Query: Blast on Scyld beowulf

[Tim Chipman]

I apologize if this topic has been hashed over a million times ; I've just 
joined the list, and read through the last year worth of archived postings. No 
clear success so though I should ask.





I'm curious about the feasibility of running sequence analysis apps (blast, 
hmmer, fasta) on a scyld-beowulf setup.





I have seen some mention of "demos done by tigr at recent conferences of blast 
on beowulf" ; other mention that "no free parallelized blast is yet available, 
use fasta instead" ; and a great archive of discussion on the topic at the 
URL,

 http://sapiens.wustl.edu/~ikorf/beowulf.html , which highlights the 
differences / possible more appropriate arrangements of batch job queues vs. 
parallelized jobs.

Additionally, most discussions I've found on the 'net (via google searches) 
pre-date scyld beowulf and typically take the approach of earlier more 
traditional beowulf clusters where each node had a full OS install, apps & 
data maintained locally (or via NFS shares) & the like.

Ideally, what I'm looking for is comments on how one can setup things to work 
with a scyld beowulf (hence more straightforward to setup / maintain) and what 
(if any) tweaks are required to encourage relatively efficient use of 
resources.

I also suspect that the soon-to-be-available new scyld beowulf (which includes 
PVM out of the box, not present previously?) will be a significant help.. but 
of course I'm still a tad unclear. Sigh.

All this to say: If anyone has any pointers or comments relating to this topic 
of (dna,protein) sequence analysis / comparison on scyld beowulf (even if it 
is to say, "blargh! you must you *this* 'other' implementation..") ... then I 
will certainly very much appreciate if people can e-mail me such info.

If I get a significant number of replies, I'll probably post a summary to the 
list of the salient points, just to create a reference which will help prevent 
other newbies like myself from asking the same questions in the near future.

Thanks very much,


Tim Chipman