[Ndn-interest] NDN Load Balancing and Alternative Path
GTS
gts at ics.uci.edu
Mon Apr 24 22:47:13 PDT 2017
My $.02:
On 4/24/17 7:58 PM, Junxiao Shi wrote:
> Hi Boubakr
>
> Yes, load balancing and usage of alternate paths apply to Interest
> forwarding only, while Data simply follows the reverse of an Interest.
> There is no "consumer address" in the Data, so that there is no way to
> send the Data on another path.
> The consumer is ultimately responsible for retrieving the Data it
> wants. If there is a link failure, the downstream (consumer or
> otherwise) can retransmit the Interest, in order to reach a node that
> has the Data (either producer or cache). It is also downstream's
> responsibility to load balance its Interests among multiple paths, in
> reaction to congestion marks.
> You may have seen schemes that add a return address to Data packets.
> Such schemes would not work with Interest aggregation: if the Data is
> routed via an alternate path according to the return address, other
> consumers whose Interests have been aggregated could not receive the
> Data because the Data no longer passes through the node that performed
> the aggregation.
That's correct.
Some of these schemes do not use the PIT at all, e.g.,
https://arxiv.org/abs/1512.07755 (to appear at IEEE ICCCN 2017).
No surprise then that "no PIT entry" implies "no interest aggregation".
IMHO, interest aggregation is a very appealing feature, in principle.
However, I wonder if there is any solid (for some definition
of "solid") published evidence of interest aggregation occurring with
non-negligible frequency under realistic traffic
scenarios and thus saving bandwidth, reducing latency, etc, etc.
Cheers,
Gene Tsudik
More information about the Ndn-interest
mailing list