PhD Oral Exam - Abbas Yekanlou, Electrical and Computer Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Collaborative and distributed edge computing networks (CDECNs) improve service responsiveness by enabling edge servers to store, share, and reuse digital objects (DOs), including computation results, processed data, AI-related artifacts, model parameters, and chunk-level data components, close to end users. While DO reuse can reduce repeated computation and service delay, it also introduces a fundamental redundancy-management challenge. Redundant copies consume limited edge storage and increase coordination overhead, yet some level of redundancy is essential for low-latency access, service continuity, resilience under failures, and adaptability to changing demand. Therefore, redundancy in CDECNs is both a source of inefficiency and a resource for robustness. This thesis develops a unified redundancy-orchestration framework that reduces unnecessary DO redundancy while preserving key system-level requirements, including latency, freshness, resilience, and scalability. To achieve this goal, the framework is developed progressively from snapshot-based redundancy management, where deduplication and relocation are jointly optimized under storage and latency constraints, to dynamic freshness-aware control, where object age, heterogeneous validity horizons, and time-varying popularity are incorporated into long-term decision-making. The framework is further generalized to large-scale resilience-aware CDECNs with chunk-dependent DOs, probabilistic server and link failures, and deadline-aware service requirements. To address these increasingly complex settings, the thesis integrates optimization, long-term stochastic control, learning-based decision-making, and scalability-aware network decomposition. The results show that effective redundancy management cannot be achieved by maximizing storage savings alone. Instead, DO deduplication and relocation must be coordinated across space, time, and network structure to balance storage efficiency, service accessibility, freshness, robustness, and scalability. The main insight of this thesis is that redundancy should be treated as a controllable system resource: excessive redundancy wastes capacity, while strategically preserved and relocated redundancy improves the reliability, adaptability, and efficiency of collaborative edge infrastructures.