Efficient Resource-Aware Hybrid Configuration of Distributed Pervasive Applications

Abstract

As the size and complexity of Pervasive Computing environments increases, configuration and adaptation of distributed applications gains importance. These tasks require automated system support, since users must not be distracted by the (re-)composition of applications. In homogeneous ad hoc scenarios, relying on decentralized configuration schemes is obviously mandatory, while centralized approaches may help to reduce latencies in weakly heterogeneous infrastructure-based environments. However, in case of strongly heterogeneous pervasive environments including several resource-rich and resource-weak devices, both approaches may lead to suboptimal results concerning configuration latencies: While the resource-weak devices represent bottlenecks for decentralized configuration, the centralized approach faces the problem of not utilizing parallelism. Instead, a hybrid approach that involves only the subset of resource-rich devices is capable of rendering configuration and adaptation processes more efficiently. In this paper, we present such a resource-aware hybrid scheme that effectively reduces the time required for configuration processes. This is accomplished by a balanced-load clustering scheme that exploits the computational power of resource-rich devices, while avoiding bottlenecks in (re-)configurations. We present real-world evaluations which confirm that our approach reduces configuration latencies in heterogeneous environments by more than 30% compared to totally centralized and totally decentralized approaches. This is an important step towards seamless application configuration.