To help kick this off, how about concentrating initially on multi-core machines? This avoids large, difficult issues of security, permissioning, highly variable overhead for data transfer, and political issues around resource contention.
There are still plenty of difficult problems even in this restricted subset of parallelism. For instance, if we posit numerous instances of J passing "control" information via sockets and working on a common mapped file (to reduce the overhead of data transmission), we still have potential problems with co-ordinating access to the common data area. Dan Bron wrote: >Don Guinn wrote: >> After this message I would like to take this out of chat and go >> to private messages so as not to burden those not interested. > I think it would be best to keep the discussion public, because the more eyes the better. Also while I'm interested in discussing > the topic, right now I don't have the time to actually contribute substantial code, and so the discussion might peter out and go > nowhere if only a few of us are involved. Keeping the topic public would allow others to volunteer if they like. > To avoid flooding uninterested Forum members, we could just pick a new subject (say, "Implementing coarse parallelism in J") and > anyone who's not interested can filter that topic out. If interesting new topics arise out of that discussion, they can be moved to > new threads, so that they don't get filtered out with the "implementing" thread. > Does anyone object? -- Devon McCormick, CFA ^me^ at acm. org is my preferred e-mail ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
