[olpc-software] [Fwd: Re: [avahi] scalability of avahi (or mdns-sd in genera

[Christopher Blizzard]

Krishna Sankar wrote:
> Good notes. Couple of random thoughts ...
> 
> A) From our perspective, my (uninformed, as I am still learning the pragmas of OLPC) view is that, for > 50 nodes, with lots of network traffic, we might have to go to a powered, infrastructure mode; (I am ready to be corrected) 

That was my hope as well.  We haven't talked about it much here, but I 
suspect that once we get a large enough school we'll have to depend on a 
server to be there to scale as a mediator.  What we _can_ do is find and 
associate services and activities through that server using zeroconf. 
At that point it's an issue of scaling administration instead of scaling 
the network infrastructure.

> B) Most probably other constraints will kick-in before we reach the ZeroConf limits
> C) A ZeroConf substrate is applicable not only at the connectivity layer but also at the collaborative layer

Correct.

> D) Unless we are leveraging the bus for other functionalities, avahi *might* be slightly heavyweight; we might have to implement ZeroConf on the bare metal

Not sure why this is.

> E) Also, we might need more integration and context-awareness between the mesh layer and the ZeroConf layer; may be even optimize some of the exchanges.

Yeah, we'll have to do some thinking about that.  I've had the thought 
"we'll only use mDNS for two hops on the mesh" but we haven't done a lot 
of thinking about that yet.

--Chris