[OpenAFS] Deploying OpenAFS on VMs

Garance A Drosihn drosih@rpi.edu
Thu, 16 Jun 2011 21:10:57 -0400


On 6/16/11 12:18 AM, Derrick Brashear wrote:

 > On Jun 15, 2011, at 8:48 PM, Russ Allbery<rra@stanford.edu>  wrote:
 >
 >>
 >> We have AFS server VMs on VMware whose vice partitions are
 >> vmdk files on an NFS network file system, and they're usable.
 >> They're slow (we're using them to store old archival volumes
 >> that don't get much traffic), but they're still usable.  Given
 >> that, if you did something slightly less awful in terms of
 >> performance, I suspect you could run a real file server that
 >> way.
 >
 > I run a real fileserver that way.

Hmm.  I wish I had thought to ask about this earlier this week!

One of the things we're hoping to do is to upgrade our AFS
servers at RPI.  We have a few thousand users, but *most* of
the time we wouldn't see significant load on our current AFS
servers.  However, we do sometimes have periods when we see
a lot of filesystem activity.

We have been pretty confident that we could put the new AFS db
servers up as virtual machines (each virtual DB server on a
different physical machine, of course).  We've been thinking
we could also put the AFS fileservers on different virtual
machines, but maybe we should keep those on physical machines?
We were expecting to put the /vicepX partitions on iSCSI
disks, and had assumed that combination would provide pretty
reasonable performance.

Maybe I need to get a better idea of how heavily our fileservers
are used in those peak periods!

-- 
Garance Alistair Drosehn                =     drosih@rpi.edu
Senior Systems Programmer               or   gad@FreeBSD.org
Rensselaer Polytechnic Institute;             Troy, NY;  USA