Violating the end-to-end principle

It’s sometimes convenient to divide communication systems into the end points that attach to a network and the network itself. This provides the framework for thinking about the end-to-end principle. This tells us that whenever possible, operations should take place as close to the end points as possible instead of being implemented in the network. Conventional wisdom tells us that the closer we follow the end-to-end principle, the easier it is to create reliable systems. This principle has guided the evolution of the Internet for many years. Is it still appropriate today?

There are certainly some cases where it’s proved to be useful to violate the end-to-end principle. It’s usually not practical to do content scanning and filtering at end points, for example. These work better when they’re implemented in the network instead, like at a gateway appliance or a firewall. That's where these functions are typically carried out these days, although it's also common to have the same functionality at the end points. An example of this is how virus scanning is often done at both an anti-virus appliance in the network as well as on a user's desktop.

Some types of encryption also work better when they’re implemented in the network instead of at an end point. This frees users from the burden of managing cryptographic keys, and can make technologies like encrypted email much easier to use. This has also proved to be a useful alternative to end-to-end encryption, and most encrypted email today is encrypted at a gateway appliance instead of at an end point.

Not all cases where it’s useful to violate the end-to-end principle involve security. Network address translation (NAT) is a useful technology that’s not implemented at end points but has nothing to do with security, but many of the examples where it’s useful to push functions away from end points seem to. Could this be a general principle: that security often needs to be implemented in the network instead at an end point? There seems to be a fair amount of resistance in the IETF to technologies that violate the end-to-end principle, so if this is true, we may never actually see standards for many useful security technologies.

Leave a Reply

Your email address will not be published. Required fields are marked *