"A Parallel and Multithreaded ERLANG Implementation" - status?

Scott Lystig Fritchie fritchie@REDACTED
Wed Jul 31 07:04:47 CEST 2002


>>>>> "jl" == Jani Launonen <jabba@REDACTED> writes:

jl> How much that would enhance performance? The pipe-communication is
jl> supposed to provide 10% improvement over TCP/IP (as
jl> http://erlang.stacken.kth.se/user.html#unixdom-0.1
jl> describes. Perhaps there are newer and faster version somewhere?)
jl> so could one expect whole lot of more throughput?

I'm the wild-eyed hacker responsible for that driver.... The 10%
figure was for communication on a uniprocessor SPARC box (Ultra 10,
perhaps?) running Solaris 2.6.  One end of the UNIX domain socket was
an Erlang application, and the other was a C application.  Sun RPC was
being used on over the socket.  That was a while ago, alas.

So, that 10% number is subject to a lot of variables.  The speed
differences between TCP sockets across the loopback and UNIX sockets
varies pretty widely from OS to OS.  At work a few months ago I cooked
up some RPC latency and bandwidth figures for an x86 box running Linux
2.2.14.  I don't have the figures in front of me, but IIRC the
differences were more than 10%, both in terms of bandwidth and
latency.

-Scott



More information about the erlang-questions mailing list