CenturionStudio.it - Fotolia

Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

What role does the end-to-end principle play in SDN?

The end-to-end principle was a core tenet of the early Internet, but where does it fit in the software-defined network? Analyst John Burke explains.

Software-defined networking (SDN), in its various incarnations, is relatively agnostic to the end-to-end principle. The end-to-end principle basically says: The endpoints in a network conversation should be responsible for guaranteeing they get the levels of reliability they need from the network. Trying to push that work to the network itself will either (a) fail, because the requisite level of reliability cannot be achieved there; or (b) make the network devices too complex and expensive, and slow down other traffic that doesn't need the added features. The Internet Protocol -- and the Internet itself, as originally built -- is an embodiment of the end-to-end principle (EEP) at work.

The EEP is often misinterpreted as advocating "dumb pipes" networking, but it's not that simple. The principle has nothing to say about intelligence in the network that is aimed at something other than providing application-to-application communications. We are building more intelligence into networks all the time! Security is a great example, such as Network Address Translation gateways or firewalls. This kind of intelligence in the network is ancillary to helping network nodes have conversations; it serves other purposes.

We have also, however, spent years building networks that directly violate the end-to-end principle -- with, for example, appliances such as WAN optimizers. Such devices dramatically improve the reliability and performance of network conversations between endpoints, e.g., by protocol-spoofing to speed up chatty applications, or local data-caching to avoid retransmission of data over high-latency connections. They do so without breaking the bank (and, generally, save money in the long run) or breaking performance for un-optimized traffic, and reduce the load on endpoints in the process.

Classic, OpenFlow-style software-defined networking (SDN) doesn't inherently support or break the end-to-end principle. We can use it to implement the dumbest and smartest of networks. SDN, however, provides a new and remarkably flexible way to distribute intelligence in the network by allowing relatively dumb devices (data plane switches) to act smart on behalf of SDN applications (taking the place of things like WAN optimizers) and at the behest of the SDN controller.

Network virtualization-style SDN looks friendlier to the end-to-end principle, since in many incarnations it treats physical networks as featureless pipes over which functional networks are laid. Again, though, those overlay networks can already include intelligence in the form of virtualized appliances, so this style of SDN is ultimately able to support or subvert the end-to-end principle as well.

Next Steps

Learn more about end-to-end network application monitoring

Will IPv6 bring back the end-to-end principle?

Are dumb networks smarter?

This was last published in September 2015

Dig Deeper on Software-defined networking