[Sigh. Please wrap your lines!]
I am in the process of designing and developing a next generation network
product line. These discussions on packet sizes and other related topics
have been of immense value to me. Thanks much and keep it up.
Of course, any assumptions you make on packet sizes, TCP connection
behavior and the like are great until there's some application shift
that comes out of nowhere that invalidates those assumptions.
If your next generation network product line now can't handle the
load, well then, you've got some real unhappy customers.
Recall that http traffic went from essentially unobservable to one
of the dominant types of traffic on the public internet in a span
of time measured in months. Clever hardware that had assumptions
about TCP connection behavior/durations and how that related in cache
sizes were now in deep trouble.
What's the next application type that's going to surprise us? Consider
the effect of NAPSTER and other peer-to-peer based applications and
how that challanges some basic assumptions on how asymmetric end-user
traffic volume is.
As a customer of much bleeding edge network hardware, when a vendor
asks me about the average packet size of the traffic on my network,
I know that I'm about to hear some bad news, excuses or qualifcation
of the forthcoming peformance numbers. And of course an ISP has
essentially no way to influence this distribution of packet sizes
since it's the users that choose the applications they're gonna run.
Louis A. Mamakos louie(_at_)uu(_dot_)net