[zeromq-dev] Theoretical whitepaper on 0MQ

[Fabien Niñoles]

2011/7/9 Martin Sustrik <sustrik at 250bpm.com>:
> I would really like to see the surveyor pattern implemented. It's the only
> the third pattern I know of that conforms to all three principles.
>
> Moreover, I guess it'll be very useful from user's point of view as it
> implements "message destribution with replies" scenario, which cannot be
> addressed today without using hacks, such as combining pub/sub and push/pull
> into a single topology.

Right now, I just changed team from the Online group to the Framework
group, so my work on ZMQ is currently limit to my free time and,
altough I had time to implement the collector pattern using the same
model as the one I wrote in Python, I'm not satisfy with its
scalability: it's too easy to either lost request or that a single
node block an entire sub-cluster.  I'm looking for some way to allow
requests to be queued at the edges without adding too much useless
latency to the protocol. My impression right now is either I don't
have a clear enough model for the pattern, or that I'm taking in the
wrong way (meaning I don't have the right model at all).

Fabien