Flexible Function Embedding for Low-Latency UAV Microservices in Programmable Edge Clouds
Bo Pang, Weiting Zhang, Deyun Gao · 2025
In the context of 6G networks, Unmanned Aerial Vehicles (UAVs) are expected to deliver services for various applications, such as real-time video streaming, environmental monitoring, and autonomous decision-making. These complex computational tasks can be broken down into smaller atomic functions in the microservices architecture, which can then be deployed in edge clouds to enable flexibility and scalability. However, minimizing end-to-end latency for UAV microservices remains a significant challenge, as many microservices consist of multiple atomic functions that are distributed across edge clouds and suffer from the performance limitations of software-based solutions. In this paper, we propose PBox, a UAV microservices acceleration framework by embedding atomic functions into programmable data planes. The atomic functions are implemented as Match-Action Tables (MATs) in programmable hardware pipelines to achieve line-rate processing and programmability. Notably, the order in which these functional MATs are applied is flexible and customizable, with the goal of reducing the number of routing hops required to complete the entire microservice. We model the optimal MAT table embedding and service routing as a nested optimization problem and propose a sampling-based genetic algorithm to efficiently accommodate the specific requirements of different UAV microservices. We evaluate PBox using four UAV applications and build prototypes on BMv2 programmable switches. Our experimental results show that PBox reduces UAV microservice completion time by 33% to 79%, while also increasing the number of completed services by up to 46%.