The research program "ENDIKTIS" has been concerned with the study of architectures and mechanisms that support the provision of quality of service for fixed and mobile networks (e.g. constant low delay in the delivery of data, small percentage of lost data, and high and constant throughput), as well as the development of tools and theoretical methodologies for performance evaluation of such applications. The main objective of the program has been the extensive evaluation of architectures and protocols that have been proposed for providing quality of service in both fixed and mobile networks via extensive and realistic analysis and comparisons of their performance. For the achievement of this objective, different techniques for the analysis of networks were used, such as, simulation tools of network architectures, pilot networks that we have developed, and mathematical analysis using formal methods for analysis of system performance.
Initially an in depth investigation and analysis of the most important factors that regularly influence the quality of service in mobile and fixed networks have been carried out, and criteria and parameters have been identified in order to proceed with the performance evaluation of network systems. Furthermore, existing and proposed architectures that support or are expected to support quality of service in the future, as the differentiated services have been examined thoroughly. Based on the criteria that have been identified, an extensive study and performance evaluation of architectures and protocols of concern have been carried out through simulations and pilot experiments. We have experimented with various parameters and different topologies. Based on the results and measurements that have become available, we have determined conditions and weaknesses of existing mechanisms for providing quality of service. Consequently, new improved proposals for robust provision of quality of service have been proposed, and have been evaluated as well.
Concurrently, research was carried out towards the development of formal methods oriented to the analysis and performance evaluation of network systems. For the modeling of these systems we have proposed the usage of the well-studied formalism of process algebras. First result of this work was the development of the application-specific process algebra QoSPA, a framework enabling the description of complex network systems compositionally and offering constructs for modeling their real-time and probabilistic behavior. The algebra offers as primitives the notions of “communication channel”, “resource failure” and operators such as “parallel composition”. This allows the easy modeling of the various layers that may constitute a network system as well as the description of probabilistic failure and randomized behavior that appeared in several of the protocols studied. The process algebra was equipped with a formally defined structural operation semantics, which maps process-algebra terms onto labeled transition systems (Labelled Concurrent Markov Chains) in a compositional manner. Finally, regarding the performance evaluation of network systems, algorithms for computing critical quality-of-service quantities, those of throughput and long-run average, were developed and implemented in a toolset.
Consequently, the above-mentioned methodologies were extended for the modeling and analysis of systems with stochastic behavior. Specifically, the process algebra was extended with a new construct for describing stochastic durations of actions, thus giving the possibility of modeling systems where random phenomena can lead to the appearance of actions at unknown times governed by exponential probability distributions. This algebra was also equipped with structural operational semantics, this time giving rise to a new type of transition systems combining probabilistic and stochastic behavior. This model, a hybrid between Discrete and Continuous Markov Chains, though little studied in the literature to date, proves to be very useful for the study of the network systems such as the ones we have been investigating, since they apply randomized algorithms within environments of highly stochastic nature. To support the performance evaluation and analysis of the model we worked in two directions: (1) we have proposed the notion of strong bisimulation and algorithms for deciding when two systems are indistinguishable with respect to the notion and (2) we have extended techniques proposed in previous models for the model checking and computation of quantities critical for the evaluation of their performance.