I've set up Sun's Online Disk Suite on a couple of systems (not AFS servers)
and have had about six months' experience using/managing it. I've been working
with production AFS servers for about six years, and my advice is to do some
serious testing in your environment before you decide to combine the two. I
think the main issue is an operational one, in recovering from the kinds of
problems you run into on any fileserver.
ODS has some attractive features, but it adds another layer of software
complexity between the physical disk and what users really want, thus it adds
to the complexity of managing an AFS server. For example, the ODS has its own
database without which the mirrored disks are essentially useless, and the
synchronization of mirrored partitions is not quick. So, if you're interested
in minimizing downtime, you need to make sure your system administrators
understand everything they did before AND the internals of ODS so they can
recover from problems quickly. (I wouldn't like to be working for a large
financial institution which is losing million$/sec while I fumble around
trying to figure out why fsck (duh! which one was I supposed to use?) isn't
working...)
If you set up the ODS in a reasonable way (mirroring partitions that are
on different disks, on different disk controllers), and do some testing
recovering from interrupted vos moves, salvages run on completely full
partitions, and other mean things you can do to a file server, I'd like
to hear about the results. Good luck!
Anne Salemme
MIT Information Systems, DCNS Athena Systems Support
