Digital Silk Road as Network Architecture

I think that networking is complex enough that only diversity can protect us from the many kinds of attack on networks and avoid the vulnerabilities of monoclonal species. The only hope for economical diversity that I see is from the community working on Active Networks. Such networks can support many different network disciplines without the cost of separate physical networks. Timeshared fiber, as it were. Current commercial fibers are shared by many users, but they all use the same bottom few layers of protocol. The routing and prioritizing of these packets is monoclonal. The active network efforts have so far has been in support not of security, but in search for protocols to support the many things you ought to be able to do with glass fibers linking piles of transistors. There are many things that are unsupported by Internet or any architecture that I know of. It is like having an OS to protect the code of one discipline from impairing another.

Quick anecdote

I recently heard a proposal to install P-wave detectors at oil refineries so that critical valves could be shut off before the more destructive and slower moving S-waves of an earthquake arrived and disabled the controls to shut the valves. Now packets travel along fibers much faster than even P-waves and here is an application where latency is the whole issue. The required bandwidth is a few bits per year, or a few nb/s. The value of those bits may be several million dollars per bit. The cost of extra latency may be thousands of dollars per msec. The benefit of the earlier signal (faster than the P-waves) may warrant significant development.

DSR is not a network architecture but it provides a framework for diversity in disciplines such as routing. Internet is sort of distributed but with few central physical and logical points of possible failure. In particular DSR provides fully distributed routing, physically and logically. If you need a route thru the net and there are nodes and connections collectively able and willing to move your traffic, and there are competent and trusted scouts and guides, then you should be able to move your data. There is no single point of failure imposed by the architecture. Bear in mind that a “point of failure” may not be a geographical place, but a universally deployed piece of code. I speculate that current available architectures are prone either to physical or logical brittleness.

I suspect that the real internet today is more reliable than its “architecture” would suggest. I suspect that there are trust barriers that protect sub components against various accidental and malicious signals. It is difficult to analyze this as these details are often secret.


The active network thrusts that I am aware of emphasize experimentation in protocols. I think that this is very good. I also think that they may have a permanent place in the network world. There are more combinations of comm function than can be provided by any fixed set of switch functions. Many of these require microsecond application specific response to arriving packets.
Another notion is the physical link that is shared between unrelated protocols. I was amazed when I learned that Apple would intermix packets of two protocols on one physical link. The packet protocol was easily determined by examining the packet. This is commonly possible but seldom needed in network design. I was (and still am) in the mindset of one low level protocol per link. That I was amazed was perhaps surprising.

Perhaps DSR packets could slip in between IP packets in a variety of physical links designed and deployed for Internet operation. Hardware packet discriminators could route arriving packets to different network nodes, or it could be pure software for more modest dual deployment.