Survivability through Run-Time Software Evolution

In this paper we present an architectural framework designed to increase the survivability of software agent nodes in a distributed system. A multi-layered model replaces the original node software. Original computational requirements of the node are retained in the lowest level. The upper layers of the model provide protective and supportive services. Model components are mutated to create behaviorally identical but structurally distinct software. The lowest level of the framework is composed of replicated components drawn from a pool of mutations. The diverse population provides for identification of faulty or compromised components through a voting technique implemented in a higher level of the architecture. Failure recovery automatically creates a different component population. Unless multiple components are simultaneously compromised, the node continues functioning during failure recovery. Preliminary test results of a prototype implementation are given. The tests show that the architecture is feasible and that, in the simulated test environment, overhead is acceptably low. Recovery from single component failures and multiple component failure is demonstrated.