Seminar Series Archive
December 7, 2018
11:00am - 12:00pm
Title:
Abstract:
The rapid advances of computer networks, cloud computing, big data, and consumer-graded electronics, result in many more computing nodes surrounding us that can offer diverse types of resources, including computation, communications, storage, and sensing, to multiple users. Integrating resources from devices scattering from the cloud data centers to end devices enables dynamic Internet-of-Things (IoT) analytics, which is referred to as cloud-to-things continuum. Deploying dedicated infrastructure to realize individual IoT analytics is time-consuming, tedious, and expensive. Therefore, research teams and even companies around the globe have been building managed, programmable, and multi-tenant smart spaces in different scales, ranging from smart buildings, campuses, communities, and cities, for the era of cloud-to-things continuum.
In this talk, I will share our recent work on dynamically deploying IoT analytics across multiple geographically-distributed computing nodes. I will discuss the problem of optimally deploying heterogeneous IoT analytics, such as object recognition and sound classification, on a given smart-space infrastructure, so as to maximize its
throughput. We mathematically formulate this deployment problem and design a suite of algorithms to solve it under different circumstances. Moreover, we conduct a detailed measurement study to derive real system models of the IoT analytics based on diverse quality levels and heterogeneous devices to facilitate the optimal
deployment decisions. We implement a testbed to conduct experiments, which show that the system models achieve reasonably good accuracy. Other novel applications enabled by cloud-to-things continuum, such as smart city and wearable Augmented Reality (AR) will also be discussed, if time permits.