Article Preview
Top1. Introduction
Recent advances in the cost, performance, and energy efficiency of IoT devices; network technologies (such as 4G and 5G) (Liu, 2013) and distributed computing architectures have led to the explosive growth of the Internet and mobile connectivity, in turn leading to new distributed applications in areas such as transportation, healthcare, mining, entertainment, and security, such as automated vehicles, augmented reality, cloud robotics, smart homes and cities, video surveillance and streaming, and Internet of Things (IoT) applications (Best-Rowden, 2018) This has led to an unprecedented growth in data as well as increased the importance of latency and regulation in handling and managing data. The new distributed applications have characteristics that may be bandwidth-hungry (video surveillance, video conferencing, traffic monitoring), latency-critical (automated vehicles, robotic surgery, safety), and may cause spikes in activity at places or times (sporting events). Applications may also require high availability, low jitter, and security. Large-scale deployment of IoTs and industrial IoT devices (Sun, 2017) is expected to play a big role in the development of smart cities, which will generate large volumes of aggregated cellular data that may choke the network. On the other hand, devices such as sensors on the power grid or on oil pipelines may host latency-sensitive applications that require low latency in order to ensure that mission-critical data is transmitted and processed in a timely manner so that potential damage to people, property, and the environment can be prevented (Wang, 2018). Online video games on consoles such as Xbox Live, where reaction times are in milliseconds, have become extremely popular recently. Such games may be hosted by a distant data centre (DC), so the presence of latency and jitter could have a significant effect on gamers’ experience and dramatically reduce their interest in the games. Virtual Reality (VR), Augmented Reality (AR), and other state-of-the-art human-computer interaction applications require low latency and rapid processing for complex rendering algorithms as well as large volumes of data that may need to be transferred between a user and a DC hosting the applications (Mao, 2017). There has been a rush of live streaming applications (Tran, 2017) such as SnapChat, Facebook Live, and YouTube Live, due to the proliferation of high definition (HD) video cameras on smart phones. Similarly, video surveillance applications will require high-performance computer resources to run artificial intelligence (AI) and machine learning (ML) technologies (Wang, 2018) to identify people and alert human operators in real time. These applications may generate gigabytes or even terabytes of data per second. All in all, there is a need for a compute infrastructure that can begin to address the challenges posed by these emerging applications. Today’s telecom networks are not even expected to handle the enormous and rapidly varying capacity demands that will arise soon. One of the challenges of using IoT applications to their fullest potential (Taleb, 2017) is figuring out how to handle network traffic between end-users and application-hosting nodes while keeping infrastructure costs low and meeting QoS requirements of end-users, such as latency and/or throughput. One way to address the challenges presented by emerging IoT applications is to move the computations closer to end-users – that is, towards the ISP’s edge network – to reduce transmission costs, decrease network latency and jitter, increase reliability, and avoid network congestion. A key idea is to create a unified ICT infrastructure using existing large-scale distributed cloud infrastructures and augmenting (Li, 2016) them with compute capacities at intermediary nodes, such as radio base stations at the ISP’s edge network and inside its internal network. As shown in Figure 1, the unified infrastructure at the edge of the network, called “Mobile Edge Clouds,” can host applications closer to the end users. This helps solve congestion problems and meets end users' performance expectations.