Avaliação do desempenho do Webassembly: um estudo de caso utilizando reconhecimento de dígitos manuscritos

Abstract

The goal of the current work is to present a study on the effectiveness of web-based applications that use webassembly. A language and binary file format called Webassembly promises significant speed gains over javascript. In the study, two applications capable of processing and classifying images with handwritten data were developed. All of the image preprocessing algorithms and neural network algorithms for classification were written in C++ for the first application and compiled for webassembly using Emscripten; the same algorithms were written in Javascript for the second application. At the conclusion of the study, benchmarking tests were conducted using a fictitious dataset in the MNIST style. In conclusion, the study on WebAssembly showed that it has superior performance results in most browsers, allowing us to take advantage of the computational power of client-side users and reducing the overload of calls to the backend. The benchmarking tests were carried out on the browsers Google Chrome, Microsoft Edge, Mozilla Firefox and Brave. In short, WebAssembly is a promising technology and deserves to be explored by developers and companies, to take advantage of all its potential and take the performance of web applications to the next level.