"Name Based Content Routing" Review

There exist a variety of methods for reducing access latency and
providing load balancing to Web servers. This is done by directing
clients to the "nearest" of a set of content servers which host
replicas of the requested Web content. The authors of this paper
argue that such mechanisms can result in unnecessary access latencies,
scaling problems (e.g. to support multiple CDNs), and that oftentimes
they defy the open nature of the Internet. Consequently, a name-based
content routing approach is proposed.

Some routers would be extended to act as both conventional IP routers
and name servers which could route packets based on the name of the
target resource, effectively adding a content layer to the Internet.
To this end, the authors describe an Internet Name Resolution Protocol
and a Name-Based Routing Protocol to support name lookup and routing
by name, respectively. Akin to BGP, NBRP would use a distance-vector
algorithm with path information to propagate routing information to
other routers. In addition to the original purpose, the advantages
of this approach are implicit support for anycast and the ability to
redirect from unreachable hosts. A variety of techniques are
described to make the proposed system feasible and scalable. Tests of
a prototype implementation are encouraging; in many cases the average
request latency can be reduced by as much as an order of magnitude.

I think the idea of adding a new content layer to the Internet is
interesting but I do not think the idea is timely nor that the paper
solves a real problem. For one, the paper starts off by claiming that
the majority of the traffic on the internet is HTTP, DNS and RealAudio
streams -- this assessment seems out of date. Presently, the
majority of the bytes on the Internet are from peer to peer networks,
especially BitTorrent, whereas RealPlayer protocols have lost most of
their users. HTTP traffic does account for a significant amount of
requests travelling on the Internet but I would argue that with the
proliferation of high speed connections, access latency is no longer a
problem. It may be possible to reduce the access latency for a page
from 50 ms to 5 ms but I do not believe most users would perceive the
difference. Additionally, I do not think a content based routing
architecture implemented in the core of the network would lend itself
to the new reality that peer-to-peer hosts have become important
providers of content -- the frequency of updates from the end nodes to
the content routing tables would be too high and too risky.
Received on Thu Nov 30 2006 - 03:01:31 EST