Discoveries in microbenchmarking with JMH

Microbenchmarking is fraught with peril. Method inlining. Dead code elimination. Constant folding. False sharing. Loop unrolling. Bimorphic and megamorhpic call sites. This talk explores these fantastic mysteries using JMH, the excellent microbenchmark harness from Oracle developed under the OpenJDK project.

Speaker

Trask Stalnaker

Trask Stalnaker is a 16-year Java programmer, author of Glowroot, Portland native and alumnus of Stanford University (BS Mathematics).

Speaker: Paul Jungwirth

Benchmarking is an important part of developing database systems like Postgres. For businesses, benchmarking can guide important decisions around selecting a database management system, or evaluate configuration changes. However, while there are many standard benchmarks out there, setting up, configuring, running, and collecting results was left as an exercise to the reader.

Benchbase is an open-source benchmarking framework from Carnegie Melon University’s Database Group (https://github.com/cmu-db/benchbase). It helps you configure, run, and collect results from various standard benchmarks. It also provides a framework for authoring your own performance benchmarks, and wiring it up to the batteries-included configuration, running, and results collection provided by it.

Paul is going to walk us through Benchbase, and some recent work he has embarked on around creating a benchmark. If you are a veteran of benchmarking, or have never done a benchmark before, this talk is for everyone.

Now a bit about Paul. He’s a freelance software developer here in Portland, and has been building applications with Postgres since 2011. Paul has been working on the Postgres project itself. He’s authored many database extensions, and his contributions to Postgres include work on GiST indexes, multiranges, and SQL:2011 application-time features.