On 23/03/13 07:36 AM, James Vincent wrote:
> Some updates:
>
>
>> It should be
>>
>> mpiexec -n 40 --mca btl ^sm Ray ...
>>
>> and not
>>
>> mpiexec -n 40 Ray -mca btl ^sm
>
> Corrected this
>
>
The upcoming v2.2.0 includes a fix so that recycling never loops infinitely.
>>
>>> -disable-recycling \
>>
>> This bug (infinite looping) was fixed in the development branch.
>
> Pulled dev version and removed this option
>
>
>>
>>> -o \
>>> coco_filtered.200.NCBI \
>>> -p \
>>>
>>> /mnt/Nexenta/archive/BC_Projects/BCProject-1-Launen-Metagenomics/analysis_recovered_reads/filtered-reads/coco/coco_filtered_left.fastq
>>> \
>>>
>>> /mnt/Nexenta/archive/BC_Projects/BCProject-1-Launen-Metagenomics/analysis_recovered_reads/filtered-reads/coco/coco_filtered_right.fastqa
>>> \
>> Is this a type or your file is really called coco_filtered_right.fastqa ?
>
> This was a just a typo ( job did not try to run like this)
>
>
> After fixing the above things I am consistently able to run to
> completion using tcp as the transport but not using sm.
This is the sad state of NUMA systems ;-)
> This is on a
> single 40 core server that has 1TB memory. The timing info in stdout
> seems to indicate using tcp is actually faster also.
Since you are using a single computer, maybe you can improve performance a
litte using the loopback network interface:
mpiexec -n 999 --mca btl self,tcp --mca btl_tcp_if_include lo \
Ray -p BigFile_R1.fastq.bz2 BigFile_R2.fastq.gz
I don't know which network interface is used by default, but lo will the
fastest here.
> So for now, we've
> just been using tcp and that seems to have cured what ails us.
>
------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_mar
_______________________________________________
Denovoassembler-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/denovoassembler-users
