Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical and Computer Engineering

Committee Chair

Mark R. Bell

Committee Co-Chair

Xiaojun Lin

Committee Member 1

Borja M. Peleato-Inarrea

Committee Member 2

James V. Krogmeier


Resource allocation appears in various forms in communication and networking systems. In this thesis, we start with relay-based nanonetwork communications and analyze the message delay with bacteria-inspired message transfer carriers. Modeling the message transfer as a renewal process, we draw the connection to the mobility model of ad-hoc networks and show that the expected message delay between two nodes is a convex function of the number of available carriers. We then design a decentralized carrier-allocation strategy such that the systematic message delay is minimized. We show that the proposed policy can effectively reduce weighted delays when different message priorities are involved. Then, we study an agent allocation problem for the deep-target delivery application. To increase the delivery ratio of the agents to the deep target, we devise an agent coordination strategy so that the allocation of target-reaching and message-broadcasting agents can be balanced autonomously. We show that the system achieves a much higher delivery ratio of agents with the proposed coordination. We end the first part of the thesis with an investigation of a novel accelerated-diffusive channel. We show that by the proposed allocation of the molecule emission times according to their different travel times, the inter-symbol interference can be effectively reduced compared to the equivalent constant-drift channel.

In the second part of the thesis, we turn the focus to the optimal service allocation for online service platforms. When the system dynamics and user features are known beforehand, the operator can decide the optimal service allocation by solving a linear program. However, in reality, these information are generally uncertain and unknown. This motivates us to devise an adaptive control policy with learning capability. The resource allocation decision is therefore coupled with its feature-learning efficiency. We show that our algorithm achieves near-optimal performance and can generalize easily to time-varying dynamics and service rates.