I'm in the process of building a large Starcluster installation. It is
going to be supported by a linux support group that is used to its own
internal support processes, and we're trying to migrate the support over to
the cloud in a way that is most convenient for the support group.
One question that has arisen: would it be better to use EBS-backed
instances, to handle various automatic update patch / kernel update
cycles? It would be more like their usual setup, in that they could point
their current updating scripts to run on an existing cluster, instead of
having to boot a new base instance to make the updates on.
However, does using EBS also add some complications? First, how difficult
would it be to build a starcluster base AMI as an EBS-backed image? Are
there complications with that?
Furthermore, if I modify a running EBS-backed instance that has been updated
a bit, then start another cluster from the same AMI without stopping the
first one, presumably the changes would have to be made again? And then, if
the changes made were slightly different on the second cluster, would I kind
of have two diverged systems? In other words, all the problems mentioned
Does anyone have thoughts about this?
Received on Mon Oct 25 2010 - 09:57:08 EDT