On Oct 11, 2007, at 5:02 PM, Lennart Poettering wrote: > On Thu, 04.10.07 10:13, Brian Gerkey ([EMAIL PROTECTED]) wrote: > >> - Independent of the Zeroconf implementation that's being used, are >> there drawbacks in general to registering and browsing large numbers >> of services? > > Yes. They cost resources, mostly in network traffic. The less you have > the less traffic is generated. > > May I ask you to elaborate a little why you need so many services per > client?
I'm adding service discovery to the Player robot device server, part of the Player Project (http://playerstage.sf.net), which I maintain. Each robot will advertise a handful of services (usually sensors, like lasers and cameras), so with real robots the resource limits are not a big deal: a few services per machine. But when working in simulation, I might have hundreds of simulated robots on a single machine, in a single process. I'd like to use the same service discovery scheme with real and simulated robots, so I need to resolve this service discovery resource issue somehow. Any advice? I could save a bit by bundling each robot's sensors into a single service per robot, but that won't be enough. I could save a lot by bundling all the (simulated) robots controlled by a single process into a single service. But then I have to implement a second-tier discovery mechanism to drill down inside each monolithic service to find out exactly what it's offering. brian. _______________________________________________ avahi mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/avahi
