Cluster recommendations

[Ori]

Hej,

We are puzzling specs for a cluster dedicated to PacBio. We start with the guide lines given on the wiki (https://github.com/PacificBioscience...1#wiki-HardReq).

But I would like some input from you guys, from your experience. Especially on memory requirements for de novo large genomes.

Thanks

[SillyPoint]

As to hardware, the old answer applies: As much as you can afford.

But I do have one possibly useful recommendation, if you are building a brand new cluster: Run it under SGE. PacBio claims to support LSF and PBS as well. But:

1) If you ever intend to run the HGAP de novo assembly protocol, you will find it is built on top of the Celera Assembler, which spins off a second tier of jobs on the cluster. CA assumes SGE.

2) The PBS submit command, as noted on the wiki, lacks a 'sync' option. This necessitates an ugly kludge (qsw.py) which polls PBS every ten seconds waiting for jobs to finish.

Your choice of cluster manager will be driven by many considerations, of course, such as existing installations and expertise.

Good luck,

--SP

[Khen]

I suspect you will need a $%#@ ton more memory than 2 GB per 8 nodes. The cluster I use has 24GB per 12 cores and I wish I had twice as much memory.

[frozenlyse]

Our cluster has 512Gb ram per 64 core node, and RAM has never been a problem.