Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!UDEL.EDU!Mills
From: Mills@UDEL.EDU
Newsgroups: comp.protocols.tcp-ip
Subject: Re:  Use of Precedence in the IP Type of Service field
Message-ID: <8904212016.aa04716@huey.udel.edu>
Date: 22 Apr 89 00:16:30 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 29

David,

So far as I know, the only IP implementation to use the precedence
field in the forwarding function is the Fuzzzball, which led a brief
but flashy life as the switching engine in the NSFNET Phase-I backbone
network prior to July 1988. These gizmos used the precedence field
as a priority indicator in a conventional FB(n) queue service discipline.
However, and this has not been widely known, if the precedence field
was zero, which is what almost all implmentations (except the Fuzzball)
used, the priority indicator was taken as the type-of-service bits read
as a three-bit number. If even those bits were zero and either the
source or destination TCP port field was 23 (TELNET), then the priority
indicator was assumed as one. 

You will note that (a) if the precedence/TOS field was zero, TELNET
won; (b) if the delay or throughput bits were set, they won over (a);
(c) if a nonzero precedence was set, they won over (b); and (c)
Fuzzballs themselves used a precedence field of all ones, so they
always won. Oh yes, TELNET usually won, but FTP usually lost. All this
in the bad old days of horrendous congestion when desperate men were
driven to desparate measures. Surely these crimes of history will never
haunt us again, at least for the next month or two.

Further culpa of mea can be found in my papers in the SIGCOMM 87 and
SIGCOMM 88 proceedings. I surely would not admit that above nonsense
in a rag like that. And, oh yes, the Fuzzballs are still around. Have
you read your clock lately?

Dave