Stefan Kaes, a Rails core developer since 2005, and author of InfoQ's 2006 most viewed article on Rails Performance, has released Railsbench 0.9.2. Railsbench is a collection of ruby and shell scripts which make measuring raw performance of rails apps a snap. Benchmarking of HTTP POST requests is now supported and several other features have been added that affect the specification of the Railsbench benchmarks.yml configuration file, and the generation of PNG graphics.
Stefan put the new version into practice by comparing Rails 1.1-stable and 1.2-stable with a test suite made of simple rendering, action cached and session creation. The suites were run with different session options using pstore, ActiveRecordStore and SqlSessionStore.
From those tests, Stefan found out that 1.2-stable was about 20% slower than version 1.1. The full report comes with remarks and advise:
Stefan put the new version into practice by comparing Rails 1.1-stable and 1.2-stable with a test suite made of simple rendering, action cached and session creation. The suites were run with different session options using pstore, ActiveRecordStore and SqlSessionStore.
From those tests, Stefan found out that 1.2-stable was about 20% slower than version 1.1. The full report comes with remarks and advise:
The question perhaps remains: Did the Rails core team sacrifice performance for features on Rails 1.2? Will there be focus on performance for the next stable release?
- Previous reports on Rails 1.2 being twice as slow as Rails 1.1 are performance FUD
- Action caching could use some attention
- Never use pstore as your session container
- With the current Ruby interpreter it seems impossible to avoid slowdowns as the code base of Rails grows, due to slow GC and large interpreter overhead. YARV might change this.