Displaying Trace Files

Computers generate trace files containing reports on system performance, status, and faults. To analyze these trace files more efficiently, we have developed a graphical technique embodied in an interactive system for displaying large trace files. Our system uses abstraction, color, aggregation, filtering, interaction, and a drill-down capability to find patterns among the reports. We apply our system and technique to analyze command accounting trace files from a Unix compute server, showing what commands were executed, by which users, when, and how long the commands ran. We identify resource intensive commands, sequences of commands initiated by a compilations, and commands run with super-user permissions.