An Empirical Study of Resource-Stressing Faults in Edge-Computing Applications
Maryam Pourreza, Priya Lakshmi Narasimhan · 2023
Our growing reliance on edge-computing applications makes it crucial to improve the reliability of edge-computing systems. With multiple classes of edge-computing applications, different types of faults, and different kinds of resources needed by the applications, it remains unclear which resource exhaustion has the most disruptive impact on the latency experienced by the edge applications. Without this information, it is challenging to determine which faults to prioritize when implementing fault tolerance for edge computing. To address this challenge, we conduct an empirical study on a representative edge computing environment using well-known edge-computing benchmark applications (DeFog and ComB) and injecting resource-stressing faults (via stress-ng and hping). Our study reveals that memory overloads, CPU cache thrashing, frequent context switching, and page faults are the biggest disruptors of latency for edge-based applications.