sustrik at 250bpm.com
Tue May 3 17:40:05 CEST 2011
On 05/03/2011 05:24 PM, Mike Santy wrote:
>> The question is whether such a solution is in any way better than simply
>> using different TCP ports for different 0MQ sockets.
> Maybe I'm missing something here, but the subport idea just sounds to me
> like something that should be implemented in a device that maps ports to IPC
> endpoints. As I presume this is not going to be a common use-case, this
> choice would keep the libzmq core simpler and move the complexity out into
> an optional device.
Sure, you can do that even now, however, it works only in "one
functionality per port" way.
For example, if you have a REQ/REP topology and a PUB/SUB topology you
can't tunnel both through a single device.
The subports are meant to solve that kind of problem. For example the
REQ/REP topology would use port 5555, subport 1 while the PUB/SUB
topology would use port 5555, subport 2.
More information about the zeromq-dev