o We might want to circulate this to the web caching group
(websys@cs). I think (and they've mentioned in meetings) that it makes
a lot of sense to coordinate our deployment strategies.
o What's the motivation for 20 machines? If each ongoing project is
getting a machine dedicated to it for days or weeks at a time, then it's
reasonable, i guess, but it might be good to make this explicit. On the
other hand, having fewer machines to find space for and maintain might
lower the barrier to entry and get more people on board. Perhaps people
could sign up for as many machines as they have space for, with a minimum
of, say, 4. If the platform turns out to be PCs, particularly
Intel-donated PCs, they're likely to be in tower cases that take up quite
a bit more space than nicely-stackable Ultrasparcs [how much space do 20
of the loom machines take up?].
o Is it necessary for the sites to all have "dedicated, reliable
gigabit bandwidth"? Some experiments (particularly measurement) might
want to be going over "representative", shared, crappy links. And some
locations might not be able to scrounge up gigabit ethernet to a
gigapop. Relaxing this requirement would be another way to lower the
barrier to entry and get more people to sign on.
o I agree with Amin that mentioning the community's experience with and
ideas about software for managing clusters (Glunix, Stefan's ideas about
using Myrinet to download your favorite UNIX/NT in minutes, ...) would
help convince people that this is feasible.
o What should we do about Joe Touch's project with the same name and
almost the same goals and mechanisms?
("The X-Bone" J. Touch and S. Hotz, Third Global Internet
Mini-Conference in conjunction with Globecom '98 Sydney, Australia
Nov. 8-12, 1998 (to appear) http://www.isi.edu/touch/pubs/gi98/)
neal