PhD Defense: “An Emulation-Based Performance Evaluation Methodology for Edge Computing and Latency Sensitive Applications”

TECoSA PhD student Manuel Olguin will defend his thesis at KTH Campus (U61).  All with an interest in this topic are welcome to attend.

ABSTRACT:  Cloud Computing has greatly impacted our daily lives by providing global accessibility and virtually unlimited scalability. However, its centralized architecture prioritizing availability and scale has limitations for real-time processing and low-latency applications like Cyber-Physical Systems (CPSs) and mobile eXtended Reality (XR). Edge Computing is emerging as a solution to these limitations by bringing computation closer to the network edge, enabling real-time decision making. A key challenge to mass deployment of Edge Computing infrastructure relates to the complexity of evaluation the performance of these systems, which stems from the tight interaction of network and compute. This dissertation addresses this challenge by introducing a methodological approach for studying trade-offs in latency-sensitive applications deployed on Edge Computing which involves emulating the client-side workload while retaining the rest of the system. This maintains the realism of network and compute effects, offering advantages in efficiency with respect to fully experimental approaches and capturing complex factors that are challenging to model analytically or in simulations.

Furthermore, this dissertation explores the implications of this methodology on the potential for optimization in Edge Computing deployments, in particular with respect to improved accuracy in the emulation. In that context, the dissertation provides a realistic model of human timings for a specific class of Mobile Augmented Reality (MAR) applications which is combined with a mathematical framework for the optimization of sampling systems. The results show that the introduced methodology improves efficiency, repeatability, and replicability compared to existing methods. By integrating workload components into the emulated software domain, it reduces complexity while considering the complex effects of network and compute factors. The dissertation emphasizes the importance of incorporating enhanced realism in client-side emulation, which enables the implementation of optimization approaches that would otherwise be infeasible, and further highlights the significance of human behavior in addition to system-related metrics in the context of MAR, a killer use-case for Edge Computing.

Details of the panel are shown below:

Principal Supervisor: Professor James Gross
Defense Chair: Professor Mats Bengtsson
Opponent: Associate Professor Yu Xiao, Aalto University, Esbo, Finland

Members of the Grading Committee:
Professor Ana Aguiar, Universidade do Porto, Portugal
Professor Per Gunningberg, Uppsala University, Sweden
Professor Klaus Wehrle, RWTH Aachen University, Germany