A Study on Optimization of Compilation Process for JavaScript Engine

Abstract

JavaScript is a standard dynamic language for web, along with HTML and CSS. Since the appearance of new standards and technologies such as HTML5, ECMAScript6, and WebGL, JavaScript has been increasingly undertaking complex computations, and even running in the server environment. Therefore, the JavaScript performance becomes a critical issue. For high performant JavaScript, modern JavaScript engines have adopted a multi-tier execution architecture based on the adaptive compilation, which compiles the source code differently based upon the hotness of each function. However, JavaScript engines still suffer from the substantial compilation overhead. This study proposes two optimization techniques for JavaScript compilation process to reduce the compilation overhead, which are concurrent parsing and ahead-of-time compilation (AOTC). Concurrent parsing approach focuses on the substantial parsing overhead during the web loading time. To reduce the parsing overhead, concurrent parsing approach performs the parsing process in advance on separate parsing threads, while the main thread directly executes the pre-parsed functions skipping the parsing process. When performed on multi-cores, this can hide the parsing overhead from the main execution thread, thus improving the overall web loading performance. Concurrent parsing technique implemented on a commonly used web browser shows a tangible improvement of the whole web loading performance for various real web apps, which is up to 32% and 18% on average. AOTC approach is proposed to reduce the whole compilation overhead. From an observation of compute-intensive JavaScript codes, the total compilation overhead accounts for almost half of the entire JavaScript execution time. This result is due to the heavy workload of the parsing and optimizing JITC. To reduce the whole compilation overhead, AOTC technique is employed for JavaScript which stores and reuses the compiled code generated in the previous run. A new AOTC method is discussed which reuses the bytecode and optimized code together for state-of-the-art JavaScript engines. Proposed AOTC method implemented on a commonly used JavaScript engine shows a substantial performance improvement for industry-standard JavaScript benchmark, which is up to 6.36 times and 1.99 times on average. Two proposed optimizations for JavaScript compilation process efficiently improve the performance of web loading and JavaScript execution with promising results.