OpenMAMA ZeroMQ 0.3 Released - PGM and IPC support added

on
2 minute read

Multicast and inter-process transports added

One of the things that has always interested me about ZeroMQ is its approach to different transports. The theory goes that the transports should be fairly pluggable with a few small code changes and it turns out that this is in fact true.

So how hard was it?

The previous release of the ZeroMQ bridge already provided URI support and the design of the bridge always had alternative transport bridges in mind so it was actually fairly straightforward. The main change that you need to make to support different transport types is to figure out whether or not you should ‘bind’ or ‘connect’ which varies according to transport types. Apart from that, everything else seems to be pretty much the same.

What does the configuration look like?

I have provided new mama.properties transport details for all the new transport types in the configuration but here they are for reference:

# Ensure loopback is running multicast for these to work
mama.zmq.transport.pub_pgm.outgoing_url=pgm://127.0.0.1;239.192.1.1:5657
mama.zmq.transport.pub_pgm.incoming_url=pgm://127.0.0.1;239.192.1.1:5656

mama.zmq.transport.sub_pgm.outgoing_url=pgm://127.0.0.1;239.192.1.1:5656
mama.zmq.transport.sub_pgm.incoming_url=pgm://127.0.0.1;239.192.1.1:5657

# Ensure loopback is running multicast for these to work
mama.zmq.transport.pub_epgm.outgoing_url=epgm://127.0.0.1;239.192.1.1:5657
mama.zmq.transport.pub_epgm.incoming_url=epgm://127.0.0.1;239.192.1.1:5656

mama.zmq.transport.sub_epgm.outgoing_url=epgm://127.0.0.1;239.192.1.1:5656
mama.zmq.transport.sub_epgm.incoming_url=epgm://127.0.0.1;239.192.1.1:5657

# Note destination here must be writable
mama.zmq.transport.pub_ipc.outgoing_url=ipc:///tmp/feeds_0
mama.zmq.transport.pub_ipc.incoming_url=ipc:///tmp/req_0

mama.zmq.transport.sub_ipc.outgoing_url=ipc:///tmp/req_0
mama.zmq.transport.sub_ipc.incoming_url=ipc:///tmp/feeds_0

One of the interesting things about pgm is it seems to lack the major pieces of functionality that is useful for the market data world - namely topic and multicast group resolution which would have vastly improved the usability of pgm. Looks like that’s something which continues to distinguish the commercial messaging middlewares like SR Labs’ Data Fabric or Informatica’s UMS from PGM (though my knowledge of PGM is weak so please correct me if I’m wrong).

So what’s the general state of the ZeroMQ bridge now?

The ZeroMQ bridge is effectively complete. It now supports all of the ZeroMQ bridges that I had planned to support and it has a 100% pass rate with the OpenMAMA unit tests.

If it’s so complete, why not provide a 1.0 release?

I might consider a 1.0 release in time, but there are a few things that I would like to see completed before doing so:

  • Issue #5: Make pool sizes configurable
  • Issue #4: Make various zeromq socket options configurable
  • Issue #2: Should see if the zeromq’s proc is a worthy queue implementation
  • Issue #1: Should move memory pools being used for message passing to queues

Once these are sorted, I think we’ll be ready for a 1.0 because it would be the finally settled-on architecture for internal queuing and it would be completely tunable. As long as these things are up in the air, I’d prefer not to call it a “1.x” release.

OpenMAMA, ZeroMQ, Scons, Architecture
comments powered by Disqus