I. Introduction

Several periods have been written into the introduction of computing to the society. Although the authors tend to differ between the exact temporary bounds, there is a general consensus in the main stages in the connectivity models and services when technological challenges have arisen. Three stages have been identified:

• The first one, in the 60s until the early 80s, when the mainframe and the dumb terminals with green screens allowed to execute the batch processing in banks, universities and others;

• the second stage, in the 80s, when a new client-server stage started. This one –together with the revolution and popularization of the web– allowed the duration of this stage until the end of the 20th century and it entailed new technological developments relative to the computing technologies and data storage; and

• the third one, thanks to the popularization of mobile devices and cloud computing. This occasioned the industrialization of the computing and data storage businesses, such as the development environments and consumption of end-services from the web –e.g., the email– (De Fuenmayor,2017).

The reader might notice that, at the end-user level and its respective needs, an adequate balance between the services demand that the internet provides is present. Even though the internet is not a service with world-wide coverage, there are several application fields from pure science to the solely entertainment, which includes large amounts of people in countries where there is a generalized access to the service.

The bandwidth needs in many countries are covered by the average user (see Figure 1). As per a study performed by Akamai Technologies (2017), the speeds per subscriber are up to 7.2 Mbps, with an annual growing of 15% in the first quarter of 2017, compared with the same time in 2016 (De Fuenmayor, 2017).

Figure 1
Global internet speed (“Mapped...”, 2017)

This leads us to the following questions:

• Does the cloud computing comply with all the emerging solution needs, not only the connectivity demands of the end-user devices (smartphones, tablets, PCs), but also devices that can transmit information of their automation objects in an autonomous way?

• Which are the alternatives to the information processing need emitted by these devices for an opportune, effective, and efficient decision-making process?

II. The New Actor: Internet of Things

Cisco Systems (2017) coins the IoT [Internet of Things] as the network of objects –i.e., electronic “stuff”– formed by programs, sensors, and connectivity used to generate value and service towards data exchange with the manufacturer, the operator, or another connected devices through advanced communication protocols without human intervention (Abdelshkour, 2015). This environment has been developed by pairing the ICT [Information and Communication Technologies] development and the technological infrastructure of the society.

IoT grants instant access to a device by guaranteeing its correct installation and behavior, the analysis of the obtained data, the interoperable connection with its corresponding LAN [Local Area Network], cloud, or other connected devices by any medium in a wired or wireless network as it is technically suitable (Bojanova, 2015).

Until now, the followed model for the gathering and analysis of the information picked from IoT is based on the cloud computing paradigm. Nonetheless, this model is no more than an instance of a traditional and centralized architecture with large (and limited) datacenters with first-level connectivity. This model is not scalable nor suitable for the challenges the IoT entails in the short and medium term, as the Arco Research Group (2017) indicates: “the transmission of the amount of data generated by IoT devices to a single location to be processed will not be technically nor economically viable”.

Despite the advantages of the cloud computing supporting and leveraging the development of the interaction between the ICT and the society in all its aspects, it is true that the number of autonomous devices with computational logic –either with low or high complexity– has invaded the life of the citizens and the organizations.

The approach from the engineering point of view has found some limitations to the growing in the connectivity demand, such as the latency and bandwidth.

• Latency: the network distance between the device and the processing central is about hundreds of milliseconds (round-trip). Many of the IoT applications will be latency-sensible such as the predicting systems (energy consumption and smart grids) and the voice controlled systems (Siri, Google Home, or Alexa).

• Bandwidth: a larger information contribution coming from devices processing audio and video is predicted, especially in cities. Surveillance cameras in public/private locations, traffic monitoring, and personal phones are some examples. The information volume to process in real-time and send it to the cloud would result in inadmissible response times, if the technological capacity required for that is on its place (Abdelshkour, 2015).

One of the options to face these realities is the so-called fog computing (Abdelshkour, 2015). The term –also coined by Cisco– establishes a “fog” model formed by a distributed collection and data analysis points generated by ads, entertainment, computing, and other aspects featuring the informatic applications housed in any network-enabled device (Bonomi, Milito, Zhu, & Addepalli, 2012).

Within the fog computing model, the available computing resources can be used in the end devices or in nodes not being employed for general purposes. Alternatively, some computing resources can be added to the existing nodes or to an edged network to ease the processing near the end user device.

Fog computing has a group of associated advantages. The first one is the reduction in the network traffic by providing a platform for the filtering and analysis of the data generated by the sensors. This is done by using resources of the devices allocated in the edge, which drastically reduces the traffic sent to the cloud. Another advantage is the reduction in the latency, especially for applications requiring real-time processing.

Hu, Dhelim, Ning, & Qiu (2017) present another advantages of the fog relative to the location sensitiveness, the geographic distribution, the low energy consumption, and the security and protection of the data privacy. The fog computing supports the mobility demands based on the location and enables the administrators the power to control users’ location and the information access from mobile devices.

