The WebKit Open Source Project

Performance

Overview

Performance is a top priority for WebKit. We adhere to a simple directive for all work we do on WebKit.

The way to make a program faster is to never let it get slower.

We have a zero-tolerance policy for performance regressions. If a patch lands that regresses performance according to our benchmarks, then the person responsible must either back the patch out of the tree or drop everything immediately and fix the regression.

Common excuses people give when they regress performance are, "But the new way is cleaner!" or "The new way is more correct." We don't care. No performance regressions are allowed, regardless of the reason. There is no justification for regressing performance. None.

We have a number of benchmarks that we use to measure performance.

  1. The i-Bench test suite was the most well-known cross-browser benchmark. We used to use this for testing the performance. The suite was developed by Lionbridge, but unfortunately they no longer support it thus the suite isn't available now. We are looking for good alternatives.
  2. Our internal page load test suite (called the PLT for short) must also not regress. The PLT contains pages that are representative of sites that are encountered in the real world. The harness itself is built into Safari and is accessible from the Debug menu. You can actually make Safari run the PLT on any set of test pages. Unfortunately, the pages we use internally at Apple contain copyrighted material and therefore cannot be open source at this time. We hope that in the future sites will be willing to donate snapshots of their front pages so that an open source cross-browser test suite can be developed with content that is not as homogenous as the i-Bench pages.

Get Involved!

How can you help improve WebKit's performance?

Test for Regressions
If you have your own performance tests, run WebKit through them daily. Make sure that the performance of the sites you care about does not regress. Test the above benchmarks on your hardware to help double-check that there have not been any regressions. Remember, the best way to stay fast is to never let your code become slower.
Open Source Benchmark
We have discussed with Mozilla and Opera the idea of an open-source cross-browser benchmark. The stumbling block to the construction of such a test suite is that we need to get buy-in from high profile sites like Google, Amazon or Yahoo to use snapshots of their front pages in the benchmark. The benefits of having your site in such a benchmark are obvious, since browser vendors will make your sites faster as they optimize for the content of the benchmark.

If you work for one of these high profile sites we encourage you to contact us if you are interested in having your company contribute content to such a benchmark.

Profile with Shark
OS X now has an excellent profiling tool called Shark. If you find operations that are slow in WebKit, we encourage you to use Shark to isolate performance problems and file bugs with that information. Here is a link to get you started using Shark.