PaperlayerView exact revision on arXiv

Source-linked AI summary

Big Data Meet Cyber-Physical Systems: A Panoramic Survey

Rachad Atat, Lingjia Liu, Jinsong Wu, Guangyu Li, Chunxuan Ye, Yang Yi

arXiv:1810.12399v1cs.LGstat.ML

TL;DR

CPS growth is producing vast, complex data while creating security challenges and motivating sustainable solutions. This survey provides a panoramic synthesis of CPS big-data sources, infrastructure, analytics, cybersecurity, and green approaches. It concludes that CPS big-data management requires coordinated solutions across these areas, with unresolved issues in security, privacy, latency, and scalability.

  • Problem

    CPS generate enormous, continuously flowing data that complicate cybersecurity, privacy protection, data management, and low-latency communication.

  • Method

    The paper conducts a panoramic survey of CPS big-data collection, storage, access, caching, routing, processing, analytics, security, and green solutions.

  • Results

    The survey synthesizes multi-cloud processing, privacy-preserving analytics, integrated security strategies, and green data-management, architecture, and software solutions for CPS.

  • Takeaways & Limitations

    CPS big-data research spans interconnected technical and sustainability challenges, and robust systems require solutions matched to application latency, resource, and security constraints.

  • Takeaways & Limitations

    Homomorphic encryption is impractical and inefficient for large datasets, while anonymization requires further evaluation across data types and performance conditions.

Abstract

from arXiv · show

The world is witnessing an unprecedented growth of cyber-physical systems (CPS), which are foreseen to revolutionize our world {via} creating new services and applications in a variety of sectors such as environmental monitoring, mobile-health systems, intelligent transportation systems and so on. The {information and communication technology }(ICT) sector is experiencing a significant growth in { data} traffic, driven by the widespread usage of smartphones, tablets and video streaming, along with the significant growth of sensors deployments that are anticipated in the near future. {It} is expected to outstandingly increase the growth rate of raw sensed data. In this paper, we present the CPS taxonomy {via} providing a broad overview of data collection, storage, access, processing and analysis. Compared with other survey papers, this is the first panoramic survey on big data for CPS, where our objective is to provide a panoramic summary of different CPS aspects. Furthermore, CPS {require} cybersecurity to protect {them} against malicious attacks and unauthorized intrusion, which {become} a challenge with the enormous amount of data that is continuously being generated in the network. {Thus, we also} provide an overview of the different security solutions proposed for CPS big data storage, access and analytics. We also discuss big data meeting green challenges in the contexts of CPS.

I. INTRODUCTION

CPS connect physical components with cyber twins and generate large, complex data across diverse applications. This survey organizes CPS big-data collection, storage, processing, analytics, security, and green solutions.

  • CPS foundations: More than 50 billion sensors were expected to connect to the Internet, averaging 6.58 devices per person by 2020.
  • CPS foundations: CPS integrate physical components and cyber twins, while IoT connects different CPS for information transfer.The cyber twin is described as a simulation model representing physical things.
  • Data and analytics: Data are acquired from context-aware computing and communications, including physical, virtual, and logical sensors, and from social computing.
  • Data and analytics: Cloud computing distributes CPS workflows across servers to support storage, processing, management, scalability, lower latency, and lower power consumption.
  • Data and analytics: CPS big-data analytics uses reduction and mining tools, including feature selection, dimensionality reduction, clustering, visualization, classification, and real-time analysis.
  • Security and sustainability: Large, continuously flowing CPS data complicate cybersecurity because unauthorized access, alteration, and false-data injection can compromise data and computational results.
  • Survey scope: The survey covers CPS big-data collection, storage, access, caching, routing, processing, analytics, cybersecurity, and green solutions.Its sections include sensing sources, cloud and multi-cloud processing, clustering, security, applications, and sustainability.
  • Survey contributions: The survey presents a broader CPS perspective than prior surveys by integrating big-data security across collection, storage, access, processing, and analytics.

III. SOURCES, GENERATIONS, AND COLLECTIONS OF BIG DATA FOR CPS

CPS big data comes from context-aware systems, communications, social computing, and remote sensing. The survey distinguishes raw and processed information and illustrates these sources and types.

  • Data sources: Sensors are increasingly deployed because advances have made them smarter, more efficient, lower-cost, and more robust.
  • Context-aware computing and communications: Context-aware data can originate from physical sensors, virtual sensors, logical sensors, middleware infrastructures, context servers, or user-provided inputs.
  • Data types: Raw data are collected directly from the environment, whereas processed data provide information that is meaningful and easier to interpret.
  • Remote sensing: Remote sensing collects object data from a distance, including multi-spatial and multi-temporal data for earth observation and climate monitoring.
  • Remote sensing: Real-time remote-sensing processing applies filtration, load balancing, and parallel processing to geographically distributed, high-dimensional datasets.
  • Remote sensing: Remote-sensing data quality can be evaluated using statistical inference and prior dataset knowledge to obtain an unbiased estimator.

