Edge computing has become an effective solution to overcome the limitations of cloud computing in supporting context-aware and delay-sensitive services in the Internet of Things (IoT) era by moving service supply and data computation from the cloud to the edge. This article deliberates the role of edge computing in improving the overall performance of IoT.
Image Credit: Ar_TH/Shutterstock
An Overview of Edge Computing in IoT
The IoT is a revolutionary approach interlinking uniquely addressable virtual and physical devices through various communication protocols. Potential devices include traditional electronic gadgets, embedded objects, wearable devices, smart tags, body sensors, bio-nano things, and smartphones.
Typically, these devices possess multiple sensors that collect environmental data, which are data-driven intelligence’s core elements. Thus, massive deployment of these devices results in a substantial increase in collected data, which must be analyzed and processed before providing meaningful results for users.
However, lightweight IoT devices possess limited computation ability, which makes the task of data processing and analysis extremely challenging. Edge computing is a feasible solution for addressing this problem. This new computing paradigm directs computational data, services, and applications away from cloud servers to network edges.
Edge computing is characterized by real-time access to network information, ultra-low latency, and high bandwidth. By leveraging edge computing, app developers and content providers could bring their services closer to users, which accelerates the service response speed.
IoT applications often require massive data transmission, privacy preservation, and real-time response. Edge computing can meet these IoT applications’ requirements on a large scale more effectively compared to cloud computing. Although the common goal of both edge computing and IoT is to ensure seamless computing anywhere and anytime, their roles are different within a system, with edge computing focusing on near-field computation, while IoT focuses on endpoint sensing.
Thus, in systems like edge-computing-driven IoT (ECDriven-IoT), these two technologies/edge computing and IoT complement each other. The ECDriven-IoT is suitable for several applications, including latency-sensitive and responsive IoT applications like smart homes and cities, smart healthcare, mobile augmented reality and virtual reality, and Industry IoT (IIoT).
Advantages of Edge Computing in IoT
High quality of services (QoS) and real-time response are the biggest advantages of edge computing in IoT. Edge computing provides shorter network latency compared to cloud computing as edge servers are located geographically closer to IoT devices. This superiority better supports real-time high-demand IoT applications.
Additionally, the amount of offloaded data to the cloud is significantly reduced as most of the data is processed in edge servers. Thus, higher QoS is ensured for real-time IoT applications using ECDriven-IoT. Low consumption of energy and high scalability are other major advantages of edge computing in IoT.
Most IoT nodes are primarily power-limited devices, and synchronizing massive quantities of sensing data to the remote cloud results in power wastage. Using edge computing, IoT nodes only require transmitting data to local edge servers, which reduces the IoT nodes’ energy consumption.
Thus, ECDriven-IoT extends the IoT nodes’ lifetime and decreases the maintenance overhead. Large-scale access requirements are one of the inherent challenges in cloud-based IoT systems. The cloud server could become the system bottleneck owing to large numbers of concurrent connections from the IoT nodes.
In ECDriven-IoT, moderate computing resources are offered by base stations/edge servers in a distributed manner, leading to good scalability satisfying the needs of large-scale IoT applications like autonomous driving or smart cities.
Edge-computing Architectures for IoT
Machine Learning-based Architectures: Hierarchical fog-assisted computing architecture (HiCH) and transferring trained models (TTM) are the major machine learning-based architectures for IoT. Common edge infrastructures utilize conventional machine learning techniques within the cloud, which results in delays in the response time.
This issue can be mitigated using the HiCH architecture based on the MAPE-K model containing four key components, including execute, plan, analyze, and manage/monitor, with each component being responsible for a different role. Enhancement in response time and latency are the primary benefits of this architecture.
A lack of training data required for training models to apply new services is a significant problem in in-home applications. Thus, TTM among different smart homes has become a major area of research. For instance, an architecture was proposed in a study that can transfer activity-recognition models (ARMs) from source homes to target homes.
The architecture has three layers: the first layer, where a set of smart homes are grouped based on their cities; the second fog-node layer, where every fog node is responsible for a particular city; and the third cloud-system layer, which enables the fog nodes to communicate with each other and manage the environmental parameters, shared settings, knowledge, and data. Although secure edge service is the biggest advantage of this architecture, it is also vulnerable to runtime attacks.
