o It turned out to be more of an ns tutorial than an ns workshop. Most
people had little experience with ns, and many questions were kinda basic.
And the pace was managed poorly; we spent a lot of time on the early intro
stuff and little time on the later juicy stuff. Still, i learned quite a
bit about ns internals and learned some good ns debugging tips.
o A lot of the attendees were doing wireless/mobile/satellite
stuff. A few were doing QoS, a few routing, one Active Networks.
o I got the impression that they are happy to take useful
contributions to ns, particularly if you supply the test suite and
docs for whatever you add.
o I also got the feeling that there is a sizable, active, supportive user
community on the ns users mailing list. I'm going to subscribe
and see what it's like.
http://www-mash.cs.berkeley.edu/ns/ns-lists.html
o Fall didn't know of any work to use ns to model busy web
servers. And the support for modeling HTTP traffic seems minimal.
o Kannan recommended the GA Tech tools as the best ones available for
synthetic topology generation - particularly the transit stub stuff.
o They don't seem to be very far along on getting ns to scale to large
simulations. NS is memory-limited for large simulations. Ex: 800MB for
2420 nodes, 2465 links. Packets & events seem to be the big hogs.
Polly Huang @ ISI is working on "selective abstraction" techniques to
speed up ns multicast simulations
(http://www.isi.edu/~johnh/PAPERS/Huang98a.ps.gz). Essentially they're
ignoring packet-level, hop-by-hop, and queuing issues. And, as you
might imagine, it doesn't sound like they've figured out how to apply
this to simulations with lots of TCP connections.
neal