Kent M Pitman  <pitman@world.std.com> wrote:
+---------------
| Heh.  I've often said this about language standards: that any one of
| us could design a better standard than us as a group, exactly because
| it would then have a single, coherent point of view and style.  When
| we introduce democracy, or any consensus process other than simple
| dictatorship, we get weird combination effects that often are not what
| anyone wanted.  SOMETIMES that's better, but often it's not...
+---------------

Reading this, I am struck by certain parallels that come up in the
design of contention-based network access protocols. Several decades
ago, people discovered that "asymmetrical" protocols [such as the
Urn Protocol, Kleinrock and Yemini (1978), or the Adaptive Tree Walk
Protocol, Capetanakis (1979)], where (subsets of) stations are treated
differently depending on recent behavior, can be much more efficient
than "symmetrical" protocols [such as Aloha, CSMA, or Ethernet], where
everyone is treated equally regardless of recent behavior.

In fact, while symmetric contention protocols may *look* fair, they
sometimes may be very *unfair* under certain conditions. For example,
the Ethernet protocol has a bug (sometimes known as the "capture effect"),
in which a station that has successfully transmitted recently has an
unfair advantage over stations that have lost the contention recently,
so much so that it can effectively "capture" the entire net for fairly
long periods of time. (Mart Molle's "BLAM" protocol fixes this, by
introducing a degree of history-based asymmetry into Ethernet CSMA/CD,
but AFAIK no Ethernet hardware manufacturer has implemented it.)

[Hmmm... There's some applicability to Lisp programming in there
*somewhere*... I think...]


-Rob

-----
Rob Warnock, 31-2-510		rpw3@sgi.com
SGI Network Engineering		<URL:http://reality.sgi.com/rpw3/>
1600 Amphitheatre Pkwy.		Phone: 650-933-1673
Mountain View, CA  94043	PP-ASEL-IA