Robust Network Services with Distributed Code Rewriting

Abstract

Nature does not know the concept of a dedicated controlling instance; instead, “control” is an emergent phenomenon. This is in stark contrast with computer networking where protocol control loops are (seemingly) in charge: while the functional aspect of a networking service can be well mastered, the dynamic behavior is still difficult to understand and even control. In this chapter, we present a methodology how to design distributed software systems that are dynamically stable and robust in execution. It is based on continuously replicating a system’s own code base in order to thwart unreliable execution and even accidental code changes. The crucial part is to design the system such that it regulates its own replication. This can be achieved by an execution environment inspired by chemistry to which we add the concept of self-rewriting programs (Quines). With a link load balancing example we show how to exploit competition and cooperation in a self-rewriting service implementation.