[OLPC Networking] Mesh multicast support

[Krishna Sankar (ksankar)]

> (also, we would want to make sure the network card is 
> filtering out multicast traffic the host doesn't want so the 
> CPU doesn't need to process data it doesn't need
> 
<KS>
	Yep, very relevant. More so, because the Marvell chip would be
doing the mesh routing even when the power is off ! 

	Again, we need to dig deeper and possibly need application level
interfaces to control this. We, as in the routing protocols realm, might
not be able to make that determination, unless there are some
declarative policies, which guide us as to what multicast packets need
to be ignored.

	BTW, from your earlier e-mail, 
</KS>
> there are existing protocols in place for sets of switches etc only to
forward
> multicast down the ports they are asked for on.
<KS>
	This really mightn't apply here because the switches/routers are
not in the multicast path (or even be aware of it !). Of course, when
the domains are connected thru a switch/router, there might be some
visibility into the multicast traffic - hard to tell.
</KS>

> -----Original Message-----
> From: networking-bounces at laptop.org 
> [mailto:networking-bounces at laptop.org] On Behalf Of Trent Lloyd
> Sent: Monday, January 22, 2007 7:29 AM
> To: networking at laptop.org
> Subject: Re: [OLPC Networking] Mesh multicast support
> 
> (also, we would want to make sure the network card is 
> filtering out multicast traffic the host doesn't want so the 
> CPU doesn't need to process data it doesn't need
> 
> This seems pretty standard on networking devices these days, 
> just thought I'd mention it because it "really matters" in 
> the OLPC world :)
> 
> Trent
> 
> On Tue, Jan 23, 2007 at 12:24:17AM +0900, Trent Lloyd wrote:
> > Hi Krishna,
> > 
> > On Mon, Jan 22, 2007 at 07:01:52AM -0800, Krishna Sankar 
> (ksankar) wrote:
> > > Good question. 
> > > 
> > > >From my limited knowledge, the mesh is more or less 
> transparent - i.e.
> > > the 802.11s routing protocols use mesh algorithms; but the routed 
> > > protocols, over the 802.11s, could be anything.
> > >
> > > But this raises a good point - may be we want to restrict 
> multicast 
> > > by partitioning the network to small domains as well as 
> keep a lid 
> > > on the number of multicast messages. There is a finite 
> possibility 
> > > that multicast could drain the resources if not contained 
> properly. 
> > > OTOH, because of the underlying mesh topology, multicast 
> might not 
> > > be as costly as it seems.
> > 
> > I actually not so much want to "restrict the domain of 
> multicast" but 
> > make sure that multicast is only sent over the mesh where it is 
> > needed, there are existing protocols in place for sets of 
> switches etc 
> > only to forward multicast down the ports they are asked for on.
> > 
> > > Trent, I assume you are exploring Avahi for OLPC. I did 
> look into it 
> > > and looks like it should work without any problems; except the 
> > > partitioning aspect. (An waiting for my modules to actually test 
> > > this) If the whole network (or some big superset in a 
> domain) is a 
> > > giant mesh, while it is good in terms of visibility of 
> services, end 
> > > points et al, it could cause lots of routing overhead 
> basically draining the power.
> > 
> > Avahi is part of it, but if every machine is running Avahi 
> anyway then 
> > it's going to effectively be "broadcasted" anyway
> > 
> > I know Lennart Poeterring (the other major author of avahi) 
> is looking 
> > into some modifications to the mDNS caching that exploits mesh 
> > specifics, i.e. if the network is too large to support mDNS 
> over the 
> > entire mesh, only speak to those within a few hops of you 
> on the mesh, 
> > etc, but this is a totally separate work...
> > 
> > I am actually thinking more in terms of wide-spread 
> multicast use, if 
> > an application uses multicast, we don't want the whole mesh 
> receiving 
> > its packets but only enough to get the message from end to 
> end to each 
> > of the machines actually subscribing to that multicast IP, I think 
> > this would be quite usefull for multi-user applications.
> > 
> > Cheers,
> > Trent
> > 
> > > 
> > > Cheers
> > > <k/>
> > > 
> > >  
> > > 
> > > > -----Original Message-----
> > > > From: networking-bounces at laptop.org 
> > > > [mailto:networking-bounces at laptop.org] On Behalf Of Trent Lloyd
> > > > Sent: Monday, January 22, 2007 6:21 AM
> > > > To: networking at laptop.org
> > > > Subject: [OLPC Networking] Mesh multicast support
> > > > 
> > > > Howdy all,
> > > > 
> > > > I'd like to know what support the mesh wireless is 
> going to have 
> > > > for multicast, i.e. will it broadcast it, or will it currently 
> > > > forward packets only to the right part of the mesh 
> using something 
> > > > similar to IGMP?
> > > > 
> > > > Cheers,
> > > > Trent
> > > > _______________________________________________
> > > > Networking mailing list
> > > > Networking at laptop.org
> > > > http://mailman.laptop.org/mailman/listinfo/networking
> > > > 
> _______________________________________________
> Networking mailing list
> Networking at laptop.org
> http://mailman.laptop.org/mailman/listinfo/networking
>