Subject: More on Assn 2

(This message is being sent to all CPSC 826 students)

I have been asked to be more specific on what constitutes
a good comparison between UDP and TCP:

Run length:  Run length should be the smallest amount of
time such that doubling or halving the run length makes no
appreciable difference in drop rate or throughput.   A few
seconds will definitely not be enough.. A minute or so
probably will, but its easy enough to find out for sure.

It would be nice to run TCP and UDP both independently
and simultaneously and report on what you observe in
both scenarios. To test the capability to adapt to
changing and width you can try a scenario such as
the following:  (pj => process j tj => time j)

p1 |---------------------------------------------------------|
p2          |-------------------------------------|
p3                   |------------------|
   t1       t2       t3                 t4        t5         t6

Compare the total goodput that you get this way with what would
be optimal.

Packet size: Use UDP packets of 1450 bytes of user data.
(This is roughly what TCP will be using).

Test platforms:  Don't use loopback at all.  netlab or
other 100 mbps sparc systems to the bladexx systems should
be fine.  Make sure you are not competing with some other
teams traffic though!!

How to test tcp: Either build your own tcp traffic generator
or use class/it25/atcpgobl atcpgen --

I have both a udpserv and an atcpgobl listening on port 33333
on 10.5.0.6 10.6.0.6 and 10.7.0.6 which are reachable from
netlab-gw over links having bandwidth of 1 MBps, 10Mbps and
50 Mbps respectively.

Due date -- Since I have been late in getting this out to
you, I will back the due date up one week until Oct 26.