Re: Use cases for IM's

[Dan Weinreb <dlw@exceloncorp.com>]

   Date: Thu, 13 Sep 2001 14:43:12 -0700
   From: "Burdett, David" <david.burdett@commerceone.com>

	   You can never guarantee that a server or its software will never
   fail. You would always design it to be reliable e.g. using recoverable
   databases for persistant storage, but yes I think you have to assume that
   the BM can be unreliable. 

I'm sorry, it's hard to both be precise enough to avoid all confusion,
and yet be concise enough to be clear.  When I said "unreliable" I
meant "the formal failure model includes a failure mode, separate from
the usual network-drops-message failure mode, in which the IM itself
drops messages", which implies that our mechanism for reliability has
to deal with such a failure (e.g. by doing end-to-end retries).

			     The question is, what is the probability and is
   the probability sufficiently low that the consequence of the risk arising
   can be ignored.

Yes.