1. What is the difference between MPI-HMMER and HMMER?

2. How does MPI-HMMER differ from the PVM implementation?

3. What do I need to run MPI-HMMER?

4. Can MPI-HMMER take advantage of parallel file systems?

5. Can MPI-HMMER utilize co-processors such as FPGAs, etc?

6. What operating systems does MPI-HMMER run on?

7. Is MPI-HMMER's output different from HMMER's output?

8. What portions of the HMMER suite have been parallelized with MPI?

9. Can MPI-HMMER be run within a batch scheduler environment such as PBS?

10. What is the minimum number of processes needed for MPI-HMMER?

1. What is the difference between MPI-HMMER and HMMER?
MPI-HMMER is an MPI implementation of portions of the
HMMER sequence analysis suite.

2. How does MPI-HMMER differ from the PVM implementation?
MPI-HMMER is not a direct port of the PVM implementation. Instead, MPI-HMMER uses non-blocking IO to effectively minimize the overhead transferring the database fragments from the master node to the worker nodes.

3. What do I need to run MPI-HMMER?
You will need the latest MPI-HMMER code available from our downloads page as well as an MPI implementation. MPI-HMMER should compile on any MPI 1.2 (and up) implementation. We have tested MPI-HMMER on MPICH 1.2.7p1, and LAM 7.1.4. If you have trouble compiling MPI-HMMER with a particular MPI implementation, please let us know.

4. Can MPI-HMMER take advantage of parallel file systems?
Not yet. This feature is currently in development and we
hope to have a working solution soon. When it is ready,
we will make it available online.

5. Can MPI-HMMER utilize co-processors such as FPGAs, etc
Current unreleased versions of MPI-HMMER are able to support FPGA accelerators. We are currently porting MPI-HMMER to the Cell BE platform and hope to include combined support for Cell, FPGA, and general purpose processors in a future release.

6. What operating systems does MPI-HMMER run on?
MPI-HMMER should run on an POSIX compliant operating
system with an MPI implementation. It has been tested on
various Linux distributions for both the PC and PowerPC
64 architectures.

7. Is MPI-HMMER's output different from HMMER's output?
The output should be largely identical with the only
exception being searches that result in multiple
identical e-values. Because the searches are being
performed in parallel, it's possible that the results
are returned in an unpredictable order. In this case,
these results could be reordered.

8. What portions of the HMMER suite have been parallelized with MPI?
Currently hmmsearch and hmmpfam are parallelized with
MPI. All other aspects of the HMMER suite are still
available, they're just not parallelized.

9. Can MPI-HMMER be run within a batch scheduler environment such as PBS?
This is a function of the MPI implementation and your cluster environment. MPI-HMMER does not depend on a particular scheduler, nor does it interact with the scheduler.

10. What is the minimum number of processes needed for MPI-HMMER?
MPI-HMMER follows a master-worker design. Therefore, in order to use MPI-HMMER you need to instantiate at least 2 processes (1 master, 1 worker).