Wavelet-based Neighborhood Control for Self-Sizing Networks

Abstract

The exponential growth of the Internet has turned it into a multiservice complex network of heterogeneous elements with dynamically changing traffic conditions. To regulate such a large scale network it is necessary to place intelligence in the nodes and find simple distributed rules and strategies that can produce meaningful and consistent behavior. These control mechanisms must be adaptive to effectively respond to continually changing network conditions. A “self-sizing” network can allocate link/switch capacity automatically and adaptively using online traffic data. Such adaptive, distributed, localized mechanisms are crucial to provide a scalable solution for controlling large, complex networks. In this paper, we propose a new, distributed self-sizing framework for locally controlled networks, which can support the stringent requirements of real-time applications in the Internet. Our unified and critical study of online resource allocation algorithms of two different classical approaches, led us to the use of adaptive multi-resolution decomposition (“wavelet”) algorithms. Our results show that by performing online resource allocation at each node based on their local knowledge, we can achieve considerable bandwidth savings and also satisfy QoS at the packet level. In our novel “neighborhood control” technique, we establish that by increasing the knowledge of some nodes so that higher self-sizing gains can be attained.