[Bioclusters] mpiBLAST

[Osborne, John]

>> Right now I have it running, but with it doesn't produce any output and
>> lamclean hasn't killed
>> the process on the first node (it has been trying for >5 minute now).
>Hmmm.. Did you lamboot your MPI environment first?

Yes.  Actually it works now.  What did the trick?

1)Rename (null).pal to nr.pal
2)Replace all (null)'s in nr.pal with nr
3)Add the substitution matrixes directly to my home directory,
mpiblast can not seem to find my data directory despite it being
specified in the .ncbirc file.  Maybe I have to add this to mpi.conf?
If it wasn't for the debug option I don't think I would have ever found
this, wise of the authors to include that one in there.
4)Make my master node n0, and have it run mpiblast as well.

I think I can probably solve 3 and 4 with some more time, but I have
spent more than a few hours on getting this thing to work already.  I'm
still not sure where the original mpiformatdb null problem arose.

>> Speaking of the first node,
>> are we supposed to set up n0 as our master node?  Is that why I get 21
>> fragments when I say divide
>> it into 20 pieces?
>No, I think the problem is mpiBLAST considers '0' to be a valid number. So
>if you have 20 nodes (0-19), you'll need to mpiformatdb for 19. I know,
>this caught me off guard too, but I just live with it.
>
Yeah, I'm living with it too.  I don't have a problem starting the count
with
zero, but then the count should end with 19 when I say divide into 20
pieces.

 -John