews bioclusters@bioinformatics.org
20 Sep 2002 16:33:55 -0400

On Fri, 2002-09-20 at 03:43, Alan Hart wrote:
> As below ...
> Let me know if you want any more detail.
> Alan Hart 
> UNIX Manager 
> Oxford GlycoSciences

> > - How long have you had your Linux cluster running?
> > 
> 	1 year

THis is about 12 more months than I'll be doing it. :)

> > - What type of distributed resource management software (ie, LSF, PBS, Sun
> > Grid Engine) do you use?
> > 
> 	Mosix Kernel level load sharing with Sun Grid queuing resource on
> top

Can you give an utter newbie (to clusters, anyway) any suggestions or
pointers on this? 

I've installed our first beowulf's first 6 nodes, and had planned next
week to try to evaluate for my users whether sge, mosix, condor, or
whatever would be the best fit. 

SGE and mosix were at the top of my "I wonder if that would work" list -
but this is the first I've ever heard of them being used together. It's
appealing, because you'd get the benefits of SGE's initial queue
management and load balancing with mosix's later load balancing.

How well does it work? How hard was it to setup together? Would you do
it that way again or just one?

Our needs here are small today, but growing fairly rapidly - paup and mr
bayes are the heaviest uses, followed by clustalw, custom simulations
and I'm starting to get some blast requests.

None of the programs above are parellized, and nearly none of our
faculty/grads know about parallel programming - I certainly don't. I was
going to try and tailor my solution to that - namely trying to run as
many serial jobs as possible, in the simplest (to my users anyway) way
as possible.

Can anyone recommend some good real world guides to a biological
Hunter Matthews                          Unix / Network Administrator
Office: BioScience 145/244               Duke Univ. Biology Department
Key: F0F88438 / FFB5 34C0 B350 99A4 BB02  9779 A5DB 8B09 F0F8 8438
Never take candy from strangers. Especially on the internet.