Support for Dependable and Adaptive Distributed Systems and Applications

Distributed applications executing in uncertain environments, like the Internet, need to make timing/synchrony assumptions (for instance, about the maximum message transmission delay), in order to make progress. In the case of adaptive systems these temporal bounds should be computed at runtime, using probabilistic or specifically designed ad hoc approaches, typically with the objective of improving the application performance. From a dependability perspective, however, the concern is to secure some properties on which the application can rely. This thesis addresses the problem of supporting adaptive systems and applications in stochastic environments, from a dependability perspective: maintaining the correctness of system properties after adaptation. The idea behind dependable adaptation consists in ensuring that the assumed bounds for fundamental variables (e.g., network delays) are secured with a known and constant probability. Assuming that during its lifetime a system alternates periods where its temporal behavior is well characterized (stable phases), with transition periods where a variation of the network conditions occurs (transient phases), the proposed approach is based on the following: if the environment is generically characterized in analytical terms and it is possible to detect the alternation of these stable and transient phases, then it is possible to effectively and dependably adapt applications. Based on this idea, the thesis introduces Adaptare, a framework for supporting dependable adaptation in stochastic environments. An extensive evaluation of Adaptare is provided, assessing the correctness and effectiveness of the implemented mechanisms. The results indicate that the proposed strategies and methodologies are indeed effective to support dependable adaptation of distributed systems and applications. Finally, the applicability of Adaptare is evaluated in the context of two fundamental problems in distributed systems: consensus and failure detection. The thesis proposes solutions for these problems based on modular architectures in which Adaptare is used as a middleware for dependable adaptation of assumed timeouts