Re: [wgs-assembler-users] ovlstore our of memory

[Walenz, B. <bw...@jc...>]

Hi-

The quick answer is to use a much much lower memory limit, like 8192 (8gb).  There is no gain to using more memory here.  It seems that the algorithm that partitions the data — already known to be be not well balanced with mixes of short and long reads — gets more imbalanced with larger memory sizes.

If you still have the *BUILDING store, you can take the size of it, divide by 512 for an excellent memory limit.

The store build, unfortunately, must restart from scratch; remove Howea.ovlStore.BUILDING then restart runCA.

b

On 10/23/12 6:38 AM, "Quan, Xueping" <x....@im...> wrote:

Hi

I am running my job to assembly Illumina and 454 reads. After running runCA 7.0 my job was killed in the ovlstore stage due to out of memory.
"/apps/wgs/7.0/wgs-7.0/Linux-amd64/bin/overlapStore  -c /ax3-savolainenlab/analysis/Plants/Genomics/Celera/fullrun1/Howea.ovlStore.BUILDING  -g /ax3-savolaine
nlab/analysis/Plants/Genomics/Celera/fullrun1/Howea.gkpStore  -i 0 -M 655360 -L /ax3-savolainenlab/analysis/Plants/Genomics/Celera/fullrun1/Howea.ovlStore.li
st  > /ax3-savolainenlab/analysis/Plants/Genomics/Celera/fullrun1/Howea.ovlStore.err 2>&1
=>> PBS: job killed: mem 1014599844kb exceeded limit 1006632960kb
"

Is it possible to set a memory limit which ovlstore will not exceed? "ovlStoreMemory" is currently set as 655360 and seems doesn't work.
And am I able to resume the job from the ovlstore stage rather than run from the beginning? It really took very long time to finish the overlapper stage.

Thanks very much!



Dr. Xueping Quan
Imperial College London