------art_1593_16208852.1188532610139
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

On 8/30/07, M. Edward (Ed) Borasky <znmeb / cesmail.net> wrote:
>
>  >
> > The most interesting feature (being able to send messages to  "spawned
> > processes" that are resident in other O/S processes or on other
> machines)
> > isn't done yet. I'm not that thrilled with how Erlang does it, by the
> way
> > (IP addresses and ranges of well-known ports) but I haven't thought of
> > anything better.
>
> Well ... you need something that's OS-agnostic and network agnostic,
> which pretty much ties you to something "universal" like TCP/IP,
> sockets, ports, etc. I don't know much about Windows or MacOS, but I'm
> reasonably sure this stuff is about as efficient as it can be on Linux
> and Solaris, and it's *so* tunable. :) Are there any "parallel
> network-connected clustering" techniques that *don't* depend on TCP/IP?



Oh it'll be TCP,  all right. I was just thinking of the way that Erlang uses
to specify PIDs, with strings containing embedded machine identifiers. But
we know that the Erlang way works, because Ericsson supposedly has
distributed applications with literally millions of distributed processes.

------art_1593_16208852.1188532610139--