Software Aging in Image Classification Systems on Cloud and Edge

Abstract

Image classification systems using machine learning are rapidly adopted in many software application systems. Machine learning models built for image classification tasks are usually deployed on either cloud computing or edge computers close to data sources depending on the performance and resource requirements. However, software reliability aspects during the operation of these systems have not been properly explored. In this paper, we experimentally investigate the software aging phenomena in image classification systems that are continuously running on cloud or edge computing environments. By performing statistical analysis on the measurement data, we detected a suspicious phenomenon of software aging induced by image classification workloads in the memory usages for cloud and edge computing systems. Contrary to the expectation, our experimental results show that the edge system is less impacted by software aging than the cloud system that has four times larger allocated memory resources. We also disclose our software aging data set on our project web site for further exploration of software aging and rejuvenation research.