Service Provision and Price Strategies in Edge Computing

Abstract

This chapter of research explores the strategic deployment of computing services at the edge of the network to optimize social welfare and system performance. It involves building models for where and how to place services within edge computing systems, considering factors such as demand, cost, and usability. Additionally, this chapter investigates dynamic pricing frameworks that allow for real-time pricing adjustments to operate resource allocation efficiently. These strategies aim to balance the economic objectives of service providers, such as profit maximization and cost minimization, with the quality of service delivered to users. The goal is to develop pricing and service placement mechanisms that are both economically beneficial and capable of meeting the latency, computational, and energy consumption demands of edge applications.

1. Introduction

Edge computing is a technology that shifts data processing from centralized data centers to locations closer to the data source, such as smartphones, factory sensors, or vehicles. This approach offers several benefits: it reduces the time and distance for data transmission, enabling faster data processing results; alleviates the load on central servers; and enhances privacy and security since data do not have to be transmitted over long distances.

Now, I will explain the connection between edge computing and several application areas shown in Figure 1:

• Industrial Internet of Things (IIoT): In industrial settings, numerous sensors and devices collect data (such as temperature, pressure, and speed). Edge computing enables rapid processing of these data near these devices, allowing real-time monitoring and optimization of factory operations without the need to send large amounts of data to remote servers.

• Wireless Sensing: In wireless sensing technology, devices detect and understand their surroundings through wireless signals. Edge computing can process this information immediately at the edge, supporting rapid responses in applications such as smart homes and security monitoring.

• Edge Intelligence: This refers to the application of Artificial Intelligence (AI) within an edge computing framework. For example, in a smart factory, AI can be used for immediate analysis of data collected from machine sensors, enabling quick decision-making, such as predicting equipment failures in advance, thereby reducing downtime.

Figure 1.

A description of application and issue in edge computing.

Within these applications, three key issues of edge computing are highlighted: resource management, economic modeling, and service provision. Here, we will use straightforward language to delve into these seemingly complex concepts.

• Resource management: the magic kitchen of edge computing

  Imagine edge computing as a magical kitchen. In this kitchen, our primary task is to ensure that each chef (representing data processing devices) has ample ingredients (data) and tools (computing resources) for efficient operation. Proper allocation of ingredients leads to faster and better-quality dish preparation (processing results). This underscores the importance of resource management in edge computing, ensuring the optimal utilization of each resource.

• Economic model: the profit rule of edge computing

  Let us discuss the economic model. In our magical world, this is likely to establish a set of rules that ensure the profitability and operational sustainability of our magical kitchen. It is like finding the perfect recipe that not only attracts customers (users) to savor our culinary creations but also ensures profitability for our kitchen. This “recipe” must balance the costs (the expenses of running edge computing) and the revenues (fees paid by users for services), ensuring fairness to users and profitability for operators.

• Service provision: the secret to customer satisfaction in edge computing

  Here, we arrive at service provision. The main goal is to ensure every customer receives the service they expect, much like ensuring every dish in the magical kitchen satisfies the customer’s taste. In edge computing, this means users can swiftly and reliably receive the data processing services they need. Additionally, this service must be secure, just as restaurants ensure the safety and hygiene of the food they serve.

In general, the importance of resource management, economic modeling, and service provision is to ensure the successful operation of a magical kitchen. Resource management ensures an efficient cooking process, the economic model secures the kitchen’s profitability and sustainable growth, while high-quality service provision guarantees customer satisfaction. When these three elements coexist harmoniously, the magic of edge computing is maximized, offering users a swift, safe, and efficient data processing experience. This is similar to a successful restaurant, where delicious dishes, reasonable pricing, and excellent service are key to retaining customers.

Therefore, when discussing edge computing, this section is not only talking about technology but also a strategy encompassing efficient resource utilization, the formulation of sensible business models, and the provision of optimal user experiences. This is the essence of the magic of edge computing and a significant reason for its emergence as a future technological trend.

The remainder of this chapter is organized as follows: it begins with an introduction to applications at the edge, followed by a detailed analysis of service provision and pricing strategies from the perspectives of resource management, economic modeling, and service delivery. In detail, Section 2 introduces applications of fog computing and human sensing. Section 3 details methods for resource management and optimization. Section 4 discusses the economic model and pricing strategies. Section 5 provides an analysis of service provision and networking. Section 6 provides a conclusion for the chapter.

2. Technological innovations and applications

Recently, researchers have explored developments in Mobile Edge Computing (MEC) and Industrial IIoT. It highlights new methods and systems that improve internet transit for MEC providers. Today, MEC is vital in telecommunications because it places computing power near the data source, reducing delay, and saving bandwidth. Section 2 also discusses fog computing – an extension of cloud computing that further reduces data processing time. Additionally, Section 2 covers wireless human sensing technologies, which allow remote monitoring of human activities and conditions. These technologies largely explain the importance of edge computing applications.

In the era of big data, cloud computing has significantly enhanced daily life and technology development. However, cloud computing’s centralized nature and potential for latency issues in time-sensitive services, along with security risks like privacy breaches during data center failures or network disconnections, are notable shortcomings. Fog computing emerges as a complementary solution, extending computation to the cloud’s edge. It primarily employs edge devices, like routers and local servers, offering enhanced security and sustainability. Unlike cloud computing’s large centralized resources, fog computing utilizes numerous distributed, renewable fog nodes, each with modest computational power. This distributed approach reduces data center workload, improves efficiency, