CAVBench: A Benchmark Suite for Connected and Autonomous Vehicles

TL;DR

CAVBench is the first comprehensive benchmark suite designed for evaluating edge computing systems in connected and autonomous vehicles, covering diverse applications and providing insights into system performance characteristics.

Contribution

It introduces a standardized benchmark suite for CAVs, enabling systematic evaluation of hardware and software in this domain for the first time.

Findings

Applications have polarized operation intensity, emphasizing heterogeneous hardware needs.

All applications demand high memory bandwidth, requiring optimized memory management.

Some applications exhibit poor data/instruction locality, suggesting cache architecture optimization.

Abstract

Connected and autonomous vehicles (CAVs) have recently attracted a significant amount of attention both from researchers and industry. Numerous studies targeting algorithms, software frameworks, and applications on the CAVs scenario have emerged. Meanwhile, several pioneer efforts have focused on the edge computing system and architecture design for the CAVs scenario and provided various heterogeneous platform prototypes for CAVs. However, a standard and comprehensive application benchmark for CAVs is missing, hindering the study of these emerging computing systems. To address this challenging problem, we present CAVBench, the first benchmark suite for the edge computing system in the CAVs scenario. CAVBench is comprised of six typical applications covering four dominate CAVs scenarios and takes four datasets as standard input. CAVBench provides quantitative evaluation results via application and system perspective output metrics. We perform a series of experiments and acquire three systemic characteristics of the applications in CAVBench. First, the operation intensity of the applications is polarized, which explains why heterogeneous hardware is important for a CAVs computing system. Second, all applications in CAVBench consume high memory bandwidth, so the system should be equipped with high bandwidth memory or leverage good memory bandwidth management to avoid the performance degradation caused by memory bandwidth competition. Third, some applications have worse data/instruction locality based on the cache miss observation, so the computing system targeting these applications should optimize the cache architecture. Last, we use the CAVBench to evaluate a typical edge computing platform and present the quantitative and qualitative analysis of the benchmarking results.