An Architecture for Content Routing Support in the Internet

Paper Review: An Architecture for Content Routing Support in the Internet

The paper proposes a new architecture for content delivery in the
Internet. Their system performs content routing based on a name-based
routing scheme. Furthermore, their design can be visualized as an
explicit content layer in the Internet.

The paper begins by describing the existing content routing schemes
and their deficiencies. Current DNS-based systems have high latencies
due to the need for several RTTs to globally located name servers.
Also, in the event of cache misses, the cost incurred is significantly
high. Similarly, these systems face scalability issues especially in
supporting multiple content provider networks. They end this first
section by stating that proprietary approaches to content routing
violate the Internet philosophy of using open, standard protocols.

In the next section, they describe their own scheme,
network-integrated content routing. They propose a content router
(CR), which acts both as a conventional IP router as well as a name
router. In their scheme, not all routers need to be CRs, but rather
firewalls, gateways and BGP-level routers would be modified. When a
request reaches a CR, it is forwarded to the next hop CR based on a
name-to-next-CR mapping. This continues until a CR is reached which
is next to the ?best? content server. To set the routing tables in
the CRs, the Name-Based Routing Protocol (NBRP) is used. NBRP is a
distance-vector based algorithm. Routing advertisements are generated
by the content servers, which may additionally append load
information. A benefit of this scheme is that requests are mapped
within one RTT and there is no need to go through caches or DNS servers.

The performance evaluation section is presented next and is based on a
C++ implementation of their router. The memory requirements of their
scheme, although large, are not unreasonable. Furthermore, their
system performed fast enough to handle a high request rate. Latency
improvements were shown to be 86 to 95 percent when compared to an
Akamai based example.

Deployment was shown to be plausible since neither IP nor the routers
need to be modified. However, CRs would have to be incorporated into
the network on top of conventional routers.

One issue that seems to pop up right away is the need to modify
routers in the Internet is significant. Their scheme would need to be
considered extremely useful for deployment to catch on. Another issue
is the true scalability of this model. The amount of content on the
Internet is vast and thus for routers to be able to maintain routing
tables for all the providers is seemingly unreasonable. Aggregation
of IP addresses keeps routing tables manageable at the IP layer.
Their scheme attempts to aggregate based on names. Although it is
difficult to imagine how this can be as effective as aggregating IP
addresses, their results seem to indicate success. One interesting
positive of their scheme is that CRs are better able to tell if the
routes they choose are actually working when compared to IP routers.
They are more responsive to change and even maintain knowledge about
load. Overall this paper provides a very interesting architecture for
content delivery. Since content delivery is one of the major
application categories on the Internet, this work has significant
potential. They tackle the problem of latencies and load balancing in
content delivery. Their scheme incorporates content delivery as a
layer in the Internet architecture which correlates to the importance
of content delivery in the Internet in general. Their scheme is
scalable and robust. They propose a standard open protocol rather
than a proprietary system. For these reasons, it seems that this
project has made a contribution to networking research.

-- Nadeem
Received on Wed Nov 29 2006 - 03:13:58 EST