CSR: Small: FaaSEdge: System Support for Function-as-a-Service at the Edge

The ubiquity of cameras and other sensors in our environment coupled with advances in computer vision and machine learning has enabled several novel situation awareness applications combining sensing, processing, and actuation. Many of these applications are latency sensitive, network bandwidth hungry, and geo-distributed. Edge computing has emerged as a new trend in catering to these computational needs by moving computers physically near the sensors, and is driven in part by low-cost processing resources such as Raspberry Pi and NVIDIA Jetson Nano.

In the 5G/6G evolution, cell phone companies are moving network-level packet processing to geo-distributed edge data centers to amortize the computational cost of the proliferation of the wirelessly connected devices. Function-as-a-Service (FaaS) (which allows users to run code without managing server infrastructure) is an appropriate allocation paradigm to increase the utilization of resource-constrained Edge sites that host such applications, but to date its design has targeted Cloud datacenters.

The project’s novelty rests in addressing the gap that exists in efficient system support for FaaS for a geo-distributed Edge ecosystem. The project’s broader significance and importance rest on the assertion that Edge computing could well be the next wave of disruption due to the emerging influence of AI and ML pervading all walks of future human endeavor.

Thus, the technology nuggets from this project could spur such disruption in the technology landscape.

The project is an end-to-end software system architecture for latency-sensitive situation awareness applications using the FaaS paradigm on geo-distributed Edge infrastructure. Specifically, the project makes the following advances to the state-of-the-art: programming idioms and associated runtime machinery that ensure the spatial and temporal correctness of the applications deployed on the Edge infrastructure without requiring distributed systems expertise which is largely unavailable to most developers; software control plane techniques for efficient federated orchestration across the Edge infrastructure while preserving the application’s service level objectives (SLOs) and ensuring spatio-temporal application correctness; data plane techniques for the efficient management of application state at each Edge site with respect to spatial and temporal correctness; and nimble execution environments to quickly launch application components while conserving limited Edge resources.

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.