Dedicated vs. distributed: A study of mission survivability metrics

A traditional trade-off when designing a mission critical network is whether to deploy a small, dedicated network of highly reliable links (e.g. dedicated fiber) or a large-scale, distributed network of less reliable links (e.g. a leased line over the Internet). In making this decision, metrics are needed that can express the reliability and security of these networks. Previous work on this topic has widely focused on two approaches: probabilistic modeling of network reliabilities and graph theoretic properties (e.g. minimum cutset). Reliability metrics do not quantify the robustness, the ability to tolerate multiple link failures, in a distributed network. For example, a fully redundant network and a single link can have the same overall source-destination reliability (0.9999), but they have very different robustness. Many proposed graph theoretic metrics are also not sufficient to capture network robustness. Two networks with identical metric values (e.g. minimum cutset) can have different resilience to link failures. More importantly, previous efforts have mainly focused on the source-destination connectivity and in many cases it is difficult to extend them to a general set of requirements. In this work, we study network-wide metrics to quantitatively compare the mission survivability of different network architectures when facing malicious cyber attacks. We define a metric called relative importance (RI), a robustness metric for mission critical networks, and show how it can be used to both evaluate mission survivability and make recommendations for its improvement. Additionally, our metric can be evaluated for an arbitrarily general set of mission requirements. Finally, we study the probabilistic and deterministic algorithms to quantify the RI metric and empirically evaluate it for sample networks.