Stardate 20010901.0629 (On Screen): Quentin sends in a link to an article about an interesting experiment which was performed where a group did a distributed computing task without the permission of remote computers. It's an interesting tour-de-force but I'm not sure it's anything to get worried about. For one thing, the amount of computing that the remote computer does is actually fairly slight: it just calculates the checksum on a packet, and responds if it's good and ignores it if it's bad. (This is, perhaps, one billionth of the amount of work done on each SETI@HOME packet.) The remote system can only do one kind of calculation (the checksum) and only responds with a single bit (pass or fail). Second, sometimes it won't respond even if it's good; the packet may not be delivered or the remote system may be overloaded and may drop it. A "pass" response is reliable, but a "fail" lack-of-response may not be. So the central system doing this would have to send each test packet multiple times. And there's non-trivial amounts of computing involved at the central system to construct all the test packets, and to send them, and to keep track of timeouts, and to manage resends. All of this means that the number of problems which could be solved this way is distinctly limited, and that the compute load on the central system to frame each query is greater than the amount of computing that each remote system actually contributes. So it's an interesting tour-de-force, but I suspect that in practice the problem could be solved much faster by simply doing it on the central system directly. I'm not too worried about this becoming wide-spread. It's an interesting concept but I don't think it's actually practical.
(discuss)
