Smart Shadow - Predictive computing resources allocation for smart devices in the mist computing environment

Abstract

The Internet of Things (IoT) is a technological revolution that has generated new opportunities in academia and industry. In this context, IoT enables the emergence of several new ecosystems and computing environments. One of these new environments that, in the view of some authors, is considered of high importance in the context of the IoT devices is Fog and Mist Computing (FMC). FMC uses computational resources located at the edge of the network, reducing the latency and bandwidth problems, when compared to the use of Cloud computing platforms focused on IoT applications, also called Cloud of Things (CoT). Both infrastructures Fog and Mist computing are located on the edge of the network, however, the Fog computing processing usually occurs at the gateway layer that connects the IoT devices with the Internet. On the other hand, Mist computing, although it is a subset of Fog computing, concentrates its processing in the direct neighborhood of the device. The FMC environment offers new opportunities and benefits, however, due to the considerable dynamism of the topology and heterogeneity of devices, new challenges also arise. This thesis focuses on the problem of how to handle with this dynamism considering the issue of predictive discovery of computational resources in this environment and, thus, proposing a predictive model based on collective knowledge of previous experiences of resource allocations used by IoT devices in this ecosystem. In the proposal, the problem is subdivided into three distinct sub-problems, as follows. The first is how to evaluate from the client device perspective if it is interesting to use the infrastructure of the Fog/Mist/Cloud computing. Subsequently, once the answer is positive for Fog or Mist computing, the work seeks to find mechanisms on how to maintain data in this highly dynamic environment of the network topology. To address this issue, the work proposes a bio-inspired self-adaptive hierarchy structure of devices that use epidemic models to address this problem. Finally, the work presents a prediction algorithm of resources based on collaborative filters combined with an estimator of temporal availability of the devices that are part of the FMC environment. The evaluation is done with simulation using the Contiki operating system and the simulator Cooja. The results suggest the effectiveness of the proposal, even in cases where the FMC environment is composed of few devices that follow a pattern of permanence behavior within the network.