• Tidak ada hasil yang ditemukan

Fog Computing

Dalam dokumen Smart Cities (Halaman 175-180)

Cloud computing is attractive to both industrial users and academic researchers for several reasons: with cloud computing, capital outlay becomes unnecessary for businesses and maintenance cost can be significantly lowered down since hardware or software resources are provided by cloud instead of the local resource deployment. Cloud centers provide excellent data storage and computing capabil-ities. However, the round-trip latency resulted from data transmission to and from cloud centers is a hindrance for delay-sensitive applications such as smart city surveillance. The IoT connects ubiquitously deployed smart devices at the network edge together, such that services and applications are enabled locally at the edge of the network. Under this context, the cloud paradigm is not the universal solution in the IoT era anymore.

The basic concept of fog computing is not new. It was studied under the umbrella of cloudlets, mobile cloud, etc. But it has attracted more and more attention as the IoT become pervasive. As the extension and complement of cloud paradigm, the fog computing supports data storage and processing at the network edge. Unlike cloud that is high in the sky, fog stays close to the ground. In this section, fog computing-related concepts, applications and benefits are presented.

7.3.1 Concepts and Architecture

Fog computing is a distributed computation paradigm that leverages the huge number of heterogeneous devices deployed at the edge of network, which are connected and collaborate with each other by sharing computation, storage and communication functionalities.

With this definition, another similar terminology is edge computing. In fact, some researchers do not consider them as two different technologies [15], although some people emphasizes that the fog computing focuses more on infrastructure while the edge computing focuses more on the deployed smart devices. In this chapter, we regard these two concepts as the same and use them interchangeably.

Due to the unprecedented increasing amount of smart devices at the network edge as well as their computational capabilities, the smart devices play multiple roles, they function as data collectors, surplus data storage, and data processing units. Accordingly, the fog computing paradigm is considered as a computing platform that brings computation and storage from centralized cloud centers to the work site at the network edges, which inspires innovative services to end users and improves the quality of user experiences especially for the latency-sensitive applications.

Figure7.2shows the basic architecture of a fog computing implementation. At the bottom layer are end users within a working area. The fog layer is inserted between the end users and the cloud centers and consists of a variety of computing

nodes. Comparing with the cloud scheme, the fog computing layer stays closely with end users and utilizes a group of smart devices at the network edge as com-puting nodes. Personal laptops, smartphones, cloudlets, or even smart routers can potentially be functioning as fog nodes. While smart devices produce urban data, they are capable of storing and processing data as well. Therefore, with the fog computing paradigm, urban data can be computed on-site, which reduces the bandwidth usage since only globally significant data will be transferred to data centers. The cloud platform is still necessary in the architecture. After the pro-cessing done at the fog layer, important metadata can be sent to the cloud center for archival purpose or further analysis for long-term traffic pattern analytics.

The definition and the illustration of computing architecture of the fog com-puting above have shown that comparing with the cloud, the fog comcom-puting based scheme is a better match to smart city surveillance tasks in terms of instant decision-making and the reduction of load on communication networks. However, another important feature of the fog computing paradigm lies in the high hetero-geneity in the computing resources at the network edge. Different from powerful cloud data center, not only each individual node possesses weaker computing capability, the management of the entire fog cell is also more complex because of dynamic and heterogeneous environments.

Table7.1illustrates a more detailed comparison between the cloud and the fog scheme. With these unique characteristics, the fog computing paradigm is Fig. 7.2 Fog computing paradigm architecture

considered more suitable for the applications with the requirement of low latency and real-time response. Furthermore, due to different computing capabilities and targeting tasks, the fog and cloud schemes are mutually complementary with each other.

7.3.2 Applications and Ongoing Efforts

The fog computing paradigm facilitates the delay-sensitive applications with min-imum response latency and instant decision-making capabilities, which are essential in urban scenarios especially for the emergent situations. A number of research works have discussed potential smart city applications that can leverage the fog computing platform.

• Connected Vehicles [16]. The concept of connected vehicles proposes an ideal platform for improving user experience in vehicles and traffic surveillance. On this platform, not only vehicles are connected, but also vehicles and roadside units communicate with each other. With its inherent advantages, the fog is ideal Table 7.1 Comparisons between fog and cloud computing paradigms

Attribute Fog computing Cloud computing

Ubiquity Fog nodes present higher availability since an enormous amount of smart devices at the network edge have the potentials to be adopted as computing nodes

Cloud centers are located remotely from the network edge

Latency Fog presents low latency since computing resource are close to users

Higher latency comparing with fog scheme due to the round-trip communication time Heterogeneity Fog nodes are heterogeneous

because of the utilization of various types of smart devices at the network edge

Normally cloud resources are deployed according to the requirements from users and the computing nodes are deployed of similarity and are provided as clusters

Computing capability

Fog layer mainly consists of smart devices at the network edge which are regarded as of normal computational capability

Cloud centers present powerful computational capabilities

Dynamicity Fog computing resources are dynamic

Cloud computing resources stay within cloud data centers Bandwidth

consumption

Fog nodes could be connected using local area network which will reduce the Internet bandwidth consumption

Transmitting all the urban data to cloud centers by Internet and other communication networks present high bandwidth usage

for connected vehicles by providing real-time applications like safety informa-tion or entertainment content delivery.

• Mobile Big Data Analytics [17]. At the edge of network, tons of urban data are produced every second. Without suffering the communication delay, Fog plat-form provides on-site big urban data processing services. Because not all urban data are globally important, applying fog computing can efficiently reduce the bandwidth consumption.

