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

Motivation: Active Networks allow users to inject and execute customized
programs in the nodes of the networks. The idea of active networks has been
ambitious and desirable at times. This is an experience paper revealing the
state of the art through means of exploring a capsule based system including
the capsule model of programmability, the accessibility of that model to all
users, and the applications that can be constructed.

Key Points:

1) The ultimate goal of ANTS is to allow untrusted users to control the
handling of their own packets within the network, yet to ensure that the
code they provide can do no harm to the users of other services even if it
is designed poorly or used maliciously.

2) ANTS is based on a capsule approach in which code is associated with
packets and run at selected IP routers that are extensible. Each user can
program routers without affecting other users' capsules providing
customization and mutual isolation between untrusted users.

3) Capsules provide a way to package a static IP datagram with "active"
functionality including code that can be executed in a router. The code
needs to be certified by a trusted authority. However, a secure overlay
network like Pastry could be used for distribution of code. Capsules are a
clean way of upgrading processing along an entire network path. This has
been a problem for network management in which administrators have to
upgrade each node manually.

4) Although, the advantage of active networks seems to be introduction of
services without being bothered with backwards compatibility, I feel that
backwards compatibility would still be needed for realistic deployment.
Although, I do feel that such a mechanism would be useful for upgrading a
small controlled network as well as experimentation and measurement of new
services. Active networks provide an interesting test bed for new ideas and
implementations.

5) Active networks make the network very fragile and one cannot predict
behavior of the network which is highly desirable in sensitive applications.
A router might crash by running some code or behave in an unexpected way due
to maybe conflicting configuration. The wide diversity of the routers makes
it difficult to achieve dependability in active networks.

6) We have come to realize that softwares do contain bugs and writing
bug-free code is too ambitious a task. Thus, active networks provide a means
of jeopardizing the network with poorly written code. This also opens new
avenues for the malicious users to make the network behave in unpredictable
ways.

7) The routers need to be fast and all though the author argues that only
fast and compact pieces of code would be allowed, it introduces overhead at
the routers which might become a problem with frequent updates.

All in all, I feel that although active networks are useful for
experimentation as a test bed for new services and implementations, they
might not be well suited for large scale internet.
Received on Thu Nov 23 2006 - 10:53:04 EST