Active Networks Review

The author of this paper is envisioning a future network architecture which
lends itself to faster and deeper innovations in network design than the
current Internet. Such a network, termed an "Active Network", would be
composed of programmable routers that allow user-supplied code to be
executed in order to create new routing algorithms and provide new
network-based services.

An active network is defined by 3 characteristics: the capsule model of
programmability, accessibility of the model by all users, and the
applications that can be built on top of such a network. This paper
attempts to address the common criticisms of the security and performance
implications of active networks by implementing an active network toolkit
(ANTS) and reporting on the authors' experiences and unresolved issues. The
ANTS system adds extensibility at the packet level. It provides a core API
which allows the users to query the environment, manipulate a temporary
store of service-defined objects, and route capsules to other nodes.

The authors' design improved on each of the characteristics of active
networks as described in their original paper that proposed active
networks. The improvements are as follows:

1) CAPSULES: Capsules are packets carrying mobile code to be executed on the
router. The concept of capsules is refined by using caches and on-demand
loading to reduce the overhead of carrying code inside the packets.
Additionally, the capsule system is made to inter-operate active routers
with non-active routers.
2) ACCESSIBILITY: Accessibility means that any user should be allowed to
execute code on the routers. The new design is able to isolate different
users' code and state but it is not resistant to Denial of Service type
attacks from misbehaving code and it requires that users' code be
authenticated through a trusted authority.
3) APPLICATIONS: Active Networks should allow for the rapid deployment of
new breeds of network services. The author has learned that the active
network model is much better at implementing new routing models than
providing new network services.

The concept of Active Nodes is a significant departure from current thinking
about network architecture, and as such, it is likely to cause knee-jerk
negative reactions when readers are initially introduced to it. However,
this paper does not do enough to assuage skeptics. The authors claim that
the performance hit from the use of Java is avoidable, but they gloss over
the fact that Java-like environments are necessary to provide certain
security guarantees. Additionally, they don't investigate the combined
effect on performance of a large scale implementation. Finally, the authors
ignore the fact that performance is part of the reason why services such as
multicast are not currently deployed. An active node's performance and
traffic load would suffer even more.
Received on Thu Nov 23 2006 - 01:37:10 EST