C. Social Computing

Social computing incorporates human behavior and context into CPS data collection, while caching and routing address communication efficiency. Participatory and mobile crowd-sensing reduce deployment costs but introduce quality, trust, and participation challenges.

  • Social computing sources: Mobile big data include user applications, network performance, service characteristics, geographic information, and subscriber profiles.
  • Social computing sources: Social computing integrates social behaviors and contexts into web technologies to support prediction of social dynamics and social-network operations.
  • Participatory sensing: Participatory sensing lets users collect and share information with groups or the public, substituting users for physical sensors and reducing deployment costs.
  • Participatory sensing: Privacy-preserving participant coordination and peer-reviewed local models can select participants or limit reported information to intermediate results.
  • Mobile crowd-sensing: Mobile crowd-sensing fuses mobile-device data with mobile-social-network services to answer more complex queries.
  • Mobile crowd-sensing: MCS applications use smartphones, incentives, prediction, and Bluetooth/WiFi gateways for cost-effective data uploading, but require sufficient participant contributions.
  • Caching and routing: Caching CPS big data can reduce exchanged traffic, latency, and energy consumption, while cache misses and interference can impose costs.

B. Big Data Communications for CPS

Big-data communications for CPS must connect massive data collection, transmission, processing, and delivery while addressing latency, scalability, reliability, and energy use. The surveyed approaches include device-to-device networking, concurrent collection, cloud and fog processing, clustering, and compression.

  • Networking infrastructure: Resilient networking infrastructure is essential to bridge CPS data collection, distribution to data centers, and delivery to intended users.The paper emphasizes fast and reliable networking for data-intensive CPS applications and services.
  • Device communications: D2D communication supports massive machine-type communications when CPS devices are located close to one another.The surveyed integration of wireless sensor networks and mobile cloud computing also targets fast and reliable sensory-data delivery.
  • Energy-aware routing: Sustainable routing trades energy efficiency against capacity usage efficiency through dedicated or shared backup protection.Shared backup protection consumes less capacity than dedicated protection but increases energy expenditure.
  • Data collection: Concurrent collection trees and dynamically assigned servers accelerate high-density or delay-sensitive CPS data acquisition and transmission.Concurrent streams collect data from common devices, while IOCP servers assign micro-sensors based on concurrent connections.
  • Summary and insights: Caching, filtering, cloudlets, D2D, SDN, backup protection, and social or geographical structures can speed CPS data handling and reduce latency.These strategies target collection, processing, and distribution across traffic and device levels.
  • Cloud processing: Exabyte-scale CPS data are impractical to process on individual machines, motivating cloud-based computing despite scalability, latency, availability, and transaction concerns.Cloud centers provide elastic, pay-as-you-go computing and storage resources, while shared-resource transactions can introduce untruthfulness, unfairness, and inefficiency.

B. Multi-Cloud Data Processing

Multi-cloud processing distributes data-intensive CPS workflows across cloud centers and combines parallel execution with clustering and related data-management techniques. The surveyed methods address workload placement, energy, access cost, throughput, fairness, and clustering scalability.

  • Multi-cloud workflows: Large-scale IoT workflows consisting of interdependent computing entities need distribution across multiple cloud centers because of their data-intensive nature.Multi-cloud frameworks are intended to support multiple applications rather than a single application type.
  • Resource placement: Virtual-machine placement can reduce energy consumption or data-access costs in national cloud data centers.The surveyed approaches consider quality-of-service trade-offs and use Lagrangian-relaxation heuristics for placement optimization.
  • Resource allocation: Multi-cloud task-distribution methods target higher throughput, fairer limited-resource usage, and lower provider operating costs.A multi-round combinational double-auction mechanism is also used to allocate virtual machines across cloud centers.
  • Data clustering: Data clustering partitions objects into groups of similar attributes and supports distributed storage, task execution, parallel computing, and query processing.The survey distinguishes hierarchical clustering from centroid-based clustering.
  • Centroid-based clustering: Traditional k-means becomes unsuitable for big-data applications as cluster counts increase because empty clusters and convergence iterations grow.Enhanced k-means variants target clustering quality, execution time, and accuracy.
  • Alternative clustering methods: Hierarchical and fuzzy clustering variants address cluster estimation, initialization quality, user-defined similarity, and partial cluster membership.Hierarchical k-means uses progressively smaller datasets to obtain higher-quality initial centroids.
  • Scalable clustering: Clustering features summarize patterns in CF-trees but can incur time and scalability costs when many data points must be scanned.Dynamic thresholds based on regional density are presented as a way to vary micro-cluster sizes.

