On Mon, Jun 16, 2008 at 7:52 PM, Andrew Lentvorski <[EMAIL PROTECTED]> wrote: > Bob La Quey wrote: >> >> On Sun, Jun 15, 2008 at 7:08 PM, Gus Wirth <[EMAIL PROTECTED]> wrote: >>> >>> Bob La Quey wrote: >>> [snip] >>>> >>>> For your amusement here is a FOSS twitter server written in Erlang. >>>> >>>> >>>> http://yarivsblog.com/articles/2008/05/28/announcing-twoorl-an-open-source-erlyweb-based-twitter-clone/ >>>> >>>> Maybe this is the next Apache :) >>> >>> Erlang! Cool, I'm studying Erlang from the "Programming Erlang: Software >>> for >>> a Concurrent World" book by Joe Armstrong that I got at one of the >>> raffles >>> (need to write a book report). Erlang is a great language for doing >>> protocols. > > Uh, yeah. Erlang is good for protocols like no other language I have ever > seen. In particular, the main programming paradigms map to state machines > phenomenally well. And, Erlang has a *great* way of reading and writing > binary blobs of bits and bytes. > >>> >>> Gus >> >> I suspect this application is a particularly good use of Erlang. So >> probably a good way to learn Erlang. If, that is, the guy who wrote it >> is any good. Always a proviso no matter what the language or application. > > Yariv seems pretty good from reading his writings. > > The only caveat I would give is that Twoorl doesn't really do what Twitter > does with the views and stuff. > > Twitter is a nasty application to have to architect. The way they set it > up--everybody has their own view of the Tweets. Consequently, the system > winds up with an O(N^2) storm of stores and reads.
Yep. That is precisely the problem. > Think about the problems that IRC would have if it was required to store all > conversations and replay the whole thing everytime someone logged on. > > Twiiter is effectively IRC with persistence. And that makes it *hard*. Yep. It is going to demand both hard thought to find a good architecture and to map it to hardware. But there is a lot of statistical overlap in those N^2 views. A good archtecture is going to have to be based on the real world statistics of use. It may be that specialized aggregations become popular and thus channel the Tweetflow into somewhat a more manageble form. Twitter is, as Dave Winer has pointed out, essentially an aggregator. In a primitive form the user had aggregates, but oviously all of the filtring and aggregation helpers that have evolved for RSS and email can be applied. Interesting problem. BobLQ -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
