Difference: ImagerTechnicalInformation (29 vs. 30)

Line: 1 to 1

META TOPICPARENT	name="WebHome"

Imager Technical Information

Line: 13 to 13

Added:

>
>

Imager has raid space (found in /ubc/cs/research/): imager-raid1, psm-raid1, spiff-raid, pai-raid.
The raid space lives on a Xyretec disk and has performance problems. Sometimes, the raid is slow. Tibi performed some tests and Mike tried to figure out what caused the raid to be slow. However, the cause is still not known. Perhaps profiling during testing might help analyze performance. If the raid is problematically slow, please notify the help desk.

Changed:

<
<

>
>

Tibi's tests show that imager-raid1 and imager-raid2 have similar performance
It is documented that sequential I/O over NFS when submitting several jobs in parallel to the SGE can bring the system to a halt. Therefore, optimizing the I/O activity when submitting jobs to the SGE on the clusters likely will improve the performance quite a bit. It is possible that some of the scripts submitted do a lot of sub-optimal I/O over the network thus slowing all of us down. Here is a document tht explains the basic SGE commands as well as some best practices to optimize the I/O over the network (page 31) including some sample code. http://www.biostat.jhsph.edu/bit/SGE_lecture.ppt.pdf