[Bioclusters] Direct connect infiniband/quadrics?

[Tim Cutts]

On 15 Mar 2005, at 11:19 pm, Farul M. Ghazali wrote:

> Hi all,
>
> Thanks for the responses so far. The applications in use are molecular
> dynamics applications (mostly amber, some gromacs) and Autodock. There
> will be others as well but they won't be taking up a lot of CPU time. 
> Just
> going from a dual to a quad-CPU machine on Amber makes a big 
> improvement
> as far as I can tell.
>
> I'll most likely be using Rocks on this cluster or manually set it up 
> if
> need be. For the time being at least it's small enough to manage (2 
> nodes
> :-)
>
> In terms of expansion, I'm trying to push through quad-CPU nodes over
> duals to minimize the cost of the interconnect if the interconnects 
> prove
> to be faster. I haven't looked at 8 or more CPU Opterons yet, but I'd 
> like
> to keep away from non-standard configurations eg. the Cray XD1.


Did you consider an 8-way SGI Altix 350?

It might suit your application well, and might even fit in that budget 
(just).

The NUMALink interconnect is very fast indeed.  Although I'd probably 
be tempted to go for a few more nodes, and use GBit for the 
interconnect.

How many jobs do these people typically want to run at a time?  If it's 
more than one at a time, then buy four quad-CPU Opteron nodes, GBit 
connected, and run up to four 4-CPU jobs at a time.  No need for the 
MPI to use the interconnect then, and the throughput of the cluster 
will be very good (although individual job turnaround will be slower).  
It's worth considering.

Tim

-- 
Dr Tim Cutts
Informatics Systems Group, Wellcome Trust Sanger Institute
GPG: 1024D/E3134233 FE3D 6C73 BBD6 726A A3F5  860B 3CDD 3F56 E313 4233