[nanomsg] Re: Broadcast protocol
- From: Martin Sustrik <sustrik@xxxxxxxxxx>
- To: nanomsg@xxxxxxxxxxxxx
- Date: Thu, 12 Sep 2013 12:27:45 +0200
On 11/09/13 22:00, Paul Colomiets wrote:
Martin, what do you think about reliable pubsub anyway?
Generic reliable multicast is impossible. The only way for it to work is
to stop the publishing when there's an slow/stuck consumer (after
buffers fill up). Given that in any sufficiently large topology there's
at least one slow/stuck consumer at any give time, the topology would
not be delivering any data, ever.
Btw, this is the reason why PGM is called "pretty-good" multicast, not
"reliable" multicast.
In the real world there can be reliable multicast solutions found, but
they are very constrained. Generally speaking, two conditions must be
satisfied:
1. The topology must be very small (typically a publisher and couple of
subscribers).
2. There must be 24/7 admin team monitoring the solution and ready to
kick-in when the topology gets stuck.
Martin
Other related posts: