Abstract

Deterministic networking systems, such as Time-Sensitive Networking (TSN) and Deterministic Networking (DetNet), require strict end-to-end delay guarantees while efficiently utilizing limited network resources. Conventional approaches typically focus on fixed traffic profiles, which can lead to suboptimal resource utilization in scheduling or admission control problems. This dissertation investigates traffic reprofiling—the proactive reshaping of traffic arrival patterns—as a complementary mechanism for improving both resource efficiency and delay performance under strict service guarantees. The dissertation consists of three parts. The first part studies bandwidth minimization under hard delay constraints for Service Curve Earliest Deadline First (SCED) schedulers. We show that traffic reprofiling can significantly reduce the required link bandwidth by smoothing burstiness and aligning traffic with the service curve abstraction. We develop algorithmic frameworks that compute optimal or near-optimal reprofiling strategies under SCED, leveraging its structural properties to achieve efficient solutions. The second part extends the study of bandwidth minimization to static priority and FIFO schedulers. Compared to SCED, these schedulers impose different analytical challenges due to their discrete priority structure or lack of prioritization. We develop corresponding reprofiling algorithms tailored to these scheduling disciplines and demonstrate that substantial bandwidth savings can still be achieved, highlighting the generality of traffic reprofiling across diverse scheduler models. The third part shifts the focus from resource minimization to delay performance in FIFO networks. Static reprofiling typically incurs a fixed shaping delay that can be overly conservative under varying network conditions. To address this limitation, we investigate dynamic traffic shaping mechanisms that adapt traffic profiles to current system states. We develop efficient algorithms, including learning-based approaches, to determine shaping policies under time-varying conditions. Our results demonstrate that, even in FIFO networks, it is possible to substantially reduce average delay while preserving strict worst-case delay guarantees. Overall, this dissertation establishes traffic reprofiling as a unifying design principle for deterministic networking. By systematically exploring how to configure traffic profiles across multiple scheduler abstractions, it provides new theoretical insights, algorithmic tools, and practical guidelines for designing networks that are both resource-efficient and performance-aware.

Committee Chair

Sanjoy Baruah

Committee Members

Chongjie Zhang; Henry Sariowan; Patrick Crowley; Roch Guerin

Degree

Doctor of Philosophy (PhD)

Author's Department

Computer Science & Engineering

Author's School

McKelvey School of Engineering

Document Type

Dissertation

Date of Award

4-29-2026

Language

English (en)

Download

Share

COinS