Data Placement-based Architectures: IFogStor, IFogStorZ, IFogStorG, and IFogStorM are the key data placement-based architectures. The IFogStorZ and IFogStor were proposed by leveraging fog-node distributions and variations to minimize the overall latency of retrieving and storing IoT data in fog nodes.
In the system architecture, the components include IoT services, data centers, fog nodes, and IoT devices. Although IFogStorZ is easy to implement, a significant loss of optimality takes place when data producers are not close to the data consumers. Additionally, the number of fog nodes and IoT services varies among subregions, leading to unbalanced subproblems. The IFogStorG was proposed to minimize the data placement strategy complexity and improve the runtime performance.
This technology can effectively manage dynamic changes within the network topology. A latency issue is generated when data consumers located in diverse geographical regions are subscribing to the same data while only a single replica of the data exists in the proper fog node.
Thus, a greedy algorithm, referred to as IFogStorM, was proposed to minimize overall latency by adopting a multi-replica data placement strategy. Results displayed that this architecture achieved a 10% and 6% greater reduction in overall latency compared to IFogStorG and IFogStorZ, respectively.
Challenges of Edge Computing in IoT
Heterogeneity of Edge Computing and IoT: IoT devices are utilized everywhere and vary across multiple scenarios. Thus, different communication protocols and hardware devices exist in IoT systems. Similarly, for edge computing, the edge node deployment architecture requires various solutions for diverse scenarios.
Thus, unifying the diversity of edge computing and IoT and making them complementary to each other while integrating both technologies remain a big challenge. The cooperation architecture of ECDriven-IoT, communication protocols, and hardware devices must form industry standards to effectively implement edge computing in heterogeneous IoT systems.
Coordination between Computing and Communication: An ECDriven-IoT system is significantly more complex compared to only edge computing or IoT-based systems. Although the communication between IoT devices and edge servers will lead to additional consumption of computing resources and power by the entire system, both IoT devices and edge servers are limited in computing and power capacity.
For instance, shifting the entire workload from IoT devices to edge servers can strain communication bandwidth and overwhelm the processing power/computing capacity of edge nodes, leading to higher costs. Thus, the workload between IoT devices and edge servers must be allocated by balancing the cost of computation and communication.
Complicated Security and Privacy Issues: Ensuring the privacy and security of systems is a persistent challenge in edge computing and IoT. However, these issues become more complicated in ECDriven-IoT due to its limited computing capability and heterogeneity.
Edge servers and IoT devices remain highly vulnerable to different attacks, and the danger to the entire system becomes higher once edge servers and/or IoT devices are compromised. Thus, a qualified ECDriven-IoT system must consider all potential security threats and their countermeasures in diverse application scenarios.
Overall, edge computing reduces latency and improves the performance of IoT applications. While there are challenges like device heterogeneity and security, edge computing also offers benefits like reduced energy consumption and improved scalability.
References and Further Reading
Kong, L., Tan, J., Huang, J., Chen, G., Wang, S., Jin, X., Zeng, P., Khan, M., Das, S. K. (2022). Edge-computing-driven internet of things: A survey. ACM Computing Surveys, 55(8), 1-41. https://doi.org/10.1145/3555308
Huang, T., Lin, W., Li, Y., He, L., Peng, S. (2019). A latency-aware multiple data replicas placement strategy for fog computing. Journal of Signal Processing Systems, 91, 1191-1204. https://doi.org/10.1007/s11265-019-1444-5
Hamdan, S., Ayyash, M., Almajali, S. (2020). Edge-Computing Architectures for Internet of Things Applications: A Survey. Sensors, 20(22), 6441. https://doi.org/10.3390/s20226441
Alwarafy, A., Al-Thelaya, K. A., Abdallah, M., Schneider, J., Hamdi, M. (2020). A survey on security and privacy issues in edge-computing-assisted internet of things. IEEE Internet of Things Journal, 8(6), 4004-4022. https://doi.org/10.1109/JIOT.2020.3015432