Constructing provenance-aware distributed systems with data propagation

Abstract

Is it possible to construct a heterogeneous distributed computing architecture capable of solving interesting complex problems? Can we easily use this architecture to maintain a detailed history or provenance of the data processed by it? Most existing distributed architectures can perform only one operation at a time. While they are capable of tracing possession of data, these architectures do not always track the network of operations used to synthesize new data. This thesis presents a distributed implementation of data propagation, a computational model that provides for concurrent processing that is not constrained to a single distributed operation. This system is capable of distributing computation across a heterogeneous network. It allows for the division of multiple simultaneous operations in a single distributed system. I also identify four constraints that may be placed on general-purpose data propagation to allow for deterministic computation in such a distributed propagation network. This thesis also presents an application of distributed propagation by illustrating how a generic transformation may be applied to existing propagator networks to allow for the maintenance of data provenance. I show that the modular structure of data propagation permits the simple modification of a propagator network design to maintain the histories of data.