MUMIPLOP: optimal multi-core IoT service placement in fog environments

Abstract The Internet of Things paradigm paves the way toward automating real-world tasks, especially in intensive domains. Nonetheless, the criticality of the intensive tasks to be automated, such as video recognition, speech recognition, or data prediction, subjects them to very strict Quality of Service (QoS) requirements, such as short response times. To achieve these low response times, it is key to optimally place the microservices in a Computing Continuum infrastructure, considering their interactions in the form of workflows, their execution times, and the latencies of the underlying network fabric. While there are some proposals in the current state of the art to optimally place the application’s microservices, these proposals do not consider the effects that multi-core processing can have over the microservices’ execution times in their model, nor have their model compared with emulated network testbeds. This work proposes the Multi-core Microservice Placement Optimizer, a system that advances the state of the art by considering multi-core processing, making use of both the parallelization characteristics of microservices and the cores available at the computing devices. Our evaluation over a smart city case study, based on a real application and a fog computing testbed, shows that MUMIPLOP’s model is more accurate than single-core models, and yields shorter response time than state-of-the-art techniques, enhancing the QoS obtained in the Computing Continuum.