D. NoSQL

The survey presents NoSQL and fog computing as responses to heterogeneous, large-scale CPS data and distributed service needs, while highlighting security and privacy concerns. It also connects clustering and mining tools with CPS intelligence and workload processing.

  • NoSQL: Fixed-schema relational databases are poorly suited to heterogeneous big-data processing, whereas NoSQL relaxes relational constraints for flexible and scalable access.The survey associates NoSQL with higher availability and faster read/write operations for unstructured data.
  • NoSQL data models: NoSQL databases store data as key-value, document, or column-oriented records, supporting different structures and schema operations.Document databases can use complex structures such as XML, while column-oriented systems simplify adding and deleting columns.
  • NoSQL systems: Operational NoSQL systems support online transaction processing, while analytical systems based on MapReduce, Hadoop, and Spark support complex decision-support queries.Decision-support workloads process larger tables than online transaction-processing workloads.
  • Fog computing: Fog computing extends cloud services toward end users through proximity, geographical distribution, and mobility support.These properties make fog suitable for latency-sensitive CPS scenarios.
  • Edge processing: Fog computing can offload core-network traffic and keep sensitive data inside the network while enabling edge data mining and distributed storage.Smart-gateway fog computing was reported to enhance health monitoring with notification services at the network edge.
  • Security and privacy: Fog security solutions face gateway-authentication issues, Man-in-the-Middle risks, wireless attacks, and possible private-information leakage.Cloud security solutions may not transfer directly to fog devices positioned at network edges.
  • Summary and insights: Multi-cloud environments can process CPS workloads more efficiently, but they require optimization for virtual-machine placement, energy, security, fairness, and cost.The paper identifies additional open issues spanning cloud, edge, and fog architectures.

B. Real-time Analytics

Real-time and cloud-based analytics transform high-volume CPS streams into structured, manageable information for decision making across healthcare, transportation, environmental monitoring, and smart cities. The survey covers mining, visualization, distributed storage, and resource prediction while noting cloud security, privacy, and ownership challenges.

  • Real-time analysis: Real-time analysis converts streaming data into structured form before applying big-data analysis tools for decision making and control.Healthcare, transportation, environmental monitoring, and smart-city applications are identified as requiring real-time decisions.
  • Streaming analytics: Hybrid-stream analytics, Spark Streaming, and mobile edge nodes are surveyed for real-time video, network-traffic, and transit-data analysis.The examples cover high-speed Internet monitoring and descriptive analytics on transit buses.
  • Visualization: Representative data forms and geographical information systems make complex large-scale CPS information easier to extract and understand.GIS supports real-time analysis in healthcare, urban planning, transportation, emergency response, and public safety.
  • Cloud analytics: Cloud-based analytics scales extraction, aggregation, and analysis of big-data streams across different granularities.Statistical analysis and machine learning reduce massive datasets to manageable sizes for information extraction and hypothesis testing.
  • Cloud limitations: Cloud analytics remains constrained by security, privacy, and data-ownership challenges.Mobile cloud computing is also used to address memory, battery, and CPU limitations in healthcare applications.
  • Resource prediction: Predicting application workloads and resource requirements can optimize bandwidth allocation and future virtual-machine provisioning.The surveyed systems use MapReduce resource prediction and linear regression to anticipate new resource and VM needs.
  • Space-time analytics: Distributed spatio-temporal indexing and cloud-based space-time analytics address storage, scalability, and retrieval challenges from widely deployed sensors.The examples include VegaIndexer and Internet-of-Vehicles applications.

1) Tools for Data Mining:

The survey reviews tools and approaches for extracting useful information from CPS big data, spanning real-time analytics, cloud platforms, security, and anomaly detection. It also highlights privacy-preserving analytics challenges and security controls.

