[OpenAFS] Re: Best way to debug "Lost contact with file server"

Andrew Deason adeason@sinenomine.net
Wed, 29 Feb 2012 00:39:19 -0600

On Tue, 28 Feb 2012 18:56:30 -0600
Benjamin Kaduk <kaduk@MIT.EDU> wrote:

> Er, I would have thought that this would cause a group of clients
> deployed near each other to all prefer the same replica, which is not
> really "random".   That is, though from the point of view of a single
> client the choice may be random, from the point of view of managing
> server load it is not.  Is that incorrect?

Well, if you get a client/server combination in a situation like that,
it's not random for either clients or server :)

That can happen, but you have to be in the right nets to have it make
any difference (as in, class a net, etc). I think we also take into
account the subnet and whether we're contacting ourselves, but in those
cases I think it makes sense.

In my experience, it's been pretty random; I don't think I've seen it
noticeably do the wrong thing for someone, but it seems not-unlikely,
and of course most people don't even look at it. I mean, most of the
time if you're talking about a "large" number of clients you're talking
about clients in various networks, so they don't get all pointed at the
same thing, but it's possible.

If you want to force the servers to be weighted equally, you can do
that, too.

Andrew Deason