[Codel] update, question

[Kathleen Nichols <nichols@pollere.com>]

ACM Queue folks are very quick and have sent us an edited version of the
paper.
We are now the bottleneck as Van has another paper with a deadline (and I'm
one of the guest editors waiting for it, so it gets priority!) but we
hope to get it
back to them before Monday. So I assume it will go up on Queue soon.

I am going to put out some of the other results asap.

So I came home and told Van about our Tuesday meeting and some of the
questions
and comments. He suggested I try to run a "datacenter" type scenario.
Well, okay,
easier said than done. ns-2 becomes very unhappy with a 10G link though
1G seems
okay with shorter simulation time than I've been running. I let the
buffers be set
to a "nominal" BDP which means they are huge for a 5ms RTT. Initially,
they fill up
and, because the control interval is longish compared to 5ms and the
pipe size is
big, it take a while (order 3 sec) for CoDel to start controlling after
which it does
so just fine. This is preliminary and not real scientific, but I was
wondering if anyone
could suggest any particuar wrinkles to this that make things more
realistic. I'm thinking
of using a smaller buffer size size it just leads to 3 uninteresting
seconds of sim time.

Getting a better wireless model might be interesting...

	Kathie