  • Real-Time Big Data Analytics: Storm provides distributed real-time big data analysis, while Splunk offers a web-based platform for searching, monitoring, and analyzing data.Storm uses master and worker nodes organized into task-specific topologies.
  • Cloud-Based Big Data Analytics: Google’s cloud analytics platform combines GFS for storage, BigTable for management, and MapReduce for cloud computing.The platform is designed to meet large-scale storage and usage demands.
  • Cloud-Based Big Data Analytics: Cloud-analysis performance can be optimized by predicting virtual-machine and resource requirements, efficiently processing workloads, and using hybrid clouds.The passage presents these techniques as ways to speed analysis and optimize computing resources.
  • Security Analytics: Machine learning detects future security anomalies by training on large datasets and identifying unusual activities in network traffic.Higher accuracy requires a large volume of training data.
  • Security Analytics: Anomaly detection approaches include grid-based Poisson modeling, multivariate Gaussian analysis, and autoencoders for smart-meter data.The cited autoencoder approach targets high-accuracy, low-latency detection of energy-consumption and operational anomalies.
  • Privacy-Preserving Analytics: Privacy-preserving analytics remains difficult because encrypted-data analysis is inefficient, costly, and non-straightforward, while k-anonymization feasibility is insufficiently evaluated for big-data types and computation time.The survey also describes homomorphic encryption and vector-similarity protocols as approaches for analytics over protected data.

C. Summary and Insights

The survey concludes that robust CPS security requires combining multiple strategies and adapting them to application latency and device-capability constraints.

  • Summary and Insights: A single security solution is insufficient for robust CPS protection, so strategies should be combined according to application requirements.The survey contrasts cryptography and advanced controls with visualization and machine-learning anomaly detection for delay-sensitive CPS.

VIII. BIG DATA MEET GREEN CHALLENGES FOR CPS

The survey frames green CPS from two directions: reducing the environmental cost of big-data infrastructure and applying CPS big data to sustainable sectors. It covers collection, storage, computing, processing, and applications.

  • Two Green View Directions: Green CPS research addresses both greening big-data systems and using big data with CPS for green applications.The survey organizes solutions into sustainable applications, green data management, green architectures, and green software.
  • Green Data Collection and Storage: Energy consumption challenges arise during sensed-data collection because limited communication ranges require objects to relay surrounding data.Data compression is identified as another response spanning storage, collection, transmission, processing, and analysis.
  • Green Communications: Improved approaches for machine-to-machine communications reduce device, radio-access-network, and core-network activity through optimized signaling without negative impacts on legacy human-to-human terminals.The cited work targets higher power savings in M2M or MTC systems.
  • Greening Big-Data Computing: Cloud centers consume substantial energy while processing data chunks in parallel, motivating sustainable computing techniques such as the MMGreen job scheduler and DVFS.DVFS adjusts virtual-machine frequencies while maintaining users’ quality of service.
  • Greening Big-Data Computing: A VM-placement and traffic-balancing framework achieved 50% energy savings by integrating application characteristics, network topology, and traffic patterns.The framework assigns VMs for energy-efficient routing and minimizes active switches while balancing traffic.

3) Green Processing:

Green processing reduces CPS energy use by limiting redundant communication, improving execution efficiency, and applying big-data methods to environmental, economic, and social applications. The survey also links smart-grid analytics with operational and privacy challenges.

  • Green Processing: CAPE uses checkpoints and distributed parallel execution to avoid unnecessary restarts after hardware or software failures and can reduce energy consumption.It executes shared-memory program threads in parallel on distributed-memory architectures.
  • Green Processing: Spate coding reduced exchanged communications by 62% while improving resource utilization during distributed big-data processing.The approach combines index coding and network coding with side information from processes sharing a physical node.
  • Green Applications: CPS big-data applications support greener environmental and economic systems through pollution monitoring, building-energy optimization, and smart-meter power management.The survey also identifies social media and participatory sensing for smart-grid management and fuel-efficient routing.
  • Smart-Grid Applications: Smart grids use large-scale sensing and automated decisions to manage energy patterns, understand user behavior, reduce power-plant construction needs, and address renewable-supply fluctuations.The cited smart-grid architecture includes data resources, storage, analysis, and transmission.
  • Smart-Grid Applications: Smart-grid cyber infrastructure introduces security risks, while collecting users’ energy-usage information can improve performance but affect privacy.The survey notes security situational-awareness techniques for threats occurring over short time periods.

B. Military Applications

Big data supports military security, satellite-communication risk assessment, smart-city infrastructure, urban analytics, and surveillance applications.

