Server-side image processing in native code compared to client- side image processing in WebAssembly

Abstract: Today, companies are developing processor demanding applications on the web, for example, 3D visualization software and video and audio software. Some of these companies have a native desktop application written in for example C++. These C++ codebases can consist of several hundred thousand lines of code, and companies would therefore like to reuse their codebase in the web version of the software. This thesis makes a performance comparison between two different approaches that can be taken to reuse the C++ codebase. The first approach is to compile the C++ codebase to WebAssembly and run it on the client-side. The second approach is to compile the C++ codebase to native code and run it on the server-side. It is not clear which approach to take if the goal is to achieve low execution times. This thesis will therefore answer the question of whether a client-side solution inWebAssembly is faster than a server-side solution in native code. To answer this question, this project work looked at one use case namely image processing. Two different web applications were developed, one that did image processing on the server-side in native code, and another one that did image processing on the client-side in WebAssembly. Execution time measurements were collected from both web applications. The results showed that for most algorithms WebAssembly was a factor of 1.5 slower than native code, without considering the additional delay from the internet that will affect the web application that performs image processing on the server-side. If this delay is taken into account, the web application that performs image processing on the client-side inWebAssembly will be faster than the server-side solution in native code for most users in the world. If the Round-Trip Time (RTT) is 1 ms the required average throughput needed to make the two web applications equally fast is 249 Mbps (Google Chrome) or 226 Mbps (Firefox). Most users in the world do not have such a high average throughput.