On 06/02/2011 08:25 AM, Martin Sustrik wrote: > On 06/02/2011 08:17 AM, Pieter Hintjens wrote: > >> Perhaps this is a new pattern, yes. Possibly in the way we once had a >> stream pattern in AMQP. > > Presumably. None of the current patterns seems to fit the problem.
Hm. Not really true. With subscription forwarding in place, the PUB/SUB pattern could be used to achieve something like this. Video receiver would subscribe for particular video. The subscription would contain an unique identifier of the subscriber in addition to the name of the video to distinguish it from other users watching the same video at the same time. The publisher would plug into the subscription mechanism using XPUB socket. When subscription is received it would start streaming the video broken into small frames. Each frame would be tagged by the subscription string to ensure that it gets to the right user. Publisher-side filtering would guarantee that the stream is forwarded via particular patch without dissipating into different paths. Moreover, accomplishing the goal with PUB/SUB would allow for all kinds of funny topologies. Say streaming video into a single device at a physical location and then distributing it to all the local boxes via multicast. Etc. As for HWM, dropping messages seems reasonable in this model. If there's not enough memory to serve the video feed, the video would hickup occassionally. Martin _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
