Statistical Methods using Thread Intensity for Performance Monitoring
Oracle Fellowship Recipient
Dipanjan Haldar, Sanhita Bangalore SreenivasaMurthy
Oracle Principal Investigator
Kenny Gross, Architect
Zhen Hua Liu
This research collaboration effort between Oracle and Northeastern University aims for a provable theory of the thread intensity statistics as a foundation for a novel performance analysis tool for the Java applications. The theory will formalize the assumptions and limitations of the predictions based on these vital statistics. Some of the key contributions will be the algorithms to classify the stack traces, observe the bifurcations of a class of stack traces, and drill-down a hierarchy of classes of stack traces. The tool will complement the capability of Oracle’s JVM performance analysis and diagnosis tools with the capability matching those of various performance analysis tools such as Oracle Database Active Session History, Oracle Sun Studio 12 Performance Analyzer, Intel VTune Amplifier, AMD CodeAnalyst, and UNIX gprof command. Using the univariate or multivariate state estimation models based on the vital statistics, we can derive the qualitative perceptions of the states of the applications, platforms (middleware, database), and infrastructures (operating systems, network, hardware). The contributions from this research will enable future research on characterizing the feedback loops involving perceptions, diagnosis, decision-making, and control actions for assuring the service level agreements of the integrated systems involving software and hardware in the cloud operations.
Kenneth Baclawski and his colleagues developed realistic simulators for the non-Poissonian stochastic processes that model file access, transaction processing and communication patterns. He has supervised hundreds of student research projects, some of which were intended for military real-time, mission-critical systems.