  • Military security: Real-time authentication of command-and-control messages is important for securing military cyber-physical infrastructures.A broadcast authentication scheme uses special digital signatures to accelerate signature generation and verification.
  • Military security: Satellite-communication risk can be assessed by combining attack ease with attack impact.Mitigating waveform, radio-frequency access, foreign presence, physical access, and traffic-concentration threats addresses exploitable vulnerabilities.
  • Smart cities: Urban sensor data can reveal dynamic city patterns, including traffic patterns used to compute routes.Sources include smartphones, smart cards, and vehicle-mounted sensors deployed indoors and outdoors.
  • Smart cities: A smart-city architecture separates technologies, middleware, management, and services into four layers.The management layer applies analytics to extract information, test hypotheses, and draw conclusions.
  • Safety systems: Computer-vision deep learning can recognize twelve human activities without prior knowledge for security monitoring.The described model is intended for surveillance and human-activity recognition applications.

D. Medical Applications

Medical CPS applications support telemedicine, health monitoring, and energy-efficient e-health services, while public-safety systems distribute emergency information through autonomous D2D networks. The public-safety model addresses reliability by selecting multicast nodes, tracking delivery delay, and modeling range and channel-error losses.

  • Medical applications: CPS health systems target telemedicine applications including cardiology, surgery, patient monitoring, diabetes management, and vital-sign monitoring.The passage links these applications to timely, efficient, and effective medical decisions.
  • Medical applications: Medical body area networks use nearby or implanted biomedical sensors and short-range wireless technologies to sense vital signals.Research focuses on reducing interference in medical frequency bands as e-health systems spread through hospitals and health centers.
  • Medical applications: The µSMS architecture reduces memory overhead, software-component load time, and event-propagation time for e-health systems.Its sensor nodes exchange event-driven information with dynamic memory and variable payloads such as GPS coordinates and home context.
  • Public safety: Public-safety networks require timely exchange of voice or data among first responders during disasters and emergencies.Network resilience and survivability are identified as foremost requirements.
  • Public-safety model: When infrastructure is unavailable, users form autonomous D2D clusters that send maps, videos, and pictures to a command center for analysis.The command center can return assistance assessments, emergency notifications, and coordination information for cluster members.
  • Public-safety model: The multicast procedure minimizes total distribution delay by selecting a node with the highest transmission rate to an unvisited node.It updates delay using the last successful reception time and stops when all nodes are visited or delay exceeds θD.
  • Reliability analysis: Link reliability combines transmission-range coverage, channel-error loss, and total-delay threshold failure under stationary-user assumptions.Receiver distances are modeled with isotropic direction and a Rayleigh-distributed distance; channel fading includes path loss and log-normal shadowing.

X. BIG DATA CHALLENGES AND OPEN ISSUES FOR CPS

CPS faces unresolved challenges in security, privacy, correctness, latency, data heterogeneity, and cross-application collaboration. The survey identifies lightweight, integrated, and cross-layer approaches as directions for addressing these issues.

  • Security: CPS security remains insufficiently addressed for critical-infrastructure integration, while heterogeneous applications complicate trust, authentication, and access control.Cryptography can impose latency, power, and key-management costs, motivating secure interoperation protocols for dynamic CPS.
  • Privacy: Privacy-preserving analytics remains limited because homomorphic encryption is inefficient at scale and anonymization requires evaluation across data types and mining techniques.Future protocols should support encrypted heterogeneous data across hierarchical CPS modules with time- and cost-efficient analytics.
  • Correctness: CPS must adapt to dynamic physical environments while operating correctly with little or no human supervision.Model-based simulation can support early failure detection and verification of complex-system integration.
  • Latency and reliability: Large-scale CPS data access, routing, transmission, and processing can obstruct ultra-reliable low-latency communications by consuming substantial time.The survey highlights faster cloud access, fewer retransmissions, and cross-layer optimization of latency, reliability, and security as research directions.
  • Data collection and processing: Future CPS communication should support dynamic on-the-fly device-to-device connectivity without preconfiguration, controllers, or infrastructure deployment.The survey also points to edge-near processing and improved information fusion for faster handling and higher-quality analysis of heterogeneous data.
  • Data integration: Homogenizing heterogeneous, unstructured data sources remains an open issue needed to accelerate analysis, draw meaningful conclusions, and enable automated decisions.Some CPS missions may also require collaboration across applications, such as mobile health and transportation systems.

XI. CONCLUSION

The survey presents CPS and big-data processing as mutually advancing technologies with potential for automated applications and services. It synthesizes CPS data workflows, security, green challenges, and open issues to frame future research.

  • Conclusion: CPS and big-data processing and analysis mutually benefit technological advancement.The survey connects this relationship with future applications, services, and opportunities involving artificial intelligence, machine learning, and neuromorphic computing.
  • Conclusion: The survey covers CPS big-data collection, storage, access, caching, routing, processing, and analysis, alongside security solutions and sustainability challenges.It also identifies challenges and open issues for CPS systems.
Loading 1810.12399v1…