[zeromq-dev] performance impact of large numbers of inproc sockets

[snacktime]

I'm going through the different strategies for message passing in a
multi threaded application, and one thing I can't find any information
on is at what point do you take a performance hit for the number of
inproc sockets you are using.  Currently I don't think I need more
then 2000 sockets, but at what point do I need to be concerned?
Message throughput would be fairly light.  Mostly we are offloading
the processing of network calls to a small pool of async reactor
threads, and passing back the results to the worker threads.

Currently we have a production app that creates a few hundred inproc
sockets and it works great.  It appears that aside from compile time
limits that we need to increase, it should scale up just fine.   But
before I jump in I thought I'd ask.

Chris