Enhancing performance of IoT applications with load prediction and cloud elasticity

摘要

Elasticity is one of the most important services of cloud computing, referring to the ability to add or remove resources according to the needs of the application or service. In particular to the Internet of Things (IoT) scope, the use of this facility becomes pertinent since IoT requires a middleware that should be capable to handle high volume of data at real-time. Data can arrive in the middleware in parallel as in terms of input data from Radio-Frequency Identification (RFID) readers as request-reply query operations from the users side. Solutions modeled at software, hardware and/or architecture levels present limitations to handle such load. In this context, this article presents Proliot - a proactive elasticity model that combines cloud and high performance computing to address the IoT scalability problem in a novel EPCglobal-compliant architecture. The model can be seen as a service that keeps the same API but offers an elastic EPCIS component in the cloud, which is designed as a collection of virtual machines (VMs) that are automatically allocated and deallocated on-the-fly in accordance with the system load. The Proliot contribution consists in a mathematical formalism that uses Autoregressive Integrated Moving Average (ARIMA) and Weighted Moving Average to predict the IoT load behavior, so anticipating scaling in and out operations and then delivering VMs as close the moment they will be required as possible. Based on the Proliot model, we developed a prototype that was evaluated with different workload patterns against two concurrents: a threshold-based reactive elasticity model and non-elastic solution. The results were encouraging in favor of Proliot, presenting significant performance gains in terms of response time and request throughput.