hassell@ncar.UCAR.EDU (Christopher Hassell) (12/14/88)
Short Scenarios You walk into your office with the same malarky to do. When you login you get the normal 'You have mail' and learn that your algorithm in the South Net just finished your last neural-net study and is going on to the poll you put in. Dang, that site being down bent the whole net out of shape, you think. It took 4 hours longer for that 10 Giganode system to converge, than you'd guessed. Someone next door just got back her bitmaps for a raytracer that was distributed in the North Quadrant, and got back the auto-poll for her system information made available by default. Her mail comes in from the public e-mail, it has all of her reference material needed. Now that the new phone adaptation is being used costs are nothing to worry about so bulk e-mail is nothing. Okay the above is nothing to Jump about. I couldn't think of enough possib- ilities. Actually games are a distinct possibility too, but enough of that. What I'm suggesting is a replacement for worms (malignant) to put their abilities to work. I've wondered for a while what a 'transient program' could do, wandering across the U.S., using few cycles on many machines. Why doesn't someone try this? It is already done with human support, and everything else in here is automated city. I've suggested this as a replacement for The Worm (to keep our idle hands busy) and allow experiments on a `Public' computer that size. I only got a response remarking how Max Headrooms would flood our screens. <eeck>. Seriously, standards can be set. Bulletproof and sealed environments for them to do their work are also possible. Comp.parallel appears to be the only relevant group left, especially for those who Haven't got a toy to play with. I personally would find the possibilities fascinating. Polls could be easily automated for the systems. MAAssive algorithms running low-priority and low-connectivity algorithms could be a blinding resource, not to mention normal massively parallel ones. As for the e-mail, it could be as smart as necessary, AIDing the phone company in finding the best time to tie up a line, making lower costs all around. Because of its versitility, priorities could be set maximising performance, and minimising cost to system (phone and others). Does Anyone see how useful it could be? I see it on the horizon, but not Now.
khb@Sun.COM (Keith Bierman - Sun Tactical Engineering) (12/15/88)
In article <3884@hubcap.UUCP> boulder!tramp!hassell@ncar.UCAR.EDU (Christopher Hassell) writes: > >What I'm suggesting is a replacement for worms (malignant) to put their > abilities to work. I've wondered for a while what a 'transient program' > could do, wandering across the U.S., using few cycles on many machines. > >Why doesn't someone try this? > >Does Anyone see how useful it could be? I see it on the horizon, but not Now. This is old hat to some. Read SIAM news to keep up to date on the progress of those using a vast array of VAXen and Suns to factor really large numbers (more than 100 digits). This is now "pretty easy" (so much for PKC :>). Worm technology isn't the answer. The "nice" way to do this is with RPC. The time isn't tomorrow...it's today. Keith H. Bierman It's Not My Fault ---- I Voted for Bill & Opus
eugene@eos.arc.nasa.gov (Eugene Miya) (12/15/88)
Distributed applications
Proposals like this are made all the time, I first heard them in 1973.
I actually saw my first distributed application in 1979 (Alto/Ethernet
based). So they predate Brunner's Sci Fi story years. Please consult
Schoch and Hupp's CACM paper [1982?] on worm programs.
The technical problems: poor programming language constructs for communications
and data protection. It has full amazed me that for all the Crays LLNL has,
with a fairly decent network, no one wrote multi-Cray applications.
The effort is too great for so little return (improvement).
To date the best example was the recent factoring of a 100 decimal digit
number (See Factoring by Email, done at DECWRL).
The best application I've heard was from a fellow at a cable TV company.
They have 10,000 68K in switching system around the US. On the day of the
Superbowl when switching is at a minimum, down load a redundant program
to make the system "the most powerful chess in the world."
The real problem isn't technical, the problem is social. Machines are regarded
as personal property, not a commons. What business is your program running
on my machine? It's an idea who's time will come.
Another gross generalization from
--eugene miya, NASA Ames Research Center, eugene@aurora.arc.nasa.gov
resident cynic at the Rock of Ages Home for Retired Hackers:
"Mailers?! HA!", "If my mail does not reach you, please accept my apology."
{uunet,hplabs,ncar,decwrl,allegra,tektronix}!ames!aurora!eugene
"Send mail, avoid follow-ups. If enough, I'll summarize."bart@speedy.cs.wisc.edu (Bart Miller) (12/15/88)
I've had a pair of students build a "useful" worm package. The idea is that you can take any program, and wrap it in a worm. The goal is to make forward progress on a computation in the face of failures. The worm keeps some predetermined number of "segments" (copies) running and restarts new ones as necessary. The restarted worms are from recent checkpoints. We have this running now 4.3BSD UNIX machines, currently HP Bobcats (68020 based). We'll have a paper and a releasable version soon. Note that this currently only works for single-process computations. --bart miller uw-madison cs dept bart@cs.wisc.edu ...!uwvax!bart