Diagnosis of application server performance problems via thread level pattern analysis

Abstract

Task level breakdown provides high level information for operation performance bottleneck analysis. Creating a task level breakdown requires prior knowledge of the application server program logic and can be very time consuming. In this paper, we present yShark, a profiling tool which generates a task level breakdown chart automatically without prior knowledge. With thread level pattern analysis, performance bottlenecks can be easily located. We illustrate the innovative features of this tool with two real-world application case studies. First, we show how performance bottlenecks from modules both within and external to the system can be easily identified by the tool; second, we present a pattern detection and pattern matching algorithm to detect duplicate sequential tasks that can be run in parallel. We also show how this tool can be useful in VMware vCloud Director production environments due to the minimal profiling overhead.