PaperlayerView exact revision on arXiv

Source-linked AI summary

Fog Computing for Sustainable Smart Cities: A Survey

Charith Perera, Yongrui Qin, Julio C. Estrella, Stephan Reiff-Marganiec, Athanasios V. Vasilakos

arXiv:1703.07079v1cs.NI

TL;DR

Centralized cloud processing of IoT data can waste bandwidth, storage, latency, and communication energy, while edge devices have limited computational and energy resources. The survey reviews fog-computing use cases and research, identifies platform functionalities and challenges, and concludes that cloud and fog capabilities must work together for sustainable smart-city IoT.

  • Problem

    Sending all collected IoT data to the cloud creates bandwidth, storage, latency, and communication-energy costs, while edge devices have limited computational capabilities.

  • Method

    The paper reviews fog-computing research, use cases, platform characteristics, connectivity and device-configuration features, and open challenges.

  • Results

    The survey identifies ten fog-computing characteristics, compares more than 30 research efforts, and identifies major platform functionalities and open challenges.

  • Takeaways & Limitations

    Cloud and fog computing need to work together because their inherited strengths and weaknesses do not address sustainable smart-city IoT challenges alone.

  • Takeaways & Limitations

    Fog devices have limited processing, storage, memory, energy, and global knowledge, constraining which analyses they can perform locally.

Abstract

from arXiv · show

The Internet of Things (IoT) aims to connect billions of smart objects to the Internet, which can bring a promising future to smart cities. These objects are expected to generate large amounts of data and send the data to the cloud for further processing, specially for knowledge discovery, in order that appropriate actions can be taken. However, in reality sensing all possible data items captured by a smart object and then sending the complete captured data to the cloud is less useful. Further, such an approach would also lead to resource wastage (e.g. network, storage, etc.). The Fog (Edge) computing paradigm has been proposed to counterpart the weakness by pushing processes of knowledge discovery using data analytics to the edges. However, edge devices have limited computational capabilities. Due to inherited strengths and weaknesses, neither Cloud computing nor Fog computing paradigm addresses these challenges alone. Therefore, both paradigms need to work together in order to build an sustainable IoT infrastructure for smart cities. In this paper, we review existing approaches that have been proposed to tackle the challenges in the Fog computing domain. Specifically, we describe several inspiring use case scenarios of Fog computing, identify ten key characteristics and common features of Fog computing, and compare more than 30 existing research efforts in this domain. Based on our review, we further identify several major functionalities that ideal Fog computing platforms should support and a number of open challenges towards implementing them, so as to shed light on future research directions on realizing Fog computing for building sustainable smart cities.

1. INTRODUCTION

IoT and big-data growth make centralized cloud processing costly and resource-intensive, motivating fog computing and a combined cloud–fog infrastructure for sustainable smart cities. The survey reviews use cases, platform characteristics, prior research, and open challenges.

  • IoT connects diverse objects that increasingly embed sensors, generating large volumes of data for smart-city applications.
  • Sending all collected data to the cloud imposes bandwidth, storage, latency, and communication-energy costs.
  • Fog computing shifts data processing and knowledge discovery toward network edges instead of sending everything to the cloud.
  • The survey aims to assess fog computing for sustainable smart cities, identify essential platform functionalities, and highlight open challenges and research directions.
  • The paper combines use-case scenarios, a literature-based taxonomy, comparisons of more than 30 research efforts, and analysis of research trends and gaps.

2. FOG COMPUTING: AN OVERVIEW

Fog computing moves storage, communication, control, and analytics toward near-user edge devices, reducing dependence on cloud backbones while retaining realistic limits on edge resources. Its benefits include lower latency, local context, availability, and potential sustainability savings, but edge capability, energy, cost, and global-knowledge constraints remain.

  • Fog computing is an architecture that uses near-user edge devices for substantial storage, communication, control, configuration, measurement, and management.
  • Fog computing pushes analytics toward leaf nodes without requiring all processing to occur at the edge.
  • Fog designs store and process more data locally, use local networks, and let edge devices operate with greater autonomy than cloud-centered systems.
  • Fog devices have lower computational capabilities and cost than cloud infrastructure, although capabilities are expected to increase over time while category differences remain.
  • Edge processing can reduce latency, improve availability during cloud disconnection, support local context, and lower communication-related security exposure.
  • Fog computing can cost more initially than dumb edge devices, while potential long-term savings may justify the additional cost.
  • Edge devices handle less data and processing, may lack global knowledge, and may be unsuitable for energy-intensive analytics.

3. USE CASE SCENARIOS

The paper presents four smart-city use cases—agriculture, transportation, healthcare, and waste management—to derive fog-computing requirements. These scenarios use context-aware sensing, fog devices, and selective or aggregated data transfer to improve sensing efficiency.

  • Overview: Four use cases—smart agriculture, transportation, healthcare, and waste management—are used to extract fog-platform functional requirements.The scenarios were adapted from two prior publications and are revisited throughout the paper.
  • Smart Agriculture: Smart agriculture uses sensor networks to monitor plant growth, performance, and climate conditions across experimental fields.The scenario includes sensor stations and other sensing devices for agricultural research.
  • Smart Transportation: Transportation sensing collects bus-based air-pollution data using context information and fog devices such as sensor kits and bus stops.The associated application analyzes and visualizes city air pollution.
  • Smart Healthcare: Wearable sensors and smartphones support public-health monitoring of air quality, sound, and movement during exercise and recreation.Fog processing can reduce latency and data communication for related applications.
  • Smart Waste Management: Waste-management systems deploy sensors in bins and trucks, while fog platforms aggregate data and upload only sufficient points to reduce communication and energy use.Short-range sensors can upload opportunistically through gateways such as street lights and garbage trucks.
  • Additional Smart-City Domains: Smart water, greenhouse-gas, and power-grid scenarios use Fog/Cloud infrastructure for monitoring, analysis, and management decisions.The described applications include city-wide water monitoring, greenhouse-gas policy support, and more efficient electricity management.

4. COMMON FEATURES IN FOG COMPUTING

The survey extracts common fog-computing features by combining use-case analysis with literature review. It emphasizes heterogeneous edge devices with constrained memory, energy, communication, processing, and hardware capabilities.

  • Feature Extraction: Common fog-platform features are extracted by analyzing smart-city use cases and reviewing relevant literature.The paper presents representative research approaches for the identified features.
  • Device Heterogeneity: Fog devices span sensors, smart plugs, mobile devices, watches, bottles, and fridges, and can perform edge analytics as IoT leaf nodes.These devices generally have less computational capability than cloud-composing infrastructures.
  • Device Constraints: Constrained-node classes are organized by memory, energy, and communication availability.The cited classifications include memory classes C0–C2, energy classes E0–E9, and communication classes P0–P9.
  • Device Categories: Microcontrollers without operating systems can serve as ICOs, whereas gateway devices usually run an operating system.The paper treats exact hardware specifications as less important than broad device-category differences.

4.1. Dynamic Discovery of Internet Objects

Dynamic discovery connects heterogeneous Internet Connected Objects (ICOs) to fog gateways, while configuration establishes gateway-to-cloud connectivity and sensing behavior. The survey identifies heterogeneity, security, dynamicity, and scalability as central concerns in this process.

  • Discovery Meaning: In this section, discovery means connecting an Internet Connected Object to a gateway device to build a fog network.The survey distinguishes this from cloud-based discovery of an ICO through a datastore or management system.
  • Discovery Challenges: Discovery must address heterogeneity across communication protocols, application protocols, discovery sequences, sensing, and capabilities.The paper also identifies security and dynamicity as major discovery challenges.
  • Discovery Patterns: ICOs and gateways can discover one another through two patterns: ICOs search for discoverable gateways, or gateways search for discoverable ICOs.Both approaches typically use low-range communication protocols.
  • Connection and Configuration: Connectivity and configuration occur both between ICOs and gateways and between gateways and IoT cloud platforms.Industrial IoT platforms distinguish sensing devices from fog gateways and provide category-specific configuration options.
  • Automation: Automated connectivity is needed because manually hard-coding connections becomes inefficient and cumbersome when IoT applications involve billions of sensors.ICO descriptions can provide sensor IDs, manufacturers, types, capabilities, data structures, and driver-level configuration information.
  • Data Collection: Data collection can use push or pull responsibility, with gateways retrieving data or ICOs sending data toward gateways and cloud platforms.Instant and interval events provide alternative sensor-capture frequencies.
  • Dynamic Configuration: Cloud platforms can design an overall sensing strategy, delegate mini-schedules to gateways and ICOs, and repeatedly reconfigure them as requirements change at runtime.Sampling rate and communication frequency directly affect ICO lifetime under energy constraints.

4.3. Multi-Protocol Support: Communication Level

Fog gateways must support multiple communication protocols because protocols occupy different architectural roles and trade off range, bandwidth, power, complexity, and topology. Protocol choice therefore depends on the IoT application and network segment.

  • Trade-offs: Communication protocols are not universally superior; each is optimized for different use cases and has distinct strengths and weaknesses.The survey compares protocols by communication range, bandwidth, power consumption, and supported network topologies.
  • Architectural placement: Protocols fit different OSI/TCP-IP layers and support communication between IoT objects, fog gateways, and cloud infrastructure.The link, network, transport, and application layers provide distinct communication functions.
  • Gateway role: Fog gateways should handle multiple communication technologies to determine which protocol fits each part of a given fog network.Short-range protocols commonly connect IoT objects to gateways, while others connect gateways to cloud infrastructure.
  • Trade-offs: TCP/IP offers broad adoption but can impose comparatively high computational, memory, packet-size, and communication-power demands on resource-constrained IoT objects.Cheaper and smaller hardware is encouraging TCP/IP adoption even in constrained devices.

4.4. Multi-Protocol Support: Application Level

Smart-city applications use multiple application-level communication models and protocols, but no single protocol is widely accepted across the domain. The survey organizes these protocols by architecture, latency, device category, and supported requirements.

  • Communication models: Smart-city application communication commonly follows Device-to-Device, Device-to-Server, or Server-to-Server models.The domain lacks a single widely accepted protocol covering its broad application range.
  • Protocol comparison: MQTT outperforms HTTP in the reported evaluation because HTTP uses more battery, is less reliable, and is substantially slower.The comparison illustrates that widely used Internet application protocols may not suit IoT requirements.
  • Protocol roles: MQTT is designed for telemetry from many sources to IoT infrastructure, whereas HTTP is a Device-to-Server protocol for distributed hypertext and media.MQTT targets IoT needs and remote monitoring; HTTP follows request-response communication.
  • Protocol characteristics: CoAP targets resource-constrained devices, while AMQP emphasizes secure, reliable queuing and routing with support for point-to-point and publish-subscribe settings.CoAP uses a 4-byte header; AMQP uses TCP to assure reliability and aims not to lose messages.
  • Frameworks: IoTivity, HyperCat, and AllJoyn provide middleware, web-catalogue, or device-communication frameworks supporting discovery, data handling, interoperability, or security.IoTivity offers discovery, transmission, data management, and device management; HyperCat exposes annotated IoT assets; AllJoyn supports proximal communication.
  • Evaluation dimensions: The survey categorizes application protocols by fog location, latency, device category, and nine requirements including scalability, security, responsiveness, and power sensitivity.The requirements also cover broadcasting, event listening, small-packet distribution, unreliable networks, and transmission cost.

4.5. Mobility

Mobility lets a small number of fog gateways serve edge IoT objects across large areas through temporary connections. It consequently makes rapid discovery, technology selection, and repeated security-aware configuration essential.

  • Mobile gateways: Mobile gateways can manage many geographically distributed edge objects by moving along paths and forming temporary connections.They can acquire data, configure objects, and evaluate device health during these contacts.
  • Mobile gateways: Mobility can reduce the need to deploy unnecessary gateways because edge objects need not maintain continuous gateway communication.Gateways are comparatively expensive because they have more sophisticated capabilities.
  • Dynamic management: Each gateway movement requires repeated discovery and security procedures so connected IoT objects remain maintained.Mobility therefore highlights dynamic discovery and configuration requirements.
  • Dynamic management: Short contact windows require gateways and edge objects to find a common communication technology quickly before either moves away.Efficient discovery is needed to initiate data communication within the available connection time.
  • Use cases: Drones and mobile robots can serve as fog gateways for applications such as disaster recovery and wildfire monitoring.Phenonet is cited as an example using mobile robots as gateways.

