[Bioperl-l] Proposed bioperl module for local running of the NCBI standalone blastpackage

Aaron J Mackey ajm6q@virginia.edu
Mon, 16 Oct 2000 16:43:34 -0400 (EDT)

On Mon, 16 Oct 2000 hilmar.lapp@pharma.novartis.com wrote:

> First, based on the paper I don't really understand how you would deal with
> a filling up a single multi-processor machine, but I haven't looked at the
> sources yet

The paper doesn't delve much into documentation/usage issues.  The
documentation tries to do a better job of that.  Basically, disperse is a
client/server modality ... the clients can run anywhere you want them to,
including all on the same machine.  In my experience, programs such as
fasta, blast and hmmer which all have threaded capabilities run more
efficiently as "one threaded client per machine" rather than "X unthreaded
clients per machine".  In disperse's case this is even more true since at
higher client numbers the interprocess communication begins to slow things
down (see, for example, the disperse paper where performance with greater
than 64 processors started to be more efficient with 32 2-threads fasta's
vs 64 1-thread fasta's)

> Second, 'Disperse' now seems to be restrictively licensed
> (academics only), so it is out of the game

This is probably only temporary while our university lawyers decide
whether disperse is patentable, etc.  (in some ways, working for a
university is not much different than working for a company - if you do
something considered very valuable, then the university wants to sell it).
No doubt our universities will feel that disperse isn't worth anything,
and will allow us to do whatever we want with it ;)