On Nov 1, 5:02pm, [EMAIL PROTECTED] wrote:
> > If your intra-campus network is that unreliable, AFS's fault tolerance
> > will not solve your basic problem: if your are losing your interbuilding
> > links frequently, you need to look at the quality of your network
> > infrastructure (or the people operating it.)
>
> I think that this statement is based on the incorrect analysis in the
> previous statement.
No, my statement is based upon taking a broader based view of
fault tolerance. AFS is part of a fault tolerance plan. It is
not the holy grail for unreliable networks. The way to deal with
unreliable networks is to fix the networks.
For example, I have my 3 dbservers as 1 in each datacenter in NY
and 1 in NJ. I still would not do this if network outages were
not rare. Does one really want to spend a lot of time with some
significant fraction of the network unable to make or see updates?
Actually, the network is critical enough to us that there are
multiple paths between the buildings. We *really* do not want
to suffer outages. Money and reputations are lost that way.
AFS is a great solution to the unreliable computer problem. Server
disk crash? No problem. But I stand by my contention that the best
solution for unreliable networks lies in the realm of networking.
Andy
