Lightweight embedded software performance analysis method by kernel hack and its industrial field study

Abstract

Despite advances in software testing technologies, there are still limitations in directly applying them to embedded software. Since the operational environment of embedded software has severe resource constraints, it is necessary to develop a lightweight testing method that has little impact on the operational environment of embedded software. We propose an agent-based performance analysis method to hack kernel performance counters that manage the system's execution information. The proposed method enables us to collect data required for analyzing performance bottlenecks and identify the causes and locations of bottlenecks with little impact on the test target system's operational environment. We introduce a test automation tool called Analytic Master of System v2.0 that we developed by employing our proposed method. Presently, Analytic Master of System v2.0 is being utilized as a standard tool for performance testing of embedded systems in the automotive industry. In addition, we suggest a guideline for performance analysis and improvement by introducing an industrial field study among our best practices, which analyze the relationship between the memory fault processing of the operating system and the application processing speed. © 2011 Elsevier Inc.