USF Home
Administrators
Hardware Status



Home
Getting Started
Support and Contacts
Software Portal
Documentation
Tutorial / Links
C Programming Linux MPI Programming
Get Account
Request Licensed Software
 

MPIBlast

  1. MPIBlast
    1. Description
    2. Version
    3. Authorized Useres
    4. Platforms
    5. Local Documentation
      1. Modules
      2. Creating and Submitting a Job
        1. Prepare a database for use
        2. Submit the Job

Last Modified: 04/30 12:38

Description

mpiBLAST is a freely available, open-source, parallel implementation of NCBI BLAST. By efficiently utilizing distributed computational resources through database fragmentation, query segmentation, intelligent scheduling, and parallel I/O, mpiBLAST improves NCBI BLAST performance by several orders of magnitude while scaling to hundreds of processors. mpiBLAST is also portable across many different platforms and operating systems. Lastly, a renewed focus and consolidation of the many codebases has positioned mpiBLAST to continue to be of high utility to the bioinformatics community.

Version

  • 1.5.0

Authorized Useres

  • circe account holders

Platforms

  • circe cluster

Local Documentation

Modules

MPIBlast requires the following module file to run:

  • apps/mpiblast/1.5.0

See Modules for more information.

Creating and Submitting a Job

Create a .ncbirc file in your home directory. It should be formatted as follows: 

  [mpiBLAST]
  Shared=/work/student/joeuser/blast
  Local=/tmp/joeuser/blast

'Shared' is a location that all of the nodes doing the computation will have access to and 'Local' should be a faster, local drive.

Prepare a database for use

Research Computing mirrors several of the available databases. They can be found in /opt/apps/ncbi-6.1/blast/fasta. If one of those will serve your needs, simply copy it from that location to your Shared location, and decompress it. 

[joeuser@host ~]$ module add apps/mpiblast/1.5.0
[joeuser@host ~]$ cp /opt/apps/ncbi/blast/fasta/drosoph.nt.gz /work/student/joeuser/blast
[joeuser@host ~]$ cd /work/student/joeuser/blast
[joeuser@host blast]$ gunzip drosoph.nt.gz

Then, you need to format the database for use. Below is an example way to do so. It assumes you want to run a job searching against the drosoph.nt database, on 4 nodes. 

[joeuser@host blast]$ mpiformatdb --nfrags=4 -i drosoph.nt -pF --quiet

Submit the Job

Below is a sample job script. It assumes that you want a hard runtime of 1 hour and that you have set up the database in 4 fragments. It also assumes that your value being searched for is in a file called blast.in, in the current working directory. You'll need to request two more processors than the number of fragments for the two coordinating processes that are created during execution. 

#!/bin/bash
#$ -l h_rt 01:00:00
#$ -N mpiblast-text
#$ -cwd
#$ -pe ompi* 6
#$ -j y
#$ -o output.$JOB_ID

module add apps/mpiblast/1.5.0
sge_mpirun mpiblast -d drosoph.nt -i blast.in -p blastn -o results.txt