Event Correlated Usage Mapping in an Embedded Linux System - A Data Mining Approach
Abstract: A software system composed of applications running on embedded devices could be hard to monitor and debug due to the limited possibilities to extract information about the complex process interactions. Logging and monitoring the systems behavior help in getting an insight of the system status. The information gathered can be used for improving the system and helping developers to understand what caused a malfunctioning behavior. This thesis explores the possibility of implementing an Event Sniffer that runs on an embedded Linux device and monitors processes and overall system performance to enable mapping between system usage and load on certain parts of the system. It also examines the use of data mining to process the large amount of data logged by the Event Sniffer and with this find frequent sequential patterns that cause a bug to affect the system’s performance. The final prototype of the Event Sniffer logs process cpu usage, memory usage, process function calls, interprocess communication, system overall performance and other application specific data. To evaluate the data mining of the logged information a bug pattern was planted in the interprocess communication, that caused a false malfunctioning. The data mining analysis of the logged interprocess communication was able to find the planted bug-patterna that caused the false malfunctioning. A search for a memory leak with the help of data mining was also tested by mining function calls from a process. This test found sequential patterns that was unique when the memory increased.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)