4.6. General Data Considerations

Fog data management must handle large volumes of data distributed across the system and decide which information merits cloud storage, immediate action, or disposal. This requires distinguishing data by origin and assessing its reliability and value.

  • Data management: Managing vast amounts of data across different system locations remains an important issue requiring further research.The survey frames data management as a central concern for fog computing systems.
  • Data value: Fog computing can convert data into information quickly, retain long-term-value information in the cloud, and use short-term-value information immediately before discarding it.Room-temperature readings used to activate cooling or heating systems illustrate short-term operational information.
  • Data value: A key open challenge is determining automatically which data items have sufficient value to retain.Context awareness and data analytics are identified as related mechanisms for making this decision.
  • Data origins: System data may be sensed directly, communicated from elsewhere, or derived through analytics and processing, then combined with stored data for decisions.The management issue concerns how reliable and useful data is, not only where it originated.

4.7. Context Discovery and Awareness

Context discovery and awareness is a key fog-platform capability because proximity to edge devices supports inference about situations, device capabilities, and data quality. Context-aware gateways can use this information to configure and interpret sensor data more appropriately.

  • Context Discovery and Awareness: Fog gateways can infer location, environmental conditions, nearby devices, capabilities, and malfunctioning devices from their proximity to edge nodes.Comparing nearby devices can help identify faulty data sources before cloud processing.
  • Context Discovery and Awareness: Context information helps determine what is happening in the field and assess sensor-data quality before data fusion.Data-quality information directly affects fused results.
  • Context Discovery and Awareness: Context-aware gateways should automatically configure semantic annotation according to their deployment location and device capabilities.The requirement follows from the risk that inaccurate input data produces inaccurate analytics results.
  • Context Discovery and Awareness: Semantic annotation links metadata to sensor data and is closely connected to context discovery in smart-city applications.Context information can be inferred from sensor data or entered by users.
  • Context Discovery and Awareness: Fog platforms require modular analytics components that can be remotely pushed to resource-limited gateways on demand.Installing every application-specific analytics module would waste gateway resources.

4.10. Security and Privacy

Security and privacy are essential fog-computing requirements because distributed devices handle sensitive data across multiple communication patterns. The paper discusses protections spanning access control, authentication, encryption, secure transport, and privacy-aware data collection.

  • Security and Privacy: Fog security and privacy encompass privacy, authenticity, confidentiality, and integrity as four broad requirement categories.Authentication verifies participants, while authenticity verifies the genuineness of data senders.
  • Security and Privacy: ICO-to-ICO communication can use link-layer encryption such as AES 128 with a shared key among participating devices.Specialized hardware may manage the encryption.
  • Security and Privacy: ICO-to-cloud communication commonly uses AES 256, with decryption keys stored in the cloud rather than the gateway.This pattern supports cloud decryption of data sent by a known ICO.
  • Security and Privacy: Raw-data collection can violate privacy, so IoT analytics should collect only the data sufficient for the task and be monitored ethically.Privacy concerns become especially critical in smart-home and wearable applications.
  • Security and Privacy: Fog gateways should communicate with cloud IoT platforms through pluggable architectures that simplify support for new cloud platforms.Fog computing is presented as complementary to, rather than standalone from, cloud computing.

5. EXISTING RESEARCH EFFORTS AND TRENDS

The reviewed research largely combines fog devices with cloud platforms, emphasizing cloud companion support and data analytics while giving less attention to application-level protocols, context awareness, and semantic annotation. The comparison also highlights mobility, communication interoperability, security, privacy, and selected local-analytics use cases.

  • Existing Research Efforts and Trends: The survey compares more than 30 representative fog-computing research efforts, each addressing at least two common fog features.The comparison is organized in Table III.
  • Existing Research Efforts and Trends: Cloud Companion Support and Data Analytics are the most popular features across the surveyed research efforts.Many applications perform smaller analytics tasks in fog devices while relying on cloud platforms for larger tasks.
  • Existing Research Efforts and Trends: Multi-Protocol Support at Communication Level, Mobility, and Security and Privacy are also frequently addressed features.Mobility may require distributed directories and protocols such as LISP, while sensitive location, identity, and time data increase disclosure risks.
  • Existing Research Efforts and Trends: Multi-Protocol Support at Application Level, Context Discovery and Awareness, and Semantic Annotation are the least implemented common features.The survey associates this pattern with limited application-protocol needs, clear scenarios, and fog computing's early stage.
  • Existing Research Efforts and Trends: A healthcare case study deploys smart gateways with embedded data mining, distributed storage, and notification services for ECG feature extraction.The case examines transmitting ECG signals to gateways for local feature extraction.
  • Existing Research Efforts and Trends: Fog-based security research includes user-behavior profiling and decoys on gateways to mitigate data-theft attacks.The cited approach uses one-class support vector machines for user-behavior profiling.

6. LESSONS LEARNED

Sustainable smart-city sensing depends on efficient resource use across sensing, communication, analytics, security, and cloud–fog coordination. The survey identifies design practices that reduce energy, data movement, deployment costs, and disruption while retaining cloud scalability and knowledge discovery.

  • Smart cities must accommodate growing populations and needs with limited resources, making efficient sensing infrastructure essential for sustainability.
  • Dynamic discovery and runtime configuration reduce deployment costs and let operators adjust sampling, communication, and sensor activation to extend infrastructure lifetime.These parameters directly affect energy consumption, especially for battery- or solar-powered ICOs.
  • Selecting communication protocols carefully, including opportunistic switching and application-level choices, can reduce energy and communication costs.The survey specifically contrasts MQTT with HTTP as an example of application-protocol impact.
  • Applying analytics early summarizes data before cloud transmission, reducing bandwidth, storage, communication-energy, and computational costs.The paper emphasizes applying analytics at the right time and place in the data pipeline.
  • Security and privacy are sustainability requirements because opposition to deployments or disruption of dependent applications can undermine sensing infrastructures.The survey also notes that cloud platforms remain needed for scalability and large-scale analytics despite fog efficiencies.
  • Context-aware sensing can reduce energy use by controlling sensing periods, protocols, transmission timing, and sampling rates.Load balancing and opportunistic sensing further help avoid redundant sensing and resource wastage.

7. CHALLENGES AND OPPORTUNITIES

The section identifies platform, intelligence, interoperability, and security challenges for fog computing in sustainable smart cities. It emphasizes that ideal platforms remain underdeveloped and require further experimentation.

  • Platform requirements: Ideal fog platforms should support experimentation with different approaches, techniques, and algorithms across fog-computing features.Both IoT and fog computing are described as comparatively immature fields requiring substantial experimentation.
  • Platform requirements: Interchangeable plug-ins should support varied communication protocols, application protocols, analytics frameworks, and feature implementations.Plug-ins should also be removable because fog gateways have limited computational resources.
  • Research gaps: Context discovery, awareness, and semantic annotation are fairly ignored in existing literature despite their potential to improve sensing-infrastructure sustainability.The comparison is based on the literature summarized in Table III.
  • Research gaps: Distributed intelligence must span smart things, buildings, fog gateways, and cloud infrastructure to support timely reactions and decisions.Relevant concerns include domain knowledge, policies, cost-efficient computation, and context- and semantics-aware real-time models.
  • Research gaps: Security challenges include cyber attacks, trust and authentication, network security, and data security across smart-city fog infrastructures.The paper links these risks to service failures, wrong emergency decisions, misbehavior, criminal activity, and broader system disruption.

8. CONCLUSIONS

The conclusion presents fog computing as a response to the inefficiency of centralized cloud processing for increasingly sensor-rich smart-city applications. The survey evaluates fog and edge-analytics research and highlights fog platforms as important for sustainable IoT infrastructure.

  • Conclusions: Increasingly inexpensive sensing technology is driving sensing capabilities into everyday objects.This trend supplies the expanding sensing base for smart-city applications.
  • Conclusions: Centralizing all sensor data at cloud nodes is inefficient from both computational and communication perspectives.Fog computing addresses this approach by using fog gateways for edge analytics.
  • Conclusions: The survey analyzes fog-computing and edge-analytics research to identify important fog-platform functionalities and field trends.The conclusion frames these findings around sustainable smart-city IoT infrastructure.
  • Conclusions: Fog computing platforms are highlighted as important for building sustainable IoT infrastructure for smart cities.The conclusion connects this importance to the survey’s evaluation of existing research efforts.
Loading 1703.07079v1…