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

Review of Active network vision and reality: lessons from a capsule-based
system
by Ivan Hernández

The paper discusses results, experiences and vision of Active Networks
from a particular implementation the ANTS toolkit. The idea of the
active networks is to allow packets flowing through the network to
dynamically modify the operation of the network (possibly only to that
particular flow). In ANTS this is achieved by providing ad-hoc
software to the routers to process certain types of packets known as
capsules. The software if provided to nodes at the edges of the
network by end-user software. Within the network a distribution system
transfer the code along the path the capsule follows when the code is
not ready. Ants uses capsules that are additional headers inside the
IP packet. At arriving to a router, each capsule is demultiplexed and
processed by the associated software.

The ISPs does not easily provide internal data to research, it is hard
to believe that they are going to provide support to install and
execute software in their routers. Furthermore, ... well known
patterns of traffic. Who can program the network? i.e. who is going
to be able to distribute code to the routers? It is not clear, the
author says that it is an open problem and that the temporal solution
is certification of code by a trusted authority. The trusted authority
has to certify that the service makes use of overall network resources
in a "reasonable" fashion. Nevertheless there is no discussion about
what is "reasonable", neither what evidence we must deliver along with
the service code to get a certification. Furthermore, this
"reasonable" metric may vary according to the type of active node,
i.e. for edge nodes, this requirements could be only small memory
requirements, but for core nodes the requirements could be constant
memory and to be able to keep up 1Gbps constant traffic that is hard
to achieve on software. The routers that a packet uses may vary in
capabilities and resources. There is no discussion about how to deal
with this kind of problems, i.e., there is no guarantee that for a
certain packet flow there is going to be an available path of active
routers. Furthermore, an active router could be flooded with code, how
the router would manage a large overhead of code to execute? It would
have to stop executing some code, to keep routing at a reasonable
manner, thus is no guarantee that the required functions are going to
be executed on the packet in a router. The author claims that in ANTS
is possible to isolate the code and the state of different
services. In addition to this, I think that the routers must be able
to virtualize their resources in order to provide a lower bound
guarantee of the resources that can be provide to the code.

Finally, it looks to me that overlay networks may be another approach
to solve the same problem but with a higher level of abstraction. In
addition, there is no killer app to this type of networks, furthermore
there is no killer-example of the use of this networks.
Received on Wed Nov 22 2006 - 22:48:44 EST