Frequent pattern mining for kernel trace data

Operating systems engineers have developed tracing tools that log details about process execution at the kernel level. These tools make it easier to understand the actual execution that takes place on real systems. Unfortunately, uncovering certain types of useful information in kernel trace data is nearly impossible through manual inspection of a trace log. To detect interesting interprocess communication patterns and other recurring runtime execution patterns in operating system trace logs, we employ data mining techniques, in particular, frequent pattern mining. We present a framework for mining kernel trace data, making use of frequent pattern mining in conjunction with special considerations for the temporal characteristics of kernel trace data. We report our findings using our framework to isolate processes responsible for systemic problems on a LINUX system and demonstrate our framework is versatile and efficient. Categories and Subject Descriptors D.4.8 [Operating Systems]: Pe...

• Type: Journal
• Year: 2008
• Where: SAC
• Authors: Christopher LaRosa, Li Xiong, Ken Mandelberg
• Keywords: Applied Computing, Frequent Pattern Mining, Kernel Trace Data, Operating Systems, SAC 2008