Evaluating Offloading Scalability Using a Multi-language Approach on Cellular Networks.

Offloading has been suggested in the literature as a mechanism to minimize problems related to computational and energy limitations commonly associated with mobile devices. So far, most offloading solutions involve only processes developed with the same programming language. In an Android environment, most of these solutions are based on Java programming language. However, recent studies have shown that Java presents problems due to high resource consumption and low performance, whether on resources constrained devices or powerful server machines. Thus, some works have evaluated offloading performance when it involves processes developed with different programming languages and have obtained good results. This paper evolves such works by conducting a study that analyzes 1) the multi-language offloading scalability and 2) the multi-language offloading performance when using networks other than WiFi. Thus, we conducted experiments in an emulated environment where a Java Android application offloaded matrices to be multiplied by Go or Java processes through three different types of networks. The results showed that the response time decreased up to 87% compared to local processing and that the multi-language approach scaled well up to 24 clients with Go servers and up to 12 clients with Java servers.