> 2. To follow up on Vern Paxon's post, there is a famous paper that shows
> Ethernet traffic is inherently fractal--that is, it's just as bursty over 1
> million seconds as it is over 1 second. This paper, by Will Leland of
> Bellcore, answered once and for all the telco types who had always
> maintained that the burstiness of data traffic would even out over time.
Yes, the paper: "On the Self-Similar Nature of Ethernet Traffic" was quite
ground-breaking in its own right. Of course there were its detractors,
saying that the network analyzed was not "typical", that THEY had traces
that showed short-range dependencies, etc. There were many a battle from
several playgrounds on the long range-dependencies of data traffic, heated
discussions would erupt at conferences, several people were fired, I even
heard that somebody was shot one night during an all-out war on the subject
behind closed doors....
Seriously though, I think the paper is entirely relevent in data traffic
pushed on some types of CSMA/CD (shared) mediums. Ethernet by nature is an
unfair and random layer-1|2 protocol. However, it is not clear that the
low-frequency components necessary for long-range dependence are present in
token-based, ring or switched systems.
Further, with new technologies on the horizon with true multiplexing
capabilites (e.g., ATM), the assumptions that ALL data traffic is
inherently self-similar may not be valid.
Maybe the guys at Bellcore could move up to the present and check
out what real (e.g., full-duplex, switched and ATM) networks look like. 8-)
Don't get me wrong, I have total respect for Will, Walter, Murad and Daniel
for their paper and the contributions that they've made to the data world.
How do you say? "Man, y'all da BOMB!"
-Jason
Received on Dec 03 1997
Intro
