Publications Repository - Gdańsk University of Technology

The idea of distributed computing systems has been gaining much interest in recent years owing to the growing amount of data to be processed for both industrial and academic purposes. However, similar to other systems, also distributed computing systems are vulnerable to failures. Due to strict QoS requirements, survivability guarantees are necessary for provisioning of uninterrupted service. In this article, we focus on reliability of overlay distributed computing systems, i.e. systems using a virtualized communications network deployed over an underlying physical network (e.g. the Internet). In particular, our article proposes a survivable distributed computing system using the 1 + 1 protection approach working as an overlay network built on top of an existing underlying communication network. Our proposal is thus in contrast to recent papers mainly related to Grid systems using a dedicated optical network to connect the computing sites. Since the considered optimization problem is NP-complete, apart from introducing the respective Integer Linear Programming (ILP) model, we also provide two suboptimal offline metaheuristic algorithms, namely: AlgTS, based on the Tabu Search method; and AlgGen, based on the genetic algorithm, to obtain the results close to the optimal ones in a reasonable time. Extensive numerical experiments, conducted to verify the efficiency the heuristic approaches against the optimal results, confirmed the benefits of our proposals. In particular, the obtained optimality gap was as low as 5–7%, on average.

Authors

• Krzysztof Walkowiak,
• Wojciech Charewicz,
• Maciej Donajski,
• dr hab. inż. Jacek Rak link open in new tab