Deep Root-Cause Performance Analysis: Integration and Navigation from a Load Testing Tool
2008 IBM Faculty Award Project
[Summary] [People] [Artifacts] [Presentations] [Education] [Related] [Sponsor]
PROJECT SUMMARY
Load testing tools such as IBM Rational Performance Tester can measure client-side response time and correlate it with collected application transaction monitoring data. Such correlation can break down client-side response time into its constituent parts on the various tiers of the server(s) under test. To avoid high overhead, some other tools can perform runtime profiling of CPU and memory usage by using JVM profiling with statistical sampling. However, such an imprecise profiling mechanism poses challenges for performance analysis. It is challenging to identify the correlation between a transaction or time-range (typically associated with different workload levels) on the client side and the profiling information at a server's JVM. Therefore, it is difficult to provide user-friendly drill-down navigation from client-side response time or time-range to determine problem spots and bottlenecks in the JVM(s) on the server sides because of a lack of correlation between the transactions or time-range driven by a load testing tool at the client side and the runtime sample data collected by the JVM profiling tools at the server sides. This project is developing new techniques and tools to help address the preceding issues in deep root-cause performance analysis.
PEOPLE
Faculty
Tao Xie (Principal Investigator)
Graduate Student
Yoonki Song (Ph.D. Student)
Past Graduate Student
Nuo Li (Ph.D. student)
Collaborators
IBM Rational Performance Tester Group (Kent Siefkes, David Chadwick, Joseph Toomey, et al.)
PROJECT ARTIFACTS
PRESENTATIONS
EDUCATIONAL ACTIVITIES
RELATED SOFTWARE & MATERIALS
SPONSOR
2008 IBM Faculty Award