[EMBOSS] Parallelizing Emboss

[Guermonprez, Paul]

hello,

Glad to hear it, a good target may be "water",
a smith waterman implementation.
We had a similar topic on [emboss-dev] lately.

Of course before parallelizing an algo the first step is
to optimize it in serial mode, you may find information
about emboss-water optimization done at intel
a few months ago here :

please find the pdf of the slides here :
http://www.guermonprez.net/projects/emboss/emboss-EBI_06_06.pdf
and the sources here (need emboss 3.0.0 to work) :
http://www.guermonprez.net/projects/emboss/emboss-runsh.tar.gz

the software speed gain ( x11 ) was measured on intel architecture
but you can get more or less the same on different hardware too.

regards, paul.


Paul Guermonprez - Intel
Sr. Software Engineer - Digital Health EMEA
email  : paul.guermonprez at intel.com
phone  : +33 1 58 87 72 41
mobile : +33 6 26 23 67 62


-----Original Message-----
From: emboss-bounces at lists.open-bio.org
[mailto:emboss-bounces at lists.open-bio.org] On Behalf Of praneet
Sent: Friday, November 03, 2006 12:59 PM
To: emboss at lists.open-bio.org
Subject: [EMBOSS] Parallelizing Emboss

Hello Everyone,

We at Sun Microsystems just got started with Emboss. So, apologies if 
you don't find the questions asked below as particularly bright.
We are evaluating Emboss and trying to figure out if it can be
parallelized.
Eventually, Emboss* may be* available on Sun Grid 
(http://www.sun.com/service/sungrid/overview.jsp and 
http://www.network.com/).


I've a few  questions

   1. What do you think is the best way to parallelize Emboss? best here
      means we are ready for a cost performance trade-off. Our options
are
         1. Input partioning or
         2. Writing mpi constructs
   2. Are there any existing guidelines/ reference implementation/ docs
      or any relevant material pertaining to parallelizing Emboss?
   3. Out of the 160 applications, which ones do you think should be
      parallelized? We don't have time to parallelize all applications
      if we go the mpi route. A starting list of 5 applications would
      serve our purpose. From what I could find on the net, palindrome
      and einverted seem to be slow. Are these two good candidates to
      get us started?

Anything else that you think may enlighten us would be heartily welcome.

Thanks in advance

--praneet
_______________________________________________
EMBOSS mailing list
EMBOSS at lists.open-bio.org
http://lists.open-bio.org/mailman/listinfo/emboss