bwa mt branch extreme memory usage?

[Kotoro]

We attempted to align some illumina reads with bwa using the multi-thread branch. We submitted the job to a cluster and it landed on a node with ~100 GB RAM and 48 cores. Ram usage exceeded the system resources and crashed the node.

We trimmed the data with dynamic trim but had not yet filtered out shortened reads, could this be responsible for the extreme RAM usage or is there a known issue with the mt branch?

(attempting to run this with the single thread sampe/samse took SO long that I suspect the memory issue might be present there as well.) I'm having a hard time diagnosing the cause of the problem.

[nilshomer]

You may want to ping the bwa folks ([email protected]) or PM lh3 (the author).

[Kotoro]

I am also contacting them there, just wondering if anybody else has tried the mt branch with the same problems.

[Richard Finney]

1) Examine the fastq input. Make sure the read names are correct. Make sure there are 4 lines per read. Make sure somebody didn't accidentally redirect stderr into the stdout at some step and mess up the input to the next step.

2) Check your genome build files (what you "indexed") for bwa to align reads against. Are you using the same version of BWA for indexing and alignment? There is a very new BWA where the indexing must be redone.

3) Where does it fail? At the "aln" step or the the "samse/sampe" step of BWA?

[Kotoro]

the aln step completes just fine, Its the sampe step that appears to have a memory leak. The program was using more than 100GB of RAM, which simply shouldn't be happening.

Remember this isn't the vanilla bwa available on the sourceforge page, I went to the git repository and tested the multi-threaded code branch.

I am also recieving some assistance from the bio bwa help mail list.


M. Gooch