• Smart Grid [18]. Fog computing provides an environment in which data col-lected from smart meters in smart grid can be preprocessed. Furthermore, fog nodes act as a bridge between raw data and the data centers. Smart grid can achieve better performance with the facilitation of the fog paradigm.

• eHealth and Smart Home [19]. In smart homes, residence data will be collected by devices embedded in the environment, i.e., wearable digital sensors. Data are processed locally in the fog layer to provide personalized services in smart home applications. eHealth is one kind of services in smart homes. Patients at home wear digital sensors or other wearable devices for health condition monitoring. Health data will be preprocessedfirst by personal services and then decision can be made whether the data should be sent to remote hospitals.

Utilizing Fog platform, personal health monitoring service at home is seamlessly connected with remote service providers.

There are more potential applications that the fog computing will improve the performance, but only several representative applications are listed here due to the limited space. What must be emphasized here is, although fog presents some unique advantages in a lot of scenarios, it does not replace cloud computing totally.

Instead, fog and cloud are complementary with each other for better improvement.

Given the inherent benefits and potential applications in the fog computing para-digm, a variety of research has been explored:

• A fog computing solution is proposed for advanced metering infrastructure in smart grid [20]. A user-friendly interface is developed and the proposed model is verified on a proof-of-concept testbed.

• A virtualized and decentralized fog-based emergency architecture was proposed to improve the smart living environments [21]. In this scheme, resources in the pool are managed by a dynamic resource service.

• The fog computing concept has been leveraged to improve the in-vehicle data processing [22], which achieves the goal of real-time processing by building virtual machines within the vehicles to facilitate on-site data processing tasks.

• The fog computing scheme has also been utilized for face identification and resolution in IoTs [23]. A prototype is proved to be effective through experi-mental studies and it shows significant reduction of computing overhead.

Besides the efforts focused on the fog-enabled applications at the network edge, there are also research conducted in terms of the resource provisioning and archi-tecture of fog computing. For example, a conceptual system to provision the computing resources of fog has been suggested, which treats the resource

provisioning as an optimization problem to optimally reduce the application delay [24]. Some researchers consider the fog layer as a suitable platform to improve the data delivery service and they use Stackelberg game theory to build a hierarchical data center architecture consisting of both the fog and cloud schemes [25]. In addition, there is a fog-enabled orchestration scheme was introduced for on-site applications and services [26], which is still in conceptual proof phase with some early experimental results reported.

7.3.3 Open Issues and Discussion

In spite of the unique advantages of the fog computing paradigm, there still exist a number of open issues waiting for further explorations by researchers. The major research challenges in the fog computing are listed as follows [17,27]:

• Resource provisioning and management: the fog nodes consist of a large number of heterogeneous smart devices. Not only the demands from different applications are variants, but also the computational resources provided by fog nodes are changing constantly. To obtain a satisfied quality of service, latency, dynamic and mobility must be taken into consideration.

• Computation capability limitation: although in the fog computing paradigm, fog nodes are ubiquitously available because of the increasing number of smart devices, the computational capabilities of fog nodes are not as powerful as cloud centers. How to ensure the quality of service is a relatively big concern.

Additionally, the variant computational capabilities make synchronization an issue as one task may be divided into multiple jobs and each of them is sent to one fog node for processing.

• Interface and programmability: like other commercial service providers, the complexity and technical details are expected to be transparent to end users. In fog computing paradigm, the challenging questions such as how to discover computation resources and how to configure the fog nodes and other technical details should not confuse the users. An easy-to-use interface should be pro-vided to ensure that users can migrate their works to fog nodes simply and obtain the outcome feedback timely.

• Standardization: there is still not a uniform standardization for fog computing.

Cloudlets, close-user servers, and personal smart devices are all well-qualified candidates to be considered as fog nodes. Therefore, a uniform standardization is very critical to address the issues like how to bill fog computing users, how to attract people to share their smart device resources as fog nodes or how to configure the fog services.

• Privacy and Security: as mentioned above, personal smart devices can be uti-lized as fog nodes. How to ensure the privacy and security of personal infor-mation becomes a big challenge. In a fog, computational resources may be shared with third parties. In addition, on the user side, how to ensure the privacy

and security of software or application results is also a concern as the task is outsourced to fog nodes.

As discussed in Sect.7.2, there are a lot of challenges in smart city surveillance, which aims at identifying anomalies occurring in normal patterns. On one hand, in the big data era, it is not practical to monitor cities all the time merely by human operators. Furthermore, monitoring tasks are extremely tedious because anomaly events occur with low probability and human operators can be unaware of changes under the flushing of big urban data. On the other hand, smart city surveillance tasks will present restrict requirements about processing time and decision-making delay since earlier anomalies are identified, greater opportunities there will be to prevent damages or disastrous consequences. Hence, conventional monitoring by human beings and resorting to cloud techniques cannot meet the challenges posed by smart city surveillance tasks.

Because of its unique advantages, the fog computing has been recently identified as an ideal platform for smart city surveillance. Utilizing fog paradigm, big urban data can be processed at where they are produced, which in turn significantly reduces the response time. Moreover, fog and cloud platforms cooperate with each other to form an integrated computing architecture in which fog can preprocess the collected data to provide services as well as discard redundant data.

Dalam dokumen Smart Cities (Halaman 175-180)