The Bloated IP Protocol
By Rui L. Aguiar, Universidade de Aveiro/Instituto de TelecomunicaC'C5es

Abstract:
Early objectives of the datagram ( :-) ) networks aimed to simply 
provide reliability. IP appeared as a way of having a simple distributed 
communication  system, without a central control entity. Today, IP is 
looked at as the communication panacea, able to provide the 
infrastructure to the "communication society", incorporating security, 
multimedia, quality of service, mobility, multicast, and so on...
    Claim: Somewhere on the way from the IP protocol to the Internet 
protocols, we lost track of what we were doing.

 

Lets start by one of the objectives of the IP protocol, years ago: 
"simple, unmanaged". Is this what our research is doing? A simple answer 
would be "yes, we are using only/mostly IP". But this neglects the 
fundamental separation between control and data planes of a 
communication network.

When we faced the problems of quality of service, we happily set our 
minds in the development of (not one) two different frameworks, 
requiring a more or less complex control plane, and adequate signalling 
protocols. The fact that these techniques were never originally planned 
to operate in a "disruptive network" environment did not affect the 
research work.

When required to incorporate multicast, the break was even more radical. 
Now, the whole approach was to develop a separate control (routing 
plane) overlaid to the IP network.

Mobility has been a disruptive challenge for IP networks. After all, 
they were never developed for route changes every second. So a different 
way of addressing this appears, with an juvenile notion of identifier 
and locator inside MIP. The problems face with these led then to the 
development of multiple proposals for speeding the mobility process, 
bringing more complexity and still a not satisfactory solution. The fact 
that current telecom networks operate easily with concepts such as 
mobility and number portability seems to have been somewhat neglected in 
our claims of mobility improvements.

Security is one of the least complicated developments - curiously it is 
one of the few that affects significantly the data place, and one of the 
few that will probably face several law problems, with requirements like 
legal data interception.

For multimedia, the developments were, to say the least, a reflection of 
the whole story of IP. Besides many other small control protocols, IETF 
came with SIP - a great small thing in the beginning. The current amount 
of features that SIP can provide today lost many years ago its original 
simplicity.

 

So, are we still trying to have simple distributed system, without a 
central entity?

Let's go back to this first objective of IP development. "reliable 
distributed system, without a central control entity". Well, root DNS 
attacks for DoS is a current major concern, a source of potential 
economic hazards - and a favourite target for high-profile hacking. On 
the other hand, communications infrastructure today is no longer the 
circuit switch of 30 years ago - we are now on the digitized 
cell/TDM/WDM world. Yes, many of these were developments influenced by 
the datagram concepts. But current restoration capabilities of telecom 
providers are several orders of magnitude above those achievable by 
route adaptation inside IP networks, with recovery times inferior to one 
second. And IP is being transported in networks with these capabilities 
- in an environment where more and more the physical access to any site 
will become a single wire administratively shared by different entities: 
even with two different ISPs, an institution may be connected to the 
Internet through a single line, which subverts the whole idea of having 
IP as a reliable communication infrastructure - if the underlying 
carrier infrastructure does not recover faults, IP may well be 
physically unable to do so.

A complete different point to this "no central entity" objective is the 
fact that this no longer means an assurance of reliability. With the 
development of virus technology, distribution may actually mean an 
increase on the number of points of failure, that can later contaminate 
the whole network. In fact, current security practices move on the 
direction of a central control entity, blocking and controlling access 
to resources.  And this goes without discussing business aspects - 
operators, service providers expect to be paid (even on community 
networks some kind of reward needs to be in place).

 

So, what are we doing?

    Claim: IP needs to be rethought - it is not now the solution, but 
the problem. We do not have any longer a simple control plane for IP 
networks. It would be an error to keep it as the basis for the future: 
it will be hard enough that we have it as legacy.

 

So, if we are talking about new networking infrastructures.... let's 
keep IP in its small place, and exploit the fundamental requirements on 
communications - and not on IP networks.