CNS Core: Medium: A User-centric Adaptation Framework for Edge-Native Applications

This proposal addresses a new class of applications called "edge-native applications" that have enormous societal value in domains such as assisting handicapped users, enforcing privacy in video streams, and enhancing the productivity of just-in-time manufacturing. Edge-native applications are simultaneously compute-intensive, bandwidth-hungry, and latency-sensitive.

These attributes pose a fundamental challenge to scalability. The goal of this proposal is to develop new techniques for efficiently supporting large numbers of users of such applications, without hurting their quality of experience (QoE).

The proposed research is organized into four thrusts. Thrust-1 investigates how on-device processing and adaptive sampling of sensor data can reduce load on edge infrastructure, while minimally impacting QoE. This thrust also creates an API between the operating system and applications for adaptation.

Thrust-2 explores how to efficiently and seamlessly move work from overcommitted edge infrastructure to underutilized sites. It investigates both an application-transparent approach that is based on virtual-machine (VM) encapsulation, and an application-optimized approach that seeks to be frugal in data transmission. Thrust-3 creates tools and mechanisms to study QoE.

Using machine learning on history-based data that is dynamically collected, it builds models of user-specific and application-specific tradeoffs for mapping application fidelity to QoE. It also creates tools for QoE debugging of edge-native applications. Thrust-4 explores how multi-fidelity applications that dynamically vary QoE can be evaluated without performing user studies.

It develops a new evaluation methodology that is based on the concept of synthetic users, also known as "droids".

Through close partnership with industry and local government, this research will accelerate the emergence of transformative edge-native applications. Through integration with education and curriculum development, this research will provide unique learning opportunities for students in Computer Science, Electrical and Computer Engineering, and Human-Computer Interaction at the undergraduate and graduate levels.

Because of the central role of applications in this research, it offers many research opportunities for a diverse group of individuals, including those from under-represented groups.

Software developed in the course of this research will be released open source via GitHub (http://github.com). Benchmarks and experimental data will be released on an institutional website (http://elijah.cs.cmu.edu). All results generated through this research will be available and actively maintained for at least five years after the conclusion of the project or after the publication of the data, whichever is first.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.