Speeding Up Soapdenovo2

[igwill]

Hi,
I am working on some de novo assemblies with Soapdenovo2 with the multi-mer option, and wanted to know if anyone has experience with getting these crunched quickly-ish.
Specifically, can Soap split itself over multiple nodes on a cluster (via SLURM)? All examples sbatch scripts I see have it use only one node and 8 cpus. I have access to more if it can use it.

Details on what I'm running currently:

My data: 33M (x2) reads of paired end Illumina, 150b read length on fragments in the 300-400bp range. Interleaved in a *.fa. Animal genome, estimating ~100-300Mb.

Relevant SBATCH info:

#SBATCH --nodes=1
#SBATCH --ntasks-per-node=8
#SBATCH --time=40:00:00
#SBATCH --mem-per-cpu=4000
SOAPdenovo-63mer all -s MYCONFIG -K 63 -m 57 -R -o TESTRUN 1>TEST_ass.log 2>TEST_ass.err

MYCONFIG:

max_rd_len=150
[LIB]
# most options just commented out, assuming defaults are fine for my short paired data
avg_ins=350
asm_flags=3
rank=1
p=TESTDATA.fa

For Soapdenovo2, what's the best combination of resources to crank up? Nodes? ntasks (and correspondingly the -p parameter for my run command)? mem-per-cpu?

Thank you!

[igwill]

Just for anyone wandering in later with a similar question, after getting some more info, we are trying this:


#SBATCH --nodes=4
#SBATCH --ntasks-per-node=8
#SBATCH –-mem=128000
#SBATCH --time=168:00:00


Should be plenty to let SOAP do it's thing as quickly as it can, hopefully.