Benchmarking Edge Computing Systems for Autonomous Vehicles via CAV Motifs

Yifan Wang, Qisong Shi, Xizi Chen, Xiaohui Peng

Abstract: Connected and autonomous vehicles (CAV) have become a typical application scenario of edge computing. The complexity and diversity of CAV workloads make them challenging to deploy, profile, and benchmark on heterogeneous edge computing systems. To address this problem, this paper proposes CAV motifs, a new approach to benchmarking edge computing systems on CAV scenarios. We extract and implement 11 CAV motifs covering 4 major autonomous driving tasks and present a performance prediction method based on CAV motifs to benchmark the edge computing systems via a few evaluation data. The evaluation results show that the predicted execution time of CAV workloads by CAV motifs is \(79.82\%\sim 97.49\%\) of real CAV workloads. However, CAV motifs have \(54.95\%\) lines of code (LOC) in the average of the original CAV workloads.

External IDs:dblp:conf/bench/WangSCP24