I am developping a multicasting application based on RTMFP, and I use NetGroup and NetStream:send to multicast my data between group member.s
It's mainly a 1:N streaming case, but sometime I have to switch the publisher. I have an aglorhytm to repair the stream which can be corrupted during this switch operation (the stream stays the same + packets are ordered and unique).
Now my "publisher switch selector" (next publisher) is random on the group ring, I try to optimize it to make the stream reparation more easy, but one question comes: what is the tree directions in a multicasting scenario? I was hoping that the broadcasting is done in increasing "peer group address" direction, is it right?
If it's the case, I can sort the peer of my group by "peer group address" (increasing sort), and use this list in my algorythm to select the next publisher.
Thanks for your wonderful work on P2P in browser,
the "tree direction" in an RTMFP Group is "outward from the publisher hopefully along the lowest latency links between neighbors". note also that a multicast stream will build several distribution trees in parallel for different slices of the stream. the group addresses of the peers are not a factor (other than in the basic selection of the group topology).
furthermore, each new multicast sender creates its own low-level multicast stream in the group. the distribution trees are separate for each low-level multicast stream. there is no advantage to selecting any one member over another to be the new publisher.
to see how multicast works, watch
and go to about 36:25. it's also worthwhile to watch that entire video.
Ok, I see, it's clear now :-)
Thanks for your response Michael