[Swift-devel] Swift-issues (PBS+NFS Cluster)

Ben Clifford benc at hawaga.org.uk
Thu May 7 12:01:59 CDT 2009


On Thu, 7 May 2009, Yi Zhu wrote:

> > Can you aggregate VM's from a few different physical clusters into one
> > Nimbus workspace?
> 
> I don't think so. Tim may make commit on it.

If you try this, an interesting comparison would be to compare with a 
separate virtual cluster running on each physical cluster using Swift's 
multisite features to spread work between those mutliple virtual clusters.

This might affect things in at least the following ways:

 In the everything is a single cluster case, the shared posix file system 
(NFS or GPFS) will perhaps encounter delays due to high latency between 
the pieces of the cluster. This may slow things down in the same way as I 
think happens for MPI-over-WAN.

 In the multi-site case, each posix shared filesystem will be closer 
(though how much closer?) to its workers. But swift will have to transfer 
data to each one, and between each one.


-- 




More information about the Swift-devel mailing list