Equivalence classes for named function networking

Named Function Networking (NFN) is a generalization of Content-Centric Networking (CCN) and Named Data Networking (NDN). Beyond mere content retrieval, NFN enables to ask for results of computations. Names are not just content identifiers but λ-expressions that allow an arbitrary composition of function calls and data accesses. λ-expressions are pure and deterministic. In other words, they do not have side effects and they always yield the same result. Both properties together are known to as referential transparency. Referentially transparent functions can be evaluated individually no matter where and in what order, e.g. geographically distributed and concurrently. This simplifies the distribution of computations in a network, an attractive feature in times of rising needs for edge computing. However, NFN is affected by a lacking awareness for referentially opaque expressions that are characterized by having changing results or side effects, i.e. expressions that depend on outer conditions or modify outer states.
The fundamental motivation of this thesis is to retrofit NFN with a clearer notion of referentially opaque expressions. They are indispensable not only to many common use cases such as e-mail and database applications, but also to network technologies such as software defined networking. We observed that many protocol decisions are based on expression matching, i.e. the search for equivalent expressions. Driven by this observation, this thesis explores possibilities to adapt the determination of equivalences in dependence of crucial expression properties such as their ability for aggregation, concurrent evaluation or permanently cacheable results. This exploration results in a comprehensive set of equivalence classes that is used for explicit attribution of expressions, leading to a system that is aware of the true nature of handled expressions. Moreover, we deliver a solution to support referentially opaque expressions and mutable states in an architecture that bases upon uniquely named and immutable data packets.
Altogether, the findings condense to an extended execution model. It summarizes how the attribution of expressions with equivalence classes influences specific protocol decisions in order to support referentially transparent as well as referentially opaque expressions. We believe that our approach captivates due to its generality and extensibility. Equivalence classes depend upon universal properties. Therefore, our approach is not bound to a specific elaboration like NFN. We evaluate the applicability of our approach in a few application scenarios. Overall, the proposed solutions and concepts are an important contribution towards name-based distributed computations in information-centric networks.