[OpenAFS] cache performance

[Phil.Moore@morganstanley.com]

>>>>> "Todd" == Todd M Lewis <utoddl@email.unc.edu> writes:

Todd> Okay, it's starting to sink in now.  I tend to think of us as a
Todd> largish AFS shop because we have lots of users, but we aren't
Todd> physically distributed (all our servers are in one room I
Todd> believe) and we're more concerned with monitoring individual
Todd> application use rather than volume use (though they are closely
Todd> related in most cases).

"Scale" is a very misunderstood concept.   Its a mutli-variable space,
and you have to be specific about which particular variables are large
in any given environment.

In fact, I've got a slide that discussed "scale", and what we mean by
it, in virtually every talk I've given in the last 10 years.   

Todd> I look forward to seeing what you come up with.  Sounds like
Todd> anything that makes managing AFS easier on your scale is
Todd> probably going to scale down in a way that's useful to the rest
Todd> of us.  Good luck.

Well, we do try to make our utilities scale down as well as up,
becuase we have to support small sites, as well as large ones.

For example, I have to architect a campus design for our large MANs,
supporting 1000's of machines, but we also need that same architecture
to scale down to the small sites, where hosts number on the 10's, not
1000's.