Reconfiguration Abstractions for Gorums

Abstract

To provide the high availability expected of a modern data service, the service needs to be capable of adapting to a multitude of scenarios, e.g., increased demand and unexpected failures. A common technique for mitigating these issues and to improve upon the service's fault-tolerance, is to replicate the service to a set of independent machines. Gorums is a novel remote procedure call (RPC) framework designed for alleviating developers of the complexity associated with building distributed fault-tolerant services. Machines have a tendency to eventually fail. Therefore, to ensure continued availability of a service and to retain its fault-tolerance, we need to be able to replace faulty machines without disrupting the user's experience through reconfiguration.

This thesis examines and extends the Gorums framework with easy to use, available, and adaptable reconfiguration abstractions. By providing a generalized out-of-the-box reconfiguration component capable of introducing reconfiguration capabilities to arbitrary services, the extended Gorums framework is able to further alleviate developers of the complex, tedious, and time-consuming processes of building truly fault-tolerant services. The reconfiguration component masks the intricacies of reconfiguration and instead presents developers with an easy to use and intuitive reconfiguration API. An experimental evaluation of the extended Gorums framework and its reconfiguration component is also presented.