Self-Healing Protocol Implementations

In this talk we consider an extended failure model for communication software: Instead of having to handle only external failures like lost or corrupted data packets, invalid routes and crashing nodes, we request the software to also handle internal errors. Internal errors can be the unfaithful execution of some instructions or the partial loss of the code base. We then ask whether software is able to detect such internal errors (self-monitoring), is able to continue operation despite such errors (resilience) and is able to correct such errors (self-healing). As a first step towards a better understanding of this problem, we restrict our attention to a simple “knock-out” criteria: The challenge is to demonstrate a program that continues correct execution despite removal of an arbitrary instructions. Having such resilient protocol implementations would permit to distribute the corresponding execution circuits over a network without having to worry about unreliable execution platforms in the same way as we expect current protocols to handle unreliable communication channels.