PhD Paper Requirement - Travis Mayberry

Dedicated vs. Distributed: A Study of Mission Survivability Metrics
Agnes Chan, Joshua Haines, Andrew Johnson, Travis Mayberry, Hamed Okhravi

Abstract:
A traditional trade-off when designing a mission critical network is whether to deploy a small, dedicated net- work 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 met- ric 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.

Full Paper

Conference

MILCOM 2011
Acceptance Rate: 35%
Program link

Student Contribution

This paper was accepted and presented by Travis Mayberry at MILCOM 2011. MILCOM is a well-regarded conference in communication network, especially for defense-related research work. Attendants are mainly researchers from defense-related laboratories such as Sandia and NRL, as well as academicians from different universities.

Travis worked on the problem in the paper with people at MIT Lincoln Lab as an RA for Spring and Summer semesters. During these two semesters, he studied the Birnbaum Importance Measure on his own and came up with the quantitative analysis and a new metric for measuring network reliability. He implemented the algorithm and ran the simulations. He also wrote the paper and presented it at the conference.

It is not clear at this point whether the work will be included as part of his thesis since Travis is interested in cryptography and will be concentrating in that area instead. Nonetheless, Travis has shown independence in carrying out research and an ability to present his work clearly, both in written and in oral forms.