[Bioclusters] SGE and NFS scalability

[Rayson Ho]

Even SGE 5.3 can handle 200 nodes easily, and SGE 6.0 with Berkeley DB
spooling won't make a big difference unless you constantly have a huge
number of jobs submitted and dispatched.

However, Berkeley DB needs the DB file either on NFS4 or on the local
filesystem... and if you want to use failover (shadow master), then you
need to use the "classic spooling" for non-NFS4 servers.

Rayson




--- Steve <slitster at rcn.com> wrote:
> We are about to start the build out of a 200 node cluster, and we
> plan 
> to use SGE 6 for job scheduling.
> The NFS device is an EMC Centera with 3 data movers, so it should
> shift 
> data at a decent speed, but I don't know about SGE scalability.
> 
> Question:
> Is anyone using SGE via NFS on this number of nodes?
> Has anyone reached a CPU/scheduling limitation on a lesser number of
> nodes.
> Would SGE with Berkeley DB be of benefit here?
> Who would recommend a local install of SGE on each node?
> 
> Cheers,
> 
> Steve
> 
> 
> 
> 
> 
> 
> _______________________________________________
> Bioclusters maillist  -  Bioclusters at bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters
> 



	
		
__________________________________ 
Do you Yahoo!? 
Yahoo! Mail - You care about security. So do we. 
http://promotions.yahoo.com/new_mail