Exceeding memory limits for variant calling

[Joshua Thia]

Hi there -

Having issues with the variant calling step. Despite setting memory limits in dDocent parameters, my submitted jobs to our uni's HPC (PBS pro) keep exceeding more memory than I specify to dDocent. This occurs in automated submissions and interactive submission.

Provided an example below where I have requested a large amount of memory on a job (100GB) and 8 cores, and then set the dDocent limits much lower (32GB) to see what would happen. As can be seen, the dDocent run accesses ~122GB, which kills the job.

Any ideas on a work around or dDocent script lines that I could edit to bypass this?

Note, because launching as jobs on HPC (and they keep crashing), I do not have the log files output by dDocent.

Appreciate any help.

Best wishes ,

~ Josh

JOB ERROR OUTPUT

=>> PBS: job killed: mem 122460784kb exceeded limit 104857600kb

ResourcesRequested:mem=100gb,ncpus=8,place=free,vmem=100gb,walltime=47:59:00

ResourcesUsed:cpupercent=800,cput=15:58:22,mem=122460784kb,ncpus=8,vmem=137878464kb,walltime=05:01:32

BASH SCRIPT TO SET UP CONFIG FILE AND EXECUTE DDOCENT

(note, same thing occurs in interactive job when I manually input parameters through dDocent interface)

### Config file

echo "

Number of Processors

8

Maximum Memory

32

Trimming

no

Assembly?

no

Type_of_Assembly

PE

Clustering_Similarity%

0.99

Minimum within individual coverage level to include a read for assembly (K1)

2

Minimum number of individuals a read must be present in to include for assembly (K2)

2

Mapping_Reads?

no

Mapping_Match_Value

1

Mapping_MisMatch_Value

3

Mapping_GapOpen_Penalty

5

Calling_SNPs?

yes

Email

n

" > $REF/ddocent_configs/varcall.config

### Call variants with dDocent

dDocent $REF/ddocent_configs/varcall.config

[Jon Puritz]

Hi Josh,

As of 2.8.12, the memory limit setting for dDocent is deprecated.  This feature was added by someone else and has never functioned as everyone assumes.  The updates with version 2.8.12 should help with memory limits, but no matter what memory will scale with the number of samples.  Is 100gb the max?  I would suggest trying 256GB or 512GB.  

Jon

--
You received this message because you are subscribed to the Google Groups "dDocent User Help Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ddocent+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ddocent/093ea6b4-9651-4b00-869b-7730494369acn%40googlegroups.com.

[Joshua Thia]

Hi Jon -

Thanks for your reply; was unaware of this change to dDocent's implementation.

I guess it's a matter of wait time; I have never run a job request with such a large RAM requirement on our HPC, so will see how that goes. 

Do you have any recommendations about core to RAM ratio? I have 574 samples.

Best,

~ Josh

[Jon Puritz]

Hi Josh,

Again, this is going to depend on your samples, both number and coverage level.  From you error, it at least seems that you need ~15GB per processor.  If you can use a version greater than 2.8.12, you should seem some memory benefits.

Hope that helps,

Jon

To view this discussion on the web visit https://groups.google.com/d/msgid/ddocent/48ba2e81-f5b3-4c07-92c7-73e032307711n%40googlegroups.com.