[Bioclusters] parallel blast???

Steve Gaudet bioclusters@bioinformatics.org
Tue, 17 Sep 2002 08:38:41 -0400

Hello Chris,

> Be careful with your benchmarks as they can be meaningless or=20
> misleading. You will find that the speed of blast distributed=20
> within a=20
> cluster or compute farm is directly related to 2 things: (a)=20
> the amount=20
> of physical memory in the compute  nodes and (b) the speed of your=20
> storage or disk I/O system.
> You can have the fastest server on earth but if you searching with=20
> blast against an NFS mounted database and your network or=20
> fileserver is=20
> slow then your blast searching speeds will be horrible. Give=20
> me a small=20
> number of speedy linux boxes and I can bring a $300,000=20
> NFS/NAS system=20
> to its knees. Storage does matter.

Anyone ever look or try solid state disks?

> Blast performance also depends on you tune your DRM=20
> (gridengine or LSF=20
> etc. etc.) and how  you adjust your workflow with respect to=20
> splitting=20
> large databases, locally caching data on compute nodes etc. etc.
> What are you trying to benchmark for? Picking the right CPU? Some=20
> people on this list may have already done this. My personal=20
> preference=20
> is Intel Pentium III's right now because:
> o P IV's are way too expensive
> o P III's are dirt cheap
> o There are a ton of dual-CPU motherboard options for the=20
> PIII allowing=20
> me flexible choices of system packaging and vendor
> o Athalon / AMDs are super fast but your motherboard choices are=20
> limited and you need to be really  careful about cooling and=20
> ventilation

Thank you,

Steve Gaudet=20
Linux Solutions Engineer
| Turbotek Computer Corp.    tel:603-666-3062 ext. 21             |
| 161 Abby Rd                fax:603-666-4519                     |
| Manchester, NH 03103       e-mail:sgaudet@turbotekcomputer.com  |
| toll free:800-573-5393     web: http://www.turbotekcomputer.com |