[lug] psuedo OT: advice on learning clusters

[hugh at math.byu.edu]

> On Sun, 2004-02-22 at 23:32, Hugh Brown wrote:
>
>> I'm interested in a computational approach and developing general
>> competency in a scientific cluster environment.
>
>
> Hi Hugh,
>
> Heres a few links that you might find helpful:
>
>   OpenMosix, a "fork and forget" kernel-based way to distribute
>   workloads across multiple x86-based systems:
>     http://openmosix.sourceforge.net/
>     [ Also a great way to turn two or more cheap-o systems into
>       a cute and useful little cluster...]
>
>   A few "batch" or "queue oriented" ways to distribute loads
>   across multiple systems:
>     http://www.openpbs.org/
>     http://www.cs.wisc.edu/condor/
>       http://www.cs.wisc.edu/condor/condorg/versusG.html
>     http://gridengine.sunsource.net/
>     http://www.platform.com/products/LSFfamily/
>     http://www.mhpcc.edu/training/workshop/loadleveler/MAIN.html
>
>   Libraries for between-node communications:
>     http://www-unix.mcs.anl.gov/mpi/mpich/
>     http://www.csm.ornl.gov/pvm/pvm_home.htm
>
> All of the above have overview or FAQ pages that will help you get an
> idea of what they do and how they could fit together.
>
> Since the early 90's, "High Performance Computing" has become more or
> less synonymous with parallel computing.  And, while the way one
> approaches parallel computing is (or should be) strongly dependent on
> the problem features, the MPI library is used on a broad range of
> applications.  For instance, its used routinely for our main project at
> work and it runs on many systems:
>
>   http://mitgcm.org/testing.html
>
> So MPI is a good place to start if you'd like to do a little coding.
>
> If you're mostly interested in the cluster admin aspects then
> concentrate on the other groups of links.  And for a concrete cluster
> example, you can see the user pages for our cluster at:
>
>   http://mitgcm.org/projects/MITGCM_CLUSTER/
>
> And remember, with just two (cheap) machines you can quickly build your
> own (genuine!) Linux cluster at home and try out almost all of the these
> cool open-source tools!  ;-)
>
> Ed
>


Thanks much.  I was noticing that dell is selling Dimension 2400n machines
w/o an operating system for $319 each (you have to click on the small
business link to find them).  I figured I could buy four (and pick up some
more ram from crucial.com) and begin my clustering fun (I just have to
persuade the purchasing committee :)