I've talked to Bill Norton about shuffling the schedule a bit and
dropping the "ANS use of the IRR for aggregation" and adding the time
to the presentation of TCP, and splitting it back into two as we had
considered earlier. The "ANS use of the IRR for aggregation" will be
presented at a later NANOG.
Slides are available for your reading enjoyment for the presentations
on TCP and HTTP that will be given.
The slides for the now two separte talks are in the directory:
the filenames, sizes, and time allocated to the presentations are:
tcp-congested.ps 346k 20 slides (15 minutes)
tcp-http.ps 459k 27 slides (30 minutes)
The first slides in each introduce the topic. These are:
TCP Response Under Loss Conditions
It is useful to understand TCP's behavior and in particular,
TCP's behavior in the presence of loss. This may help dispell
some of the myth regarding emminent collapse of the Internet.
This also provides background needed to understand how an
application's behavior can affect their own performance, why
variations in TCP's algorithms can dramaticly affect performance
and how network equipment can affect TCP's performance.
What can be done about HTTP small TCP flows?
It's well known that HTTP comprises the majority of traffic
for most Internet service providers. It is also well known that
HTTP as practiced today creates a large number of small TCP flows
which do not take advantage of TCP's congestion avoidance due to
frequently spending most of their time in TCP slow start. This
presentation demonstrates the effects of tuning browser response
and benefits of HTTP 1.1 persistent connections. HTTP proxy and
other possible improvements are mentioned.