Date of Award

Summer 8-2016

Embargo Period


Degree Type

Dissertation (CMU Access Only)

Degree Name

Doctor of Philosophy (PhD)


Electrical and Computer Engineering


Bruno Sinopoli


Resource allocation schemes play an important role in large-scale smart infrastructures to ensure efficiency and fairness among users. However, designing good resource allocation schemes is challenging due to technical limitation, policy barriers, and cost of change. The goal of this thesis is to develop a methodology to design model-based, principled and practical resource allocation schemes. Given the diverse characteristics of infrastructures, it is difficult to design unified models and algorithms. Instead, we employ a case-study-based approach on two representative smart infrastructures: Internet video delivery and electric power networks. We further generalize the insights and develop a principled qualitative guideline to design resource allocation schemes in smart infrastructures. The Internet video delivery system employs a protocol-based resource allocation scheme: network bandwidth is implicitly allocated by transport layer protocol (TCP) while client- side video players adapt video quality based on application layer protocol (MPEG-DASH) to optimize users quality of experience (QoE).We study 1) how client-side video players improve users QoE by employing Model Predictive Control-based bitrate adaptation algorithms and 2) how to achieve multiplayer QoE fairness by router-side bandwidth allocation policies. We prototype and evaluate the algorithms in real video players. On the other hand, market-based schemes are adopted in real-time economic dispatch in electric power systems to satisfy demand by lowest-cost generation. However, such schemes can lead to power imbalances and market inefficiency when slow generators fail to follow system operators command. We study 1) how system operators can mitigate power imbalance by employing a centralized, two-stage robust dispatch and 2) how the market design can be improved by penalizing non-complying generators. Based on the lessons from the case studies, we develop a general methodology to design resource allocation schemes: First, develop a formal model capturing system objectives, dynamics, and constraints; Second, identify key practical constraints that have major impact on the choice of schemes; Finally, design model-based schemes that respect practical constraints for short-term and obtain insights to inform protocol or market improvement in the long run. We envision that a mathematical theory can significantly improve the future resource allocation ecosystems.