[OpenAFS] calculating memory

Thomas Kula kula@tproa.net
Fri, 28 Jan 2011 23:46:35 -0500

On Fri, Jan 28, 2011 at 10:38:37PM -0500, Gary Gatling wrote:
> [...]

> I gather real servers aren't an option `cause management really
> likes moving most everything into VMware. We already moved all our
> license and web servers into VMware and we have some other weird
> servers working in it also. Even Windows infrastucture like domain
> controllers and stuff. If everyone says its a bad idea I can make an
> argument though. :)

The general philosophy I've developed over the past couple of 
years is treating VM machine exactly like I'd treat any other
physical piece of hardware: I figure out my resource requirements
and if it can meet them, I'd use it. This is where it's key,
though, to have good measurements of that --- here's the cpu
it uses, here's the memory, here's the disk and network throughput
I need. Get some metrics, find out how loaded one of your
fileservers is throughout the course of a day or a week or
whatnot. Then set up a fileserver in the VM and try to duplicate
that load there. If it works, great, you've convinced yourself
that it works. If it doesn't, then you can go back to management
and tell them it doesn't work and back that up with data. [1]

Another thing to consider, especially in a virtual environment,
is what happens with the failure of the physical host any number
of VMs are on. The classic case is, for example, having all of
your afs db servers on the same physical host. But you'll also
want to look at your environment and see if particular combinations
of hosted VMs work poorly if they are on the same physical host.
People seem to layout VMs so that a particular physical host can
handle things under normal circumstances. But what happens in
unusual circumstances, and will that cause even more problems.
Say, do you have some i/o intensive process that pounds afs at the
same time? If they are both hosted on the same physical box,
the combination may end up crippling both systems. If you've
got cases like this, then you can point them out and start asking
how the folks responsible for the VM layer of things are going
to work with you to situate things properly.

This is no different, really, than when everything was on
physical hardware and you had to ask "if we lost this rack what
happened" or "hey, we've got all this network intensive stuff
on one switch, and we're killing ourselves because of it". The
environment has changed, but the kinds of questions you have to
ask fall in the same general categories of identifying resource

I think there are places quite happily hosting fileservers in
virtualized environments. I'm certainly doing that at home and
have no issue with it. On the other hand, at work I sometimes
wonder if putting our fileservers in our virtualized environment
would get us anything than being able to pat ourselves on the
back and go "See! We *virtualized*. Go us."

[1]: Whether or not you can convince your local management with
     real data or if they're surrounded by some sort of reality
     nullification field is, of course, entirely up to your
     local circumstances....

Thomas L. Kula | kula@tproa.net | http://kula.tproa.net/