abyss-pe and abyss-map

Question

[malabady]

Hi,

In my abyss-pe command line, I use "np=10" since I am running abyss-mpi. At the mapping stag in which abyss-pe calls abyss-map, it work only on on thread!

Here is the leading part of the abyss-pe command:

$ /usr/local/abyss/1.9.0/bin/abyss-pe np=10 k=64 name=AbC -d \

Here is the abyss-map command that abyss-pe executes:

$ abyss-map -j1 -l64 P_R2_001_val_2.fq.gz P_R1_001_val_1.fq.gz Ab10_10-3.fa \
  | abyss-fixmate -l64 -h pe150_1-3.hist \
  | sort -snk3 -k4 \
  | DistanceEst -j1 -k64 -l64 -s200 -n10 -o pe150_1-3.dist pe150_1-3.hist

Is there any way to make abyss-pe run abyss-map multi-threaded, i.e j10 for instance? I thought that abyss-pe will always make j=np, but this is not true. I am using abyss v1.9.

Thanks

[GouthamAtla]

my wild guess is that abyss-pe is a wrapper to abyss-map, hence the abyss-pe executes 10 abyss-map commands in parallel. Can you check with top and see how many programs are being executed ? If you are using np=10, either you should see 10 abyss-map processes or 1 abyss-pe process with 1000% CPU usage. If you just see one process of abyss-map with 100% CPU usage, you need to install the mpirun (openMPI) program properly. If openMPI is not configured properly, some programs silently skips the parallel running and uses single core.

[malabady]

It is on abyss-map with 100% CPU usage!

But if openMPI isn't installed properly, shouldn't I have had the same problem in the earlier stages in the assembly? All steps before abyss-map were parallelized on the 10 nodes and worked just fine!

What do you think about running abyss-pe np=10 j=10 ......? I tried it in a dry run and abyss-map commands has j10. still didn't test this in a real run.

[GouthamAtla]

np=10 should ideally use 10 cores. But in past I saw few programs that silently run on single CPU, without any errors or warnings if mpi is not installed properly. I am not sure about the previous steps, but can you verify that your openmpi is installed properly ?

Answer 1

[benv]

Hi malabady,

Use j=10 in addition to np=10 in your abyss-pe command line.

j specifies the number of threads to use, for programs in the ABySS pipeline that are multithreaded (e.g. abyss-map). np specifies the number of MPI processes (a.k.a. "MPI ranks") to ABYSS-P, which is the first stage of the ABySS pipeline. ABYSS-P is the only program in the assembly pipeline that uses MPI and it is not multithreaded. The remainder of the pipeline runs on only a single machine.

Edit: fix typo and clarify that ABYSS-P is not multithreaded.

Edit 2: Correction: Use j=10 in addition to np=10. Mention that the later stages of the pipeline only run on a single cluster node.

[malabady]

I see. But if I use j=10 and not np=10, wouldn't this affect how ABYSS-P is run at early stages of the assembly? how about passing both flags, i.e. np=10 j=10 to abyss-pe?

[benv]

Yep, my mistake. In your case, you should use both np=10 and j=10. (I've corrected my answer.) The j=10 will have no effect on ABYSS-P since it is not a multithreaded program.