[zeromq-dev] Does ZMQ "Over Send" Using OpenPGM

[Steven McCoy]

On 20 October 2010 03:13, Bob Beaty <bobbeaty at comcast.net> wrote:

> All,
>   OK, I've got more interesting news... in my example at:
> http://gist.github.com/634738 if I
> take out all lines in the range 68->94 *except* 80 and 81, I get the same
> numbers that my transmitter is showing. I added logging to my transmitter to
> say what URLs it was sending on.
>   If I put the rest of the connections back in - eventhough they have no
> data sent to them, I get this 10x receipt message count. It's as if the
> messages sent on those two URLs are getting repeated on all the others, and
> the ZMQ_SUB socket is picking them up and returning them to the recv()
> method.
>   This would account for the problem, but it shouldn't be happening this
> way. Right?
>   Am I missing something terribly obvious?
>
>
>
That's a lot of groups.  Most hardware and operating systems only support
20, plus there is a common problem with groups on the same port.  I would
follow TIBCO's recommendation and only use unique ports, i.e.
239.22.3.1:1001, 239.22.3.1:1002, etc.

The problem stems because the receiver listens to the port, and then adds
the multicast group.  Anything received on that port is processed, including
broadcast traffic.  Whilst on Linux you can bind to a multicast group you
cannot on other platforms and more importantly for PGM you cannot receive
unicast traffic and so reliability is not possible.

You should also verify what groups your nodes are already subscribed to, for
example NTP on 224.0.1.1 (NTP.MCAST.NET), 224.0.0.251 (mDNS) and more often
than not 224.0.0.1 (ALL-SYSTEMS.MCAST.NET) which leaves only 17 groups.  You
can use "netstat -g" for an active list.

-- 
Steve-o
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20101020/8cda9439/attachment.htm>