Migração de uma plataforma de offloading para a abordagem de microsserviços

Abstract

Mobile devices are becoming increasingly present in people’s daily lives. However, the mobility provided by mobile devices imposes limitations such as less storage and processing capacity. Despite the substantial improvement of new generations of smartphones and other mobile devices, the amount of information and complexity of new applications created for these devices still impose certain restrictions on processing specific tasks, particularly concerning power consumption. This is a problem, specially for context-aware mobile applications, a particular class of mobile apps that use information gathered from the users’ execution environment to adapt their behavior to improve the user experience while using such apps. A promisse approach to mitigate this issue is Mobile Cloud Computing (MCC). In the context of MCC, some solutions emerge to assist in the decentralization of data processing and operations, also reducing the energy consumption of devices. One is the technique known as offloading. Over the last few years, various platforms for supporting offloading have been proposed, among them, the Context Acquisition and Offloading System (CAOS). Currently, CAOS has problems due to its monolithic architecture, such as tight coupling and lack of scalability. These two aspects are strongly connected. A recent approach that has received much attention to address monolithic systems is the use of microservices. The present study aims at proposing the migration of the monolithic version of CAOS into a microservices architecture, and consequently, to achieve the benefits that this architecture provides. This new version is called CAOS Microservices (CAOS MS). We performed two experiments to evaluate the CAOS MS. The former measured possible performance penalties that the microservices architecture could have suffered concerning the monolithic version. The latter verified scalability aspects provided by CAOS MS. Our experiments show us that CAOS MS presents similar performance than its monolithic version, but with improved scalability support.