The decentralized architecture of the fog computing ensures the proximity to the client in the data analysis, which allows to perform faster big data analysis, to deploy better services based on the location, and to execute more powerful capacities for the real-time decision-making processes. Due to the deployment of services near the end user, this model presents considerable advantages in relation to the data security and the protection of the user privacy.

As the main disadvantage, it has been informed that this model might have a reduced tuning fork of technological platforms for development and construction, such as its presentation when the interoperability standards are established to reach the majority of devices (Abdelshkour, 2015). This topic really impacts the advance that the standardization of solutions and platforms have had within the computing cloud. Cisco shows a summary of the aspects to be considered when fog and cloud are put in a weighting scale (see Table 1). On the other hand, the role of the servers and the information processing are shown in Table 2.

Table 1.
Fog vs. cloud (Abdelshkour,2015)

Table 2.
Features of the fog solutions compared to the cloud ones, expected results and behaviors (Abdelshkour, 2015)

III. Architecture and Quality of Service in Fog Computing

Understand on how to improve the QoS [Quality of Service] in the IoT networks is becoming a challenge. The fog computing approach is based on getting the end users as close as possible to the cloud to improve the general performance without being limited to a particular architecture (such as the cellular networks). The fog computing is an intelligent layer located between the cloud and the IoT, which provides low latency, location knowledge, and generalized geographic distribution for the IoT (see Figure 2).

Figure 2
General frame for the IoT-fog-cloud architecture (Yousefpour, Ishigaki, & Jue, 2017)

The basic way the IoT, fog, and cloud nodes operate and interact between them is exposed in the next paragraphs. The IoT nodes process their requests locally, sending it after to a fog or cloud node. After that, the fog nodes can process the received requests; if they cannot, they forward the task to another fog node in the same domain. If these latter are unable to process the requests, it is sent to the cloud, where the nodes there will process the requests and send the responses to the IoT nodes (Yousefpour et al., 2017).

All the steps entail delays in the service, which is no more than the required time to attend a request; that is, the time interval between the time a IoT node sends a service request and the moment when it receives response from that request.

The decision to perform a task is based on the response time of a fog node, which depends of several factors: the amount of computing power required for the task, the state of the queue, and the processing capacity of a fog node. It is important to consider the different processing times according to the various tasks. In other words, the heavy and light processing tasks should be distinguished (Chin-Feng, Dong-Yu, Ren-Hung, & Yin-Xun, 2016). For instance, the requests sent by temperature sensors towards fog nodes to calculate the average room temperature can be seen as light processing tasks; whilst a request sent from a video camera to read the license plate of a vehicle to the fog nodes is an examples of a heavy processing task.

When the fog nodes receive requests from IoT nodes participating in an application, they must distinguish between light and heavy requests. For that, the requests must have a field identifying the request type on their header; e.g., the traffic class field in the IPv6 header.

There are two interaction modes for fog nodes: a centralized mode, where a central authority controls the interaction between the fog nodes; and a distributed one, where the fog nodes interact with their neighbors using a universal protocol (Cardellini, Grassi, Lo Presti, & Nardelli, 2015).

The centralized interaction mode can be seen as a central resource orchestration, where the central node –which knows the topology and the state of the fog nodes in a domain– is present. The centralized mode is easier to implement, since there is no need of a distributed communication through certain protocols. Furthermore, the central node might be used as a medium to push the fog applications and perform updates to the nodes within its domain.

The distributed mode is the more suitable for scenarios where the fog nodes are not static, or when the fog network is designed in an ad hoc manner. Besides, in this mode, there is no need of having a dedicated node to act as a central one; which represents a reduction in the implementation cost, but a larger failure vulnerability by having a single failing point.

The delays in the propagation are given to the fog nodes as input parameters. As per the interaction mode, when an estimated sample of the waiting time is received –either from the central node or from another one–, the fog node updates the corresponding waiting time in the accessibility table.

The selection of the distributed mode in the fog nodes –as a strategy to provide QoS– is the most efficient for a dynamic scenario entailing billions of connected devices. The strategy is mainly based on adjusting the number of controllers depending on the amount of resources (Cardellini et al., 2015).

In Figure 3, the blue arrow tagged with number 1 illustrates de deployment of fog controllers when the controller layer is down (CL1 in yellow) to correctly handle the high control delay produced by the fog controller. In addition, when the amount of the border resources grow, the new controllers must be deployed in CL1 to avoid that overwrittes in the existing ones, as the arrow tagged with number 2 shows. Nevertheless, the collateral efect is that, when the higher the amount of controllers in CL1 is, the higher the traffic due to the communication between controllers will be, causing a larger average in the delays. To address this, it is possible to implement a new controler layer (CL2) as the arrow tagger with number 3 presents: a new layer with a controller per area to avoid high latency in the communication with the cloud controller. For example, even when the amount of controllers in CL1 is relatively low, the implementation of CL2 –containing 1 single controller per area– can reduce the average communication between areas.

Figure 3
Evolution of the fog nodes (Chin-Feng et al., 2016)

IV. Fog Computing Applications in Smart Cities

The smart and sustainable cities paradigm has gained relevance to face the challenges of the modern urbanization. The ICT use allows higher efficiency in the operations and urban services, improves the life quality of the citizens, and encourages the environmental sustainability.

A smart and sustainable city is an innovative one using the ICT and other means to improve life quality, efficiency in the services and urban operation, and competitiveness. This, by ensuring the satisfaction of needs of the present and future generations relative to economic, social, environmental, and cultural aspects (ITU, 2017).

In the scientific literature, several applications and use cases of fog computing for smart cities are reported. In Bangalore, as part of a pilot project, some surveillance cameras, environmental sensors, edge computing platforms (e.g., Raspberry Pi), and fog computing platforms (e.g., NVIDIA Jetson TX1) have been installed across the city (Varshney & Simmhan, 2017). The use of fog computing for urban monitoring through surveillance cameras is one of the scenarios where better results have been achieved, by reducing the latency when the sharing of processing loads with the cloud is performed. Further, this offers more efficient real-time (or nearly real-time) responses (Gupta et al, 2017). Mohamed, Al-Jaroodi, Jawhar, Lazarova-Molnar, & Mahmoud (2017) describe another fog computing applications in intelligent transportation systems, intelligent energy, intelligent water, and environmental monitoring.

These use cases have presented good practices and learned lessons about the manner to implement fog computing in intelligent cities, and they were analyzed to propose a roadmap seeking to face the implementation of the fog computing within a project starting to be developed in Havana (Cuba). In this project, the experimentation of technological prototypes seeking to transform the city towards a sustainable and intelligent city is proposed.

V. A Roadmap for Mobile Applications in Smart Cities Based on Fog Computing

The Cuban Union of Informatics [UIC, Unión de Informáticos de Cuba] is starting an experimentation project of smart and sustainable cities in Havana, where it should start with the creation of an urban laboratory in old Havana. This project –approved by the National Computation Program of Science, Technology, and Innovation for the 2018-2021 period– includes in its objectives to develop a line of intelligent transportation systems aiming to offer different information types (and recommendations) to the people in a determined zone, establishing a traveler information system using LED displays, tactile screens, or mobile devices.

There are experiences that work as precedent, such as a spatial recommendation system sensible to the traveler context, which recommended sightseeing places through mobile devices with GPS and by using the maps in the Spatial Infrastructure Data of the Cuban Republic (González, Delgado, Capote, & Cruz, 2013) are presented to the end user. The GPS signal and other data in a mobile device and from the vehicles must be transmitted to a central control point in the cloud and return with the recommendations to the traveler in almost real-time. In spite of the fact that, at the prototype level, the system worked with an acceptable performance, in general terms, the reduction of the latency and to take advantage of the geographically distributed configuration of the fog computing systems in real smart cities applications is the desired scenario.

In order to build a new version of the sightseeing recommendation system in the city and using the fog computing context, a general roadmap has been proposed. This roadmap includes the following four general steps (research tasks):

1. Define a fog computing architecture for information systems to the travelers in smart cities. In this step, it is necessary to discover the resources availability capable to act as controller, to configure the discovered resources, and to create a controller topology by establishing a hierarchical relation amongst them.

2. Develop the necessary mechanisms to deploy the services in the fog layer, which can be adapted to mobile devices. This step also includes to consider existing standards and open platforms (e.g., Docker) and extend them to include the necessary elements, such as efficient virtualization technologies applicable to the majority of physical nodes as possible.

3. Develop fog management systems, such as QoS, QoE [Quality of Experience], service continuity, network efficiency, and load balance. All of them according to the defined architecture of the traveler recommendation system using fog computing (step 1). This, pursuing an optimum strategy to distribute fog service actions relative to the cloud processing tasks.

4. Integrate all the components of the recommendation system and test its performance in real scenarios of old Havana.

VI. Conclusions and Future Work

Fog computing complements the cloud paradigm seeking to reduce the latency and the unnecessary internet data traffic using geographically distributed resources closer to the source (sensors). From this new paradigm, a countless number of new applications towards all the society dimensions can emerge.

The new network architectures must use the cloud and fog resources to improve the QoS in the data transmission and reduce the processing delays. For this, the impact of the decisions the fog nodes take relative to the QoS in real time for the NGN [Next Generation Networks] is relevant. This QoS control can be done through the control of the topologies and considering crucial parameters in hierarchical architectures such as the number of layers or the ability to manage each controller.

New challenges will come together with fog computing, however, its low latency and its location at the edge of local networks –with less bandwidth demand–, considering that only what cannot be solved in the fog is sent to the cloud, make it particularly interesting to implement the Internet of Things in an incremental way and learning from the experience. The described roadmap to develop recommendation systems sensible to the context and compatible with mobile environments for smart cities based on fog computing offer a starting point to experiment its advantages and limitations in a real scenario.

References

Abu Abdelshkour, M. (2015). IoT, from cloud to fog computing. Retrieved from: https://blogs.cisco.com/perspectives/iot-from-cloud-to-fog-computing

Akamai Technologies. (2017). Q1 2017 State of the Internet / connectivity report. Retrieved from: https://www.akamai.com/us/en/multimedia/documents/state-of-the-internet/q1-2017-state-of-the-internet-connectivity-report.pdf

Arco Research Group (2017, March 12). El papel de la tecnología de lógica reconfigurable como respuesta a los retos del Internet de las Cosas. Retrieved from: https://arcoresearchgroup.wordpress.com/2017/03/12/el-papel-de-la-tecnologia-de-logica-reconfigurable-como-respuesta-a-los-retos-del-internet-de-las-cosas/

Bojanova, I. (2015). What makes up the Internet of Things? Retrieved from: https://www.computer.org/web/sensing-iot/content?g=53926943&type=article&urlTitle=what-are-the-components-of-iot-

Bonomi,F., Milito, R., Zhu, J., & Addepalli, S. (2012). Fog computing and its role in the Internet of Things. In: Proceedings of the first edition of the MCC Workshop on Mobile Cloud Computing, (pp. 13-16). New York, NY: ACM.

Cardellini, V., Grassi, V., Lo Presti, F., Nardelli, M. (2015). On QoS-aware Scheduling of Data Stream Applications over Fog Computing Infrastructures. In: 2015 IEEE Symposium on Computers and Communication (ISCC), (pp. 271-276). doi:10.1109/ISCC.2015.7405527

Chin-Feng, L., Dong-Yu, S., Ren-Hung, H., Ying-Xun, L. (2016). A QoS-aware streaming service over fog computing infrastructures. In 2016 Digital Media Industry & Academic Forum (DMIAF). IEEE. doi:10.1109/DMIAF.2016.7574909

Cisco Systems. (2017). Fog computing and the Internet of things: Extend the cloud to where the things are [white paper]. Retrieved from: https://www.cisco.com/c/dam/en_us/solutions/trends/iot/docs/computing-overview.pdf

De Fuenmayor, A. (2017, Feb. 20). Cloud computing, ¿el final de una era? [blog ThinkingBig]. Retrieved from: https://aunclicdelastic.blogthinkbig.com/cloud-computing-iot/

Gonzalez, G., Delgado, T., Capote, J. L., & Cruz, R. (2013). Context-aware recommender system based on ontologies. In: H. Onsrud, & A. Rajabifard (Eds.), Spatially enablement in support of economic development and poverty reduction, (pp. 227-243). Reston, VA: GSDI.

Gupta, H., Vahid-Dastjerdi, A., Ghosh, S.K., Buyya, R. (2017). FogSim: A toolkit for modeling and simulation of resource management techniques in internet of things, edge and fog computing environments. Software: Practice and Experience, 47(9), 1275-1296.

Hu, P., Dhelim, S., Ning, H. Qiu, T. (2017). Survey on fog computing: Architecture, key technologies, applications and open issues. Journal of Network and Computer Applications, 98, 27-42.

International Telecommunications Union [ITU]. (25/ 4/2017). The Manizales Manifesto. Retrieved from: https://www.itu.int/en/ITU-T/Workshops-and-Seminars/gsw/201704/Documents/Manifesto-Manizales-05-04-2017-Eng-Final.pdf

Mapped: The world according to Internet connection speeds - Telegraph (2017, April, 9). Retrieved from: http://www.telegraph.co.uk/travel/maps-and-graphics/countries-with-fastest-internet-connection-speeds/

Mohamed, N., Al-Jaroodi, J., Jawhar, I., Lazarova-Molnar, S., & Mahmoud, S. (2017). Smartcityware: A service-oriented middleware for cloud and fog enabled smart city services. IEEE Acces, 5, 17576-17588. doi:r 10.1109/ACCESS.2017.2731382

Varshney, P., & Simmhan, Y. (2017). Demystifying fog computing: Characterizing architectures, applications and abstractions. In Fog and Edge Computing (ICFEC), 2017 IEEE 1st International Conference on (pp. 115-124). IEEE.

Yousefpour, A., Ishigaki, G., & Jue, J. P. (2017). Fog computing: Towards minimizing delay in the internet of things. In Edge Computing (EDGE), 2017 IEEE International Conference on (pp. 17-24). IEEE.

Author notes