Development of compute-intensive web applications with native desktop performance

Abstract

A web browser Execution Environment (EE) with the capability of hosting and executing Compute-Intensive Applications (CIAs) with native-like performance, have long been sought after. To that end, various technologies have been developed, the most prominent one being WebAssembly Programming Language (WASM) and to a lesser extent JavaScript Programming Language (JS). However, having CIAs achieve native desktop performance within a web browser EE has been elusive, primarily because it lacks the required systems architecture and design with which to support it. With that in mind, an empirical hypothesis of: ”A web browser EE that can host and execute CIAs on any device with native desktop performance can be created”, and a research question of: “What software architecture and design does a new web browser EE need to comprise of in order to be able to host and execute CIAs with native desktop performance”, were formed. Methodologically, this study is rooted in Design Science Research (DSR), together with Concept-Knowledge (C-K) theory, and a positivist philosophical position. Furthermore, a deductive approach, experimental strategy, mono-method quantitative data collection and cross-sectional time horizon were used. The delineation was limited to the web browser EE only. Additionally, all living organisms were excluded from the study, limiting ethical considerations to the development of a new web browser EE. The Structured/Systematic Literature Review (SLR) aided in identifying the primary literature which, as expected, articulates the promising results of WASM, while other technologies such as Rich Internet Application (RIA) was also touched upon. Further to that, we also introduced problem areas within benchmarking, that being Operating System (OS) noise, after which we provided insights into how one might go about mitigating against OS noise. We then presented our system architecture and design, based on Virtual Machines (VMs) and their lightweight container sibling. To that end a Linux-based prototype called System23 (SYS23) was presented, incorporating components such as Control Groups (Cgroups), namespaces and Secure Computing Mode (Seccomp), which together form the SYS23 enclave. We then benchmarked the prototype using PolyBench/C, which was liberally used with the literature that was discovered. The findings suggest that the prototype supports our hypothesis and achieves a performance improvement over its WASM counterpart and comes to within 0.45% of its native equivalent. Through that, the singular research question was answered and a foundation for future research to build upon was also provided. Furthermore, our methodological, theoretical and practical contributions moved our field of study forward.