[arch-d] Chaff in protocols

["Martin Thomson" <[e-mail]>]

Esteemed colleagues,

The IAB is looking for historical examples of deliberate additions to a protocol that are added not to enable new behaviour, but to test that the ability to do so remains viable.

This comes up in the context of assessing the concept of "grease" (see draft-ietf-tls-grease), a practice of adding meaningless extension points to messages in the hope that it will reveal, and perhaps prevent, intolerance toward real extensions.  This is a practice we have seen deployed with some success in TLS.  That has produced some enthusiasm about the practice.  I think that it would be fair to say that the IAB maintains some reservations about the general applicability of the technique.

Are there protocols (in either specification or deployment) that have experimented with this sort of thing in the past?

The concept of fuzzing software isn't new (see https://en.wikipedia.org/wiki/Fuzzing) and testing for resilience to unusual behaviour is similarly well-established (e.g., https://github.com/Netflix/chaosmonkey).  But are there cases of this in network protocols prior to its use in TLS?

Regards,
Martin