Towards parallelizing the layout engine of firefox

The Mozilla Firefox browser currently accounts for ∼25% of the total web browsers market segment share, establishing itself as the second most popular browser world-wide after Microsoft's Internet Explorer. With the recent adoption of a tracing JavaScript Just-In-Time (JIT) compiler in Firefox 3.5, its performance has improved significantly, especially for web pages that make heavy use of JavaScript. Currently, the heavy performance hitter component of Firefox is its layout engine. According to our extensive performance measurements and analysis on representative applications running on Intel platforms, the layout engine of Firefox accounts for ∼40% of its total execution time, with the Cascading Style Sheets (CSS) rule-matching process being the hot part of layout. We have developed a tracing and profiling mechanism that has enabled us gain in-depth insight into the CSS rule-checking runtime characteristics. This data has proven extremely helpful in determining that the CSS rule-matching component is a suitable target for parallelization. The collected trace information has also guided us to an effective parallel implementation of the Firefox CSS rule-matching component that delivers large performance speedups on popular web pages, including almost 90% of Mozilla's page-load tests that comprise ∼400 distinct web pages from all over the world. For these tests, the entire page-load process shows performance changes ranging from - 0.5x to +1.84x, with an average improvement of ∼1.1x, when two worker threads cooperate in doing the layout. To our knowledge, this is the first fully functional parallel implementation of CSS rule matching that delivers significant speedups in a world-class browser.