Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!mailrus!tut.cis.ohio-state.edu!ucbvax!RELAY.PROTEON.COM!hs
From: hs@RELAY.PROTEON.COM
Newsgroups: comp.protocols.tcp-ip
Subject: Randomnicity and the Token Ring
Message-ID: <8908282021.AA24495@monk.proteon.com>
Date: 28 Aug 89 18:45:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: Proteon, Inc., Westborough, MA  01581  [(508)898-2800]
Lines: 19


          Hi All: Mike O'D is I'm sure relating back to a wonderful
          LBL paper in which one node was indeed locked out every
          time. This happened even though the ring is egalitarian and
          allowed one packet per customer.

          If the timing of trnsmit and receive are just right, then,
          when one guy gets his packet into the buffer, the second
          node gets its packet refused. By the time the second figures
          out that it should retransmit, the first node gets the
          buffer....But, there was no randomness claused by vagueries
          of OS, head seek times and what have you. Given a small
          amount of randomness this could not persist. Also, our newer
          products (and 802.5 and FDDI) have multiple buffers so it is
          very unlikely in these newer designs.

          But, the point is made that deterministic access does not
          imply deterministic delivery. I can bring the horse to water
          but I can't make it drink. Howard