Separated Linux Process Logging Mechanism for Embedded Systems

This paper describes a low-overhead and prolonged Linux process logging mechanism that is separated into networks for embedded systems with small memory. It is useful to log the behavior of Linux processes in Android in order to debug, test, and learn about the embedded systems. However, the logging system interferes with process scheduling and requires huge storage to capture the process behavior. We propose a separated Linux process logging mechanism to solve these problems. Two features are utilized to develop the mechanism.The first is an improved logging performance. When acquiring context switch log data from an operating system, some data may be lost because of limited memory. To solve this problem, we propose a low-overhead data generating, low-latency data gathering and sending mechanism. This has a lightweight binary and string data format that uses a context-dependent log compression algorithm.The second feature is the construction of a log data management system that is available over a network. The logging mechanism is separated into two parts - a Logging Environment (LE) and a Logging Monitoring Environment (LME) - to reduce the effect of the target device and increase the volume of log data stored area. The LE consists of a log generation part in the target embedded device and a log collection part on the server. The LME consists of various monitoring and debugging tools, such as a log analyzer or log visualizer. In order to utilize the tools on the LME, the generated log data is exported to the log collection server via the network by an embedded device. The collected log data can be converted to an application-friendly format, such as Common Trace Format (CTF) or JavaScript Object Notation (JSON), on the log collection server to reduce the overhead of the embedded device in the LE and be easily available to the LME tools. We utilize an HTTP-based communication protocol between the LE and LME.As a result of an evaluation, the latency of data gatheri- g and the overhead of data generation are decreased by 50%. The scheduling interference and the volume of log data also decrease. Our method is also comfortable transferring data across the network as well as CPU loading.