A common design of the Internet of Things (IoT) system relies on distant Cloud for management and processing, which faces the challenge of latency, especially when the application requires rapid response in the edge network. Therefore, researchers have proposed the Fog computing architecture, which distributes the computational data processing tasks to the edge network nodes located in the vicinity of data sources and end-users to reduce the latency. Although the Fog computing architecture is promising, it still faces a challenge in mobility when the tasks come from ubiquitous mobile applications in which the data sources are moving objects. In order to address the challenge, this article proposes a proactive Fog service provisioning framework, which hastens the task distribution process in Mobile Fog use cases. Further, the proposed framework provides an optimization scheme in task allocation based on runtime context information. A proof-of-concept prototype has been implemented and tested on real devices.
TopIntroduction
The information systems designed for integrating the Internet of Things (IoT) (Gubbi et al., 2013) are usually applying the global centralized model, in which the IoT devices rely on distant management systems. Such a model is considered to be a drawback in terms of agility (Bonomi et al., 2012). In many real-time ubiquitous applications such as augmented reality, environmental analytics, ambient assisted living, etc., mobile device users require rapid responses. However, the latency caused by the distant centralized model is too high, even though the mobile Internet speed has improved significantly during the last few years. To address this problem, Fog Computing (Fog) (Bonomi et al., 2012) introduces data pre-processing with the computers in the vicinity of the data sources and end-user applications located in the edge network of IoT systems.
In general, Fog computing resources, which are known as Fog nodes, are mediating devices that connect the edge network with the Internet. Some typical examples are industrial integrated routers (e.g., Cisco 829 Industrial Integrated Services Routers), home hubs or set-top boxes that are employed as wireless Internet access points together with embedded virtualization technologies (e.g., Virtual Machines) or containerization technologies (e.g., Docker containers (https://www.docker.com)), which allow clients to deploy software onto them. Compared to the traditional distant Cloud computing model, which requires sending all the data to the Distant Data Center (DDC) for the processing, Fog can provide much better agility.
Although Fog-driven IoT system provides explicit enhancement in performance, it also faces numerous challenges in terms of connectivity (Zhang et al., 2015), discoverability (Troung-Huu et al., 2014), efficient deployment (Ravi & Peddoju, 2014; Guo et al., 2016; Ceselli et al., 2017; Lin & Shen, 2017) and so on. While many of the previous works focused on Fog deployment for specific use cases, this paper aims to address the mobility issue raised in the case of integrating Fog with ubiquitous mobile applications.
Imagine a mobile ubiquitous care application that needs to provide real-time environmental information to its user by continuously collecting and processing data derived from the surrounding environment while its user is moving in outdoor areas. For improving the efficiency, the mobile device (i.e. delegator) is distributing its computational tasks to vicinal Fog servers (i.e. workers). However, the delegator may need to repeatedly resend the tasks to different Fog nodes, due to the dynamic nature of the mobile environment, where the limited wireless signal coverage of the Fog nodes could cause failure in delivering results.
Consequently, it raises a question:
How can the system avoid the situation that requires the delegator to re-send tasks to the other workers due to the failed process result delivery?
In order to address the question, this paper proposes a proactive task distribution framework for mobile Fog environments. The proposed framework consists of two core schemes:
- •
Proactive task distribution, which is an extension of the Work Stealing scheme (Loke et al., 2015) that provides the mechanism to hasten the speed of task distribution.
- •
Context-aware Work Stealing, which provides an optimal decision-making mechanism that helps workers (Fog nodes) to decide how they should participate in the distributed processes.
In essence, the contribution of the paper is to study the potential of applying context-aware Work Stealing scheme in Fog computing towards improving the mobility-awareness. The study provides new insights about how distributed systems can achieve the high-performance process migration in the edge networks. Although the study is based on a specific ubiquitous application use case, the involved theoretical design still provides an important foundation for the discipline of mobile distributed computing.