[Time] Name | Message |
[00:41] travisbrady
|
sustrik: is the example at the bottom of this page out of date? http://api.zeromq.org/zmq_getsockopt.html the simple program I wrote to emulate it is crashing http://pastie.org/1004700 on Mac OS X
|
[01:36] foysavas
|
zedas: mikko: mongrel may have a good name to it, but sequels tend to suck. besides what ever happened to naming it gaegogi?
|
[02:10] zedas
|
foysavas: too offensive
|
[05:55] sustrik
|
travisbrady: yes, the example is broken
|
[05:55] sustrik
|
the last parameter should be pointer to size not the size itself
|
[05:55] sustrik
|
thanks for spotting this
|
[05:58] travisbrady
|
sustrik: sure, glad to help. I'm working on a client lib and polling and getsockopt are all I've got left
|
[05:59] sustrik
|
great!
|
[06:00] travisbrady
|
I've got the simple server and the chat examples working nicely. need to do the perf stuff next
|
[06:01] sustrik
|
yes, that's always interesting
|
[06:01] sustrik
|
seeing how much your language runtime adds to latency
|
[06:02] travisbrady
|
Any idea which languages (outside of c/c++) are fastest currently?
|
[06:02] CIA-17
|
zeromq2: 03Martin Sustrik 07master * r9858447 10/ doc/zmq_getsockopt.txt : getsockopt documentation fixed - http://bit.ly/aeBuDC
|
[06:03] sustrik
|
C is the native API
|
[06:03] sustrik
|
so it's definitely the fastest
|
[06:03] sustrik
|
however, C++ is just a thin wrapper
|
[06:03] sustrik
|
that gets optimised away anyway
|
[06:03] sustrik
|
so it should be as fast as C
|
[06:05] travisbrady
|
But apart from those is java introduce the least latency? Haskell?
|
[06:05] travisbrady
|
pardon me, I mean to say "DOES java" not "is Java", must be time for bed
|
[06:09] sustrik
|
depends on message size
|
[06:09] sustrik
|
brian grager done a lot of work on using zero-copy in python binding
|
[06:10] sustrik
|
so the performance for large messages is equivalent to C
|
[06:12] sustrik
|
compiled languages on the other hand may have the least performance impact on small messages (List, .NET)
|
[06:12] sustrik
|
Lisp*
|
[06:12] sustrik
|
Java
|
[06:29] umesh
|
I have a client server code using req-rep sockets ...
|
[06:29] umesh
|
I start a client first ...
|
[06:30] umesh
|
if I don't start the server it wiats indefinitely ... is there any way to detect error like server is not up or connection reset ... if server goes down before client send etc ...
|
[06:30] umesh
|
waits*
|
[06:36] sustrik
|
no
|
[06:39] umesh
|
no way through zmq_poll as well ?
|
[06:44] sustrik
|
no, 0MQ is async library
|
[06:44] sustrik
|
the message gets sends once it's possible
|
[06:44] sustrik
|
at the time you call zmq_send it's just queued
|
[06:44] sustrik
|
if you want to use your 0mq server with sync client
|
[06:45] sustrik
|
i would suggest simply using raw TCP connection on the clients side
|
[06:45] sustrik
|
there was a discussion about writing a sync 0mq client couple of times
|
[06:45] sustrik
|
but nodody really did it
|
[06:49] umesh
|
ok ... thanks
|
[12:02] CIA-17
|
zeromq2: 03Martin Sustrik 07master * r341e8a2 10/ AUTHORS : test commit - http://bit.ly/d13f9p
|
[15:37] CIA-17
|
rbzmq: 03Brian Buchanan 07master * rdb6d541 10/ .gitignore : ignore *.tmproj - http://bit.ly/c0AocM
|
[22:18] rgl
|
zedas, I'm almost convinced to start using fossil. the last sentence on your post was brutal heheh
|
[22:54] zedas
|
rgl: hehe, i calls it like i see it. :-)
|