Survivability in layered networks

Abstract

In layered networks, a single failure at the lower (physical) layer may cause multiple failures at the upper (logical) layer. As a result, traditional schemes that protect against single failures may not be effective in layered networks. This thesis studies the problem of maximizing network survivability in the layered setting, with a focus on optimizing the embedding of the logical network onto the physical network. In the first part of the thesis, we start with an investigation of the fundamental properties of layered networks, and show that basic network connectivity structures, such as cuts, paths and spanning trees, exhibit fundamentally different characteristics from their single-layer counterparts. This leads to our development of a new crosslayer survivability metric that properly quantifies the resilience of the layered network against physical failures. Using this new metric, we design algorithms to embed the logical network onto the physical network based on multi-commodity flows, to maximize the cross-layer survivability. In the second part of the thesis, we extend our model to a random failure setting and study the cross-layer reliability of the networks, defined to be the probability that the upper layer network stays connected under the random failure events. We generalize the classical polynomial expression for network reliability to the layered setting. Using Monte-Carlo techniques, we develop efficient algorithms to compute an approximate polynomial expression for reliability, as a function of the link failure probability. The construction of the polynomial eliminates the need to resample when the cross-layer reliability under different link failure probabilities is assessed. Furthermore, the polynomial expression provides important insight into the connection between the link failure probability, the cross-layer reliability and the structure of a layered network. We show that in general the optimal embedding depends on the link failure probability, and characterize the properties of embeddings that maximize the reliability under different failure probability regimes. Based on these results, we propose new iterative approaches to improve the reliability of the layered networks. We demonstrate via extensive simulations that these new approaches result in embeddings with significantly higher reliability than existing algorithms.