Abstract (eng)
Quality of Service (QoS) is an important aspect in distributed, service-oriented systems. When several concrete services exist that implement the same functionality, the choice of a service instance among many can be made based on QoS considerations, objectives and constraints. Typically considered properties are performance, availability, and costs. In this thesis, aspects of the QoS-aware service selection problem are studied in the context of a distributed, service-oriented system from ATLAS, a high-energy physics experiment at CERN, the European Organization for Nuclear Research. In this so-called TAG system, data and modular services are distributed world-wide and need to be selected and composed on the fly, as a user starts a request. There are two conflicting optimization viewpoints. The service selection is modeled as a dynamic multi-constrained optimal path problem, which allows considering QoS attributes of service instances and of the network. The dynamic aspects of the system are included in the problem definition, as they represent a specific challenge. To address these issues regarding dynamics and conflicting viewpoints, this work proposes a service selection optimization framework based on a multi-objective genetic algorithm capable of efficiently dealing with changing conditions by using a persistent memory of good solutions, and a stepwise adaptation of the mutation rate. A system and QoS attribute ontology as well as a description of dynamics of distributed systems build the basis of the framework. The presented approach is evaluated in terms of optimization quality, adaptability to changes, runtime performance and scalability.