[Nfd-dev] NFD Call: Multicast Suppression Slide Deck
David R. Oran
daveoran at orandom.net
Thu May 9 08:12:37 PDT 2019
I just had a chance to look at this. Jittering forwarding/replies does
help, as numerous studies in the past for reliable multicast protocols
have shown. So I think what you’ve got, modulo perhaps more clever
adaptation to the timers based on load on the subnet is a good policy.
What seems entirely missing here is any way to adaptively switch between
unicast and multicast. There’s no point in continuing to multicast
once you have enough state to “lock onto” a next hop that satisfies
your interests.
Any node/forwarder receiving a multicast Interest knows this by the
destination L2 address being multicast rather than unicast. That means
you should never multicast Data in reply if the Interest was sent
unicast. Conversely, if you sent an Interest multicast and receive any
reply (either multicast or unicast) you can remember the L2 address of
the first response (after validation, of course) and subsequently
unicast your interest there until either you get a NAK for no path or
congestion (or timeout) or you start seeing RTTs increasing, when you
can revert to multicast.
In order to reduce the probability that two consumers are
de-synchronized by enough that they would lock onto different next hops
for the same interests, if you receive a multicast interest and you
still have PIT state (or keep some extra intro for short period after
satisfying an interest) you should respond immediately with a unicast
Data reply. If you don’t have such state, then invoke your multicast
reply jitter timer. This ensures that for a given interest within a
reasonable time window multiple consumers lock onto the same next hop.
Intuitively this should cut down the amount of multicast by a dramatic
amount.
DaveO.
On 29 Apr 2019, at 0:05, Ernest Andrew McCracken (emccrckn) wrote:
> As per discussed on the last NFD call these are the slides for a
> proposed suppression mechanism for multicast suppression for both
> interest and data traffic on a wireless medium. The assumption here
> is the nodes participating are in infrastructure mode. I will also
> try to find the bluejeans archive call for this.
> _______________________________________________
> Nfd-dev mailing list
> Nfd-dev at lists.cs.ucla.edu
> http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev
DaveO
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.lists.cs.ucla.edu/pipermail/nfd-dev/attachments/20190509/0fcb8b16/attachment.html>
More information about the Nfd-dev
mailing list