What's Wrong with My Benchmark Results? Studying Bad Practices in JMH Benchmarks
Performance issues can have a devastating impact on the perceived quality of the software. To avoid such problems, the performance of a software system needs to be thoroughly tested and microbenchmarking is a widely used method for precise performance evaluation of specific units of program code.
Microbenchmarking frameworks, such as Java’s Microbenchmark Harness (JMH), allow developers to write fine-grained performance test suites at the method or statement level. However, due to the complexities of the Java Virtual Machine, developers often struggle with writing expressive JMH benchmarks which accurately represent the performance of such methods or statements.
In this paper, we empirically study bad practices of JMH benchmarks. We develop a tool that leverages static analysis to automatically identify 5 bad JMH practices. Using this tool, we empirically investigate the occurrence of bad JMH practices on 123 open source Java-based systems and found that each of these 5 bad practices is prevalent in open source software.
Further, we manually fix 105 benchmarks across 6 projects and quantify the impact of each bad practice in multiple case studies. Our analysis shows that bad practices often significantly impact the benchmark results, distorting the performance counters with large effect sizes.
To validate our experimental results, we constructed seven patches that fix the identified bad practices in 57 benchmarks from six of the studied open source projects, of which six were merged into the main branch of the project. In this paper, we show that developers struggle with accurate Java microbenchmarking, and provide several recommendations to developers of microbenchmarking frameworks on how to improve future versions of their framework.
The contributions of this paper are as follows: 1. Our study is the first to investigate the prevalence of bad JMH practices on real open-source projects and find that bad JMH practices are common and widespread in Java projects. 2. Our study is the first to quantify the impact of bad JMH practices of real benchmark results. Our results show that bad practices often significantly impact benchmark measurements. 3. We provide a static analysis tool that identifies the occurrence of bad JMH practices in JMH microbenchmarks. This tool can be executed through a batch command using Maven, Gradle or Ant and can be embedded into the CI pipeline of a software project. Also, our tool can be integrated with Eclipse IDE, where the warnings about bad JMH practices are showing directly in the editor view and can be used as a guideline for developers during benchmark development.
The full-paper is published in the IEEE Transactions on Software Engineering and can be found at https://ieeexplore.ieee.org/document/8747433
Date of Publication: June 27, 2019