Fog Computing

1 Latency constraints. The fog shares the same fundamental characteristics as the cloud in being able to perform different computational tasks closer to the end-user, making it ideal for latency-sensitive applications for which their requirements are too stringent for deployment in the cloud.

2 Network bandwidth constraints. Since the fog offers the possibility of performing data processing tasks closer to the edge of the network, lowering in the process the amount of raw data sent to the cloud, it is the perfect device to apply data analytics to obtain fast responses and send to the cloud for storage purposes only filtered data.

3 Resource constrained devices. Fog computing can perform computational tasks for constrained edge devices like smartphones and sensors. By offloading parts of the application from such constrained devices to nearby fog nodes, the energy consumption and life-cycle cost are decreased.

4 Increased availability. Another important aspect of fog computing represents the possibility of operating autonomously without reliable network connectivity to the cloud, increasing the availability of an application.

5 Better security and privacy. A fog device can process the private data locally without sending it to the cloud for further processing, ensuring better privacy and offering total control of collected data to the user. Furthermore, such devices can increase security as well, being able to perform a wide range of security functions, manage and update the security credential of constrained devices and monitor the security status of nearby devices.

The previous section introduces the edge computing paradigm as a solution to the cloud computing inefficiency for data processing when the data is produced and consumed at the edge of the network. Fog computing focuses more on the infrastructure side by providing more powerful fog devices (i.e. fog node may be high computation device, access points, etc.) while edge computing focuses more toward the “things” side (i.e. smartphone, smartwatch, gateway, etc.). The key difference between the edge computing and fog computing is where the computation is placed. While fog computing pushes processing into the lowest level of the network, edge computing pushes computation into devices, such as smartphones or devices with computation capabilities.

Fog computing architecture is composed of highly dispersed heterogeneous devices with the intent of enabling deployment of IoT applications that require storage, computation, and networking resources distributed at different geographical locations [21]. Multiple high-level fog architectures have been proposed in the literature [22–24] that describe a three-layer architecture containing (1) the smart devices and sensor layer which collects data and send it forward to layer two for further processing, (2) the fog layer applies computational resources to analyze the received data and prepares it for the cloud, and (3) the cloud layer, which performs high intensive analysis tasks.

Bonomi et al. [10] present a fog software architecture (see Figure 2.4) consisting of the following key objectives:

Heterogeneous physical resources. Fog nodes are heterogeneous devices deployed on different components, such as edge routers, access points, and high-end servers. Each component has a different set of characteristics (i.e. RAM and storage) that enables a new set of functionalities. This platform can run on multiple OSes and software applications, deriving a wide range of hardware and software capabilities.

Fog abstraction layer. The fog abstraction layer consists of multiple generic application programming interfaces (APIs) enabling monitoring and controlling available physical resources like CPU, memory, energy, and network. This layer has the role of making accessible the uniform and programmable interface for seamless resource management and control. Furthermore, using generic APIs, it supports virtualization by monitoring and managing multiple hypervisors and OSes on a single machine with the purpose of improving resource utilization. Using virtualization enables the possibility of having multitenancy by supporting security, privacy, and isolation policies to ensure the isolation of different tenants on the same machine.

Fog service orchestration layer. The fog service orchestration layer has a distributed functionality and provides dynamic and policy-based management of fog services. This layer has to manage a diverse number of fog nodes capabilities; thus, a set of new technologies and components are introduced to aid this process. One of these components is a software agent called foglet capable of performing the orchestration functionality by analyzing the deployed services on the current fog node and its physical health. Other components are a distributed database that stores policies and resource metadata, a scalable communication bus to send control messages for resource management, and a distributed policy engine that has a single global view and can perform local changes on each fog node.

Figure 2.4 Fog computing architecture [10]. ( See color plate section for the color representation of this figure )

In this section, we present two illustrative use cases for both the fog and edge paradigms with the purpose of showing key features, helping us to further understand the concept and applicability in real-world applications.

The rapid growth of big data frameworks and applications such as smart cities, smart vehicles, healthcare, and manufacturing has pushed edge computing among the major topics in academia and industry. With the increasing demand for high availability in such systems, system requirements tend to increase over time. The stringent requirements of IoT systems have recently suggested the architectural placement of a computing entity closer to the network edge. This architectural shifting has many benefits, such as process optimization and interaction speed. For example, if a wearable ECG sensor were to use the cloud instead of the edge it will consistently send all data up to the centralized cloud. As a consequence, in such a scenario it will cause high communication latency and unreliable availability between the sensor and the centralized cloud. In real-time, safety-critical IoT use-cases, devices must comply to stringent constraints to avoid any fatal events. In this scenario, the latency delay introduced by sending the data to the cloud and back is inadmissible. Thus, in case of a critical event detected by the sensor, a local decision must be taken by the edge device, rather than sending data to the cloud.

Edge computing is well suited for IoT deployments where both storing and processing data can be leveraged locally. For example, consider a smart home where the sensory information is stored on the edge device. Simply by doing encryption and storing sensory information locally, edge computing shifts many security concerns from the centralized cloud to its edge devices. In addition, IoT applications consume less bandwidth, and they work even when the connection to the cloud is affected. Furthermore, edge devices may assist in scheduling the operational time of each appliance, minimizing the electricity cost in a smart home [25]. This strategy considers user preferences on each appliance determined from appliance-level energy consumption. All such examples can benefit from the edge computing paradigm and to demonstrate the role of this paradigm in different scenarios, we describe in this section two possible use cases in healthcare [12] and a smart home [3, 15].