Re: How do you setup an Alpha cluster

["Wes Bauske" <wsb paralleldata com>]

Gary A. Allen wrote:
> 
> Folks,
> 
> A bunch of us at the STA Branch of the NASA Ames Research Center are sold on
> the idea of using an array of Alphas running on PCI bus motherboards to do
> scientific computations.  Our ideas are currently embryonic:  We are thinking
> of having several desktop Alphas with Linux and Windows-NT partitions
> (Windows-NT for shrink-wrap software and Linux for doing real scientific work)
> and also having a large cluster of Alphas dedicated to scientific computations
> using parallel algorithms. The large cluster would run with Linux as the
> operating system and use applications compiled with the best legal compiler
> available for Linux.  Our machines would all be based on the new OEM Samsung
> chips and motherboards.   All of the generic PC stuff (cases, CRTs, keyboards,
> video cards, hard disks, etc.) would be purchased at bottom dollar from local
> vendors.  We would assemble the machines in-house and wire them together using
> high speed Ethernet connections.
> 
> The basic idea is our engineer/scientists would use their desktop machines
> for code development or pure scalar computations and farm out their extremely
> parallel algorithm CPU intensive work to the Alpha cluster.  The primary
> scientific application would be trajectory and computational fluid dynamic
> simulations of spacecraft entering planetary atmospheres.
> 
> Great minds work alike!  I understand that a bunch of people at Los Alamos and
> other places have built up Alpha clusters based on PCI bus motherboards and
> the special effects for Titanic came from the same sort of architecture.  I'm
> a newbie humbly seeking advise from people who have actual experience building
> these systems.  Is this basic concept a dumb idea?   For example: would fast
> Intel or AMD chips running Linux be a better approach.  Is using Linux a dumb
> idea?  Should I (shudder) be using only Windows-NT?  What are the hidden
> pitfalls one encounters with an Alpha cluster that do not appear until after
> the system has been built and running for a few weeks?  Is using an Ethernet
> message passing approach the best basis for an Alpha cluster running parallel
> code?   What's the optimal top CPU speed for an Alpha cluster?  For example
> if the Ethernet latency is the primary data bottleneck then a cluster based
> upon the more expensive 800 Mhz. CPUs might perform no better than 533 Mhz.
> CPUs.  Our goal is to maximize CPU-cycles/dollar.  We understand that there
> are faster architectures out there (Cray T3D) but we can not afford them.
> 
>                             Thanks in advance!
>                             Gary A. Allen, Jr.
> 

Wait for the 21264's. They should be out within a month or so.
I found SPECINT95/FP95 on them of 30/45. Compare that to the
fastest Intel's of 16.5/13.7. So, depending on whether you do
int work or FP work, you're 2-3 times faster on the 21264.
I'd like to see a Linpack on the 21264 too but haven't found it.

Also, if you're not set on Alphas, wait until Oct. 5 and see
what IBM is going to announce. (tested it under non-disclosure)
Quite good price/performance if you can work a deal with your
local IBM rep. Those are my top two boxes I'm looking at now.
Oh yeah, IBM's boxes run headless without a fuss. You need
a serial switch box though to get to their consoles. The Alphas
I've seen require a KVM switch and video cards in them to
function properly. I find KVM switches quite useful. I run
two KVM's against 10 mixed PC/Alpha/IBM boxes.


Wes