Analysis and Optimization of Javascript Engines

JavaScript is an interpreted computer programming language embedded into web pages that allows the creation of sophisticated solutions in the client-side web. It consists of a core programming language together with a host environment, namely, the Document Object Model (DOM) provided by the web browser. JavaScript has to be interpreted and executed by a dynamic translator inside the web browser, unlike desktop applications, which execute optimized code for a specific operating system. This dynamic translator is commonly called the JavaScript engine. Google uses V8 in Google Chrome and Node.js, Apple uses Nitro in Safari, Mozilla uses Rhino in Firefox and Microsoft uses JScript engine in Internet Explorer. Important efforts to improve the performance of these engines have been done in the last years. In this work, we focus on V8 JavaScript engine, which is an open source and widely used JavaScript engine developed by Google that ships with Chrome web browser. V8 is a dynamic compiler and therefore, program execution and code generation have to be efficiently synchronized in order not to affect responsiveness. It focuses on optimizing hot functions (i.e. those that execute more often). In this regard, V8 integrates two compilers, one that runs fast (i.e. has light overhead) and produces generic code (Full-codegen); and one that does not run as fast but generates more optimized code (Crankshaft). There is a significant number of executed instructions that are not part of the JavaScript application itself. These instructions can be considered as overhead produced by the V8 engine, which is a consequence of the characteristics of the JavaScript language along with the dynamic compilation nature of V8. This work deals with V8 targeting x86-64 processors. We have quantified two kinds of overhead in the JavaScript execution in steady state. The first one is the V8 runtime overhead needed to support native execution (i.e. garbage collector, compilation, etc.). The second one is the additional checks and guards introduced by the dynamic nature of JavaScript. These two types of overhead represent 35% and 25% respectively of the total execution time in average There is a great opportunity to improve the performance of JavaScript platforms by reducing these overheads. Therefore, we propose using the insights gained from this study to improve certain parts of V8, concretely the overhead due to the dynamic nature of JavaScript. The optimizations that we propose are based on hw/sw codesigned techniques that incorporate new x86-64 instructions tailored for dynamic languages. 2. PROPOSAL