[zeromq-dev] [PATCH] Moved tests off 5555 (conflict with Eclipse)
mato at kotelna.sk
Wed May 4 16:37:56 CEST 2011
sustrik at 250bpm.com said:
> On 05/04/2011 01:41 PM, Martin Lucina wrote:
> > Think of the case where you're distributing work, and you want different
> > work sets to use different topologies for QoS purposes. So you end up
> > wiring up a new topology for each work set, and you want a bunch of
> > transient ports to use for that topology (which are communicated to other
> > nodes by a well-known service they connect to).
> I see. I think I recall the problem. The root of it was that the
> services were actually stateful, right? Each topology was set to handle
> particular state.
You're right that the services in that case were stateful, but the
stateful-ness of them had nothing to do with using separate topologies. The
reason that was done was to completely separate out "traffic for data set
A" from "traffic for data set B". Since said trafic could have potentially
been gigabytes of data, this seemed like a good idea.
> Thus, IMO, we should think about how to handle state in services rather
> than focus on a particular little hack.
I appreciate your concern for the big picture, but I'm proposing to replace
the use of a hack which people are actively employing (calling bind X times
in a loop) with a robust and simple solution directly supported by the TCP
stack; why do you think my proposed solution is a hack?
More information about the zeromq-dev