LINUX CLUSTER: vendors & critical issues

[Rob Myers]

ivan-

1. Test your code on each platform you are looking at.  Vendors are
likely to let you benchmark your code for such a large purchase.

2. Don't forget that heat and power will be an issue for a large
cluster.

3. Be aware that the proprietary intel compilers are free.

happy numerical modeling!

rob.

On Mon, 2002-06-10 at 20:57, Ivan Oleynik wrote:
> Dear Colleagues,
> 
> We are going to acquire a Linux cluster for running MPI type
applications
> such as molecular dynamics, quantum mechanical density functional
plane
> wave and localized orbitals programs. This involves heavy numerical
> calculations including standard linear algebra such as
> eigenvalues-eignevectors solvers (sca-LAPACK), integration of Newtons
> equations of motions for up to million particles, etc. We prefer to
get as
> "turnkey" system as possible would like to avoid studying hardware
issues
> related to Beowulf DIY.
> 
> I would really appreciate if someone could recommend a good vendor,
who
> enthusiastically works with academics, understands and adequately
responds
> to customer needs and requirements, provides good warranty support of
> hardware and, the most importantly, sells reliable state-of-the-art
> cluster systems.
> 
> I would also like to get some technical advice regarding critical
issues
> we need to keep in mind during the process of choosing configuration.
To
> be specific, I am thinking of 24 node cluster with the following node
> configuration: dual AMD Athlon 2000+ (2100+) MP processor, 2Gb DDR
SDRAM &
> 40Gb IDE hard drive. Obviously, the most important question is the
choice
> of interconnect.  I am aware of 3 standard options: Gigabit Ethernet,
> Myrinet, and Dolphi Wulfkit. I would appreciate very much if someone
could
> share an experience regarding which of these 3 choices gives an
optimum
> balance of performance (especially for heavy numerics) and price.
> 
> 
> Thanks very much,
> 
> 
> Ivan Oleynik
> 
>
------------------------------------------------------------------------
> Ivan I. Oleynik                       E-mail :
oleynik at chuma.cas.usf.edu
> Department of Physics
> University of South Florida
> 4202 East Fowler Avenue                  Tel : (813) 974-8186
> Tampa, Florida 33620-5700                Fax : (813) 974-5813
>
------------------------------------------------------------------------