The next step: supercomputer via disparate nodes
Although this is not strictly on topic, it seems strongly related to a kind
of kernel development. I checked out distributed.net, and they seem to have
a much different approach to the question. They are more concerned with
parcelling out portions of a task, not with making an actual usable
computer.
Would anyone care to make predictions about when we might see a computer of
this kind (and what obstacles stand in the way):
1) each node consists of a niced user-space program running on top of linux
(i.e. participation in the whole does not require the dedication of all
resources).
2) no node is more or less vital to the system than any other (!)
3) nodes can connect and disconnect at will via ppp or other mechanism
(contributing cpu cycles for the time they are connected) without seriously
disrupting the operation of the whole (if the node does not arrange to
reconnect within a certain time (negotiated based on the assigned job), the
job is reassigned). The way I envision it, a daemon is configured to watch
(at intervals) for an internet connection. When it finds one it negotiates
and connects to the whole.
4) nodes can locally select to include or exclude jobs (via config file)
initiated by particular users. PGP authentication ensures identity of user.
The effect of this is that most users won't actually be able to use the
power of the computer, but various worthy causes will, because participants
will include them in their config file.
5) any node can submit a job to the whole (so once you're connected, that's
your login session).
6) a mailing list or newsgroup would serve for debates about the merits of
particular tasks, in which various worthy causes could state their cases.
7) various "administrators" could maintain their own lists of worthy causes,
which participants could choose to mirror, to avoid the noise of the
discussion groups while still donating their cpu power to the supercomputer.
Zack
Reply to: