From: Ioannis Giotis (giotis@cs.washington.edu)
Date: Mon Nov 15 2004 - 01:45:55 PST
One of the main problems in improving internet protocols today is the large
effort required to deploy a new protocol. Furthermore, most of the improved
protocols can only show benefit, when both ends are upgraded to it,
something which does not generally encourage early upgrades. Finally, safety
issues require that the OS manufacturers are the ones that get to implement
these protocols.
The authors propose a scheme, named self spreading transport protocol, that
lies between the network and the application socket layer and is able to
compile future protocol upgrades. STP acts in a virtual machine fashion and
is able to offer backwards compatibility while at the same time upgrades
protocols when it is asked by another end. STP is built to offer a lot of
flexibility to accommodate future protocol extensions.
The authors present some known TCP extensions implemented in a STP setup and
show examples of its performance. They also present general arguments about
the abilities of STP in supporting all sorts of extensions. In general, the
idea seems very nice as it offers another layer of abstraction and clearly
meets its goals in terms of extensibility.
On the other hand, there are two issues of concern. First, performance is
certainly not improved, as the extra layer will require both additional
complexity and computation time. Perhaps, this might no be an issue nowadays
with modern computers, however a lot of non-PC machines have started
accessing the internet (cell phones, PDAs, refrigerators, etc.) making CPU
power an important issue. Secondly, safety concerns can be raised by the
added complexity. As we see daily more and more exploits of the current
protocols, one could easily argue that more complexity will only the number
of bugs and possible exploits.
Overall, the idea is good on paper, but mainly due to the reasons mentioned
above it could take a long time for something like this to become
widespread.
This archive was generated by hypermail 2.1.6 : Mon Nov 15 2004 - 01:45:55 PST