[zeromq-dev] ZMQ_PAIR sockets in 3.x ZMQ / IPC in winows implementation

john skaller skaller at users.sourceforge.net
Thu Jan 19 02:16:25 CET 2012

On 19/01/2012, at 9:58 AM, Martin Sustrik wrote:

> On 18/01/12 22:07, Pieter Hintjens wrote:
>>> ZMQ_PAIR sockets don't really fit in to the zeromq philosophy. Including them at all was probably a mistake.
>> I strongly disagree, finding them perfect for a quite common use case :-)
> Both of the above are true.
> 0mq is mainly about communication between multiple endpoints which makes 
> PAIR kind of out of scope.
> However, using pair where raw TCP connection would be appropriate makes 
> the task much easier for the user.
> The trade-off here is between ideological purity and ease-of-use.

No there isn't: actually this is a challenge to carefully rethink the abstractions

You are claiming the "most abstract" case involves multiple endpoints


Pieter is claiming 

(a) the 1-1 special case of that is particularly useful

(b) if I read the comments correctly, he's claiming more than that:
it isn't just a M=1, N=1 specialisation of the M-N case but also offers
additional synchronisation assurances.

and *I* claim you're both right. What's missing is the way to bind the additional
functionality onto the M-N case.

If ZMQ_PAIR were removed, what would you have to do to recover it?
Why can't that be put into a library function?
How can this be generalised so the functionality extends to M:N?

I am guessing the difference is something like this:

* Without ZMQ_PAIR all topolgies are dynamically constructed,
  which requires quite a lot of work and is error prone.

* With ZMQ_PAIR one particular topology is statically guaranteed
  by ZMQ in the sense that it does the dynamics of the connection
  of the fixed topology for you (its responsibility, not yours).

and from this description the missing abstraction pops out:
the ability to configure more general topologies statically and
have zmq make all the connections for you.

So roughlty speaking "ZMQ_PAIR" is a special case of
"network topology graph".

john skaller
skaller at users.sourceforge.net

More information about the zeromq-dev mailing list