On-line Measurement-based Capacity Allocation Schemes

Abstract

Today's high-speed packet-switched networks are faced with the task of handling an increasing amount and variety of services, requiring different QoS constraints. To cope with this demand, the networks need dynamic and measurement-based resource allocation algorithms. For this task, the choice of appropriately accurate but also "practically implementable" algorithms is crucial.

In this thesis, we perform a comparative study of alternative on-line algorithms, we analyze their complexity, and perform comparisons via simulation experiments. Our motivation is to use these algorithms in the data plane of "self-sizing" frameworks, and make use of their output in taking control plane decisions either locally or globally, in an "on-line" fashion.

Due to the dynamic characteristics of the algorithms, we encounter the choice of time resolution, namely the setting of measurement time scale and window. After numerous simulations, we gain insight on the critical effect of these choices on the performance of the algorithms. We deduce that the time scale parameter itself is to be determined "dynamically" so that measurement-based algorithms can perform successfully independent from the varying traffic conditions. Finally, we demonstrate the effectiveness of this new approach over the static one, in our measurement-based capacity allocation algorithms.