[nanomsg] Re: Fanout pattern
- From: Martin Sustrik <sustrik@xxxxxxxxxx>
- To: nanomsg@xxxxxxxxxxxxx
- Date: Sat, 19 Jan 2013 12:20:06 +0100
On 18/01/13 21:24, Paul Colomiets wrote:
I think uniformity principle should be taken up to the point "It
should be *possible* (or better *easy*) to build a uniform topology"
rather than "It must be impossible to build a non-uniform topology".
Sounds like a reasonable approach. Let's allow for non-uniform
topologies and rather focus on the *easy* part. Can we possibly simplify
it to the level where it becomes *trivial*?
By the way, why have you kept "pair" pattern?
1. It's useful for unit tests.
2. Scaling to 2 boxes is the least scalable pattern, yet, it's a
scalability pattern nontheless. It seems there's a continuum in
scalability. pair is the least scalable pattern. fanin is somewhere in
the middle (too much data sources will overload the sink at some point).
pubsub is among the most scalable patterns.
3. I was opposed to pair pattern in 2010 and 2011, because back then
people generally didn't get the idea that a single socket can represent
multiple connections and used pair for everything, basically as a fancy
replacement for TCP connection. I think the awareness of the community
have grown beyond that point today though.
Martin
Other related posts:
