Pretty much all the "layer 2+" switches (all the 10GbE unit that CASPERites are using, to the best of my knowledge) support proper multicasting through IGMP snooping. I figure we can do this setup part through the PPC without needing to complicate the yellowblock controller with IGMP stuff. The yellowblock will just need to decide whether or not to forward packets to the fabric or PPC based on IP and port, in the same way that it currently does for unicast.
WRT multicast TX, this was indeed implemented on the BEE2 (I recall playing with this some years ago) by Jouko in 2009. Emails predate archive, from 05 Feb 2009. I'm not sure what happened to this patch, but it's simple enough to re-add. Jason On 01 Oct 2013, at 19:28 , Jayanth Chennamangalam wrote: > Yes, you'll need either IGMP snooping or an IGMP proxy on the switch. Most L3 > switches support IGMP snooping. > > Jayanth > > > > On Tue, Oct 1, 2013 at 1:21 PM, David MacMahon <dav...@astro.berkeley.edu> > wrote: > Hi, Jason, > > On Oct 1, 2013, at 4:35 AM, Jason Manley wrote: > > >> Ideally, the 10 GbE block would be modified to recognize multicast > >> destination IP addressed and then add in the derived MAC address > >> automatically instead of getting it from the ARP table. Then it could > >> send both multicast and unicast packets. > > > > I think this is easy, and I thought it had actually already been done. > > I agree that it would not be very hard (not sure about "easy" :-)), but I > don't think it's already been done. > > > > >> It would be another matter to get the core to listen to multicast packets > >> (especially multiple multicast groups). I think that would be a little > >> more complicated (unless you set its IP address to a single multicast IP > >> and that's the only multicast group you want it to join). > > > > This is the tricky part. We've thrown around a number of ideas. A true > > multicast receiver is probably beyond the scope of our needs, or > > willingness to implement. For our purposes, I think we can architect things > > that a single receiver only needs to subscribe to consecutive/adjacent > > multicast addresses. In this case, we can implement something like a > > multicast subnet mask, where the least-significant-bits of the address are > > configurably ignored. Has anyone built a multicast receiver on an FPGA? > > I think building a multicast receiver is non-trivial. Doing multicast "the > right way" requires that hosts implement IGMP (Internet Group Management > Protocol) and that the switch(es) implement IGMP snooping. Using a switch > without IGMP snooping will result in multicast packets being delivered to > every port, which can overload ports leading to dropped packets. I think > using a switch that supports IGMP snooping with hosts that do NOT implement > IGMP will lead to multicast packets not being delivered to those hosts. > > Maybe the multicast group membership can be statically configured in the > switch? > > Dave > > >
