[1213] in Commercialization & Privatization of the Internet
Re: Interesting new player
daemon@ATHENA.MIT.EDU (Stan Hanks (bcm))
Sun Aug 25 16:05:52 1991
Date: Sun, 25 Aug 91 15:06:38 CDT
From: "Stan Hanks (bcm)" <stan@karazm.math.uh.edu>
To: swb@MITCHELL.CIT.CORNELL.EDU
Cc: com-priv@psi.com, edtjda@magic322.chron.com
Scott Brim asks:
>Stan, since you're interested in using capabilities like multipath
>routing, how do you deal with the possibility of datagrams arriving out
>of order (which wouldn't be very transparent behavior)?
And right you are. That is a real problem for most of the non-routable
protocols (and some of the routable protocols!). In the initial implementation,
the answer was "well, you lose". Currently, we're experimenting with using
the IP type-of-service field to denote that a given flow is not
multi-path routable. For such flows, source routing is used. This is
one of those little details that still needs to be documented...
As an ultimate fallback position, MFS has a *huge* DS3 backbone from which
whole ethernet or token ring circuits could be TDM'd. That would also
guarantee a full 10 (or 4 or 16) Mbs for you, no matter what else was
going on in the FDDI ring. It's more expensive than the FDDI solution,
natch, but it's still cheaper than regular DS3s.
For most of the traffic that we've seen so far, the out-of-order delivery
is not an issue. There have been isolated cases of people wanting to ship
Synoptics' Laticenet management protocol or LAT, but we knew in advance
that they couldn't deal with out-of-order delivery and could work around
it. There haven't been any surprises with protocols that we thought
were OK on dealing with out-of-order delivery which proved not to be the
case.
Stan
