Uma arquitetura baseada em Dew Computing e em Processamento Multi-linguagem para melhoria do desempenho computacional de aplicações móveis

Abstract

Low processing power and power limitations are typical constraints faced by most mobile devices when processing computational tasks. Several researches indicate computational offloading as a technique to face this challenge. In it, computationally and/or energetically limited devices transfer tasks to be executed on machines with greater capacity, saving time and resources. However, previous studies have demonstrated that adopting programming languages is inefficient when performing such tasks, and network latency impacts computational offloading performance levels. This thesis proposes the DADOS Architecture (Dew Architecture for Distribution of Offloading Servers), which attacks these two problems by incorporating the multi-language approach and the Dew Computing paradigm into computational offloading. When using the multi-language approach, DADOS allows interaction between processes developed in different languages, which enables the adoption of more efficient languages in offloading server processes, speeding up the execution of tasks and saving device resources. Furthermore, by using the Dew Computing paradigm, which reduces dependence on the network by allowing part of the remote services and data to be processed on the mobile device, DADOS allows bringing offloading processes into the device, mitigating adverse network effects. Experiments were conducted using real devices to validate the initial version of the architecture and evaluate the impact of the approach it offers on the execution of mobile tasks. The experiments evaluated the performance of three approaches (Local, Dew and Cloudlet) against three main metrics (response time, energy consumption, and efficiency) in two scenarios with different levels of overload on the network. The results were promising for the approach provided by DADOS, Dew. Comparing only the Dew and Local approaches, it was observed that the former was up to 7.2x faster and consumed up to 4.6x less energy than the latter in both scenarios. Between the Dew and Cloudlet approaches, Dew outperformed Cloudlet in specific conditions: when dealing with large volumes of data, Dew transmitted up to 2.6x less data. In environments with high network overhead, Dew processed tasks up to 4.5x faster than Cloudlet.