Active network vision and reality

Paper Review: Active network vision and reality: lessons from a
capsule-based system

This paper examines the open research area known as active networks,
an architecture in which custom programs are executed in the network.
The paper reports on two years of experience in designing and
implementing active network applications using the ANTS toolkit. They
contrast their findings with the original vision for active networks
by looking at capsules, accessibility and applications.

In their system, applications obtain customized network services by
sending special packets known as capsules via active nodes, which are
programmable routers. A routine inside the capsule is executed at
each active node is passes. Using the API calls of the ANTS toolkit,
novel network services including forwarding routines can be
implemented for widespread use. Applications can use the new service
simply by obtaining the code through a directory service and then
utilizing the corresponding capsules.

In terms of implementation, end-user software provides code to the
edge nodes while a light-weight code distribution system disseminates
the code in the core of the network. The ANTS toolkit has been
designed to be deployed incrementally in the Internet, which is a
beneficial feature.

Capsules are intended to operate in software-based routers. Capsule
code is carried by reference, using caching, rather than value. The
question the paper then addresses is whether or not capsules are
feasible. To answer this question, packet forwarding performance is
examined. Of the various types of packet forwarding, only capsule
forwarding limits performance in nodes. It was shown that the cost of
capsule forwarding was minimal and would not impact network
performance significantly. Furthermore it was stated that ANTS could
be implemented in software and does not necessarily require hardware
implementations.

Even though services are implemented with mobile and untrusted code,
ANTS defeats all possible threats. Another major issue is resource
management since end systems have greater control over how their
packets are treated by the core of the network. This area of ANTS
remains in need of work. Their system requires signing from a trusted
authority and thus an open system would not guarantee that a node
couldn?t affect others when acting maliciously.

Using their system they have implemented services such as host
mobility, multicasting, MTU queries, web cache routing. Several
services are identified which are inherently difficult to deploy
currently, but with the help of ANTS could be implemented.

Although it was shown that ANTS could be implemented in software, for
real deployment I feel that it should be implemented in hardware to
meet the heavy demands of Internet traffic. I consider this paper a
difficult read. Background information was not sufficient to get a
clear understanding of what they were trying to accomplish. This
paper seems to be geared towards those who are engaged in active
network research. One strong aspect of this work is that it presents
a new way of thinking of the network core in order to support new
services. It seems to violate the end-to-end agreement, although it
may be the case that the future of the Internet is headed in this
direction inevitably. The open question that remains then is: Are
these new services, such as multicast, even necessary? If not, then
this work is simply interesting from a design perspective, but without
practical need. If not, then further work should be pursued in this
field to see if active networks can support next-generation
applications and services.

-- Nadeem Abji
Received on Thu Nov 23 2006 - 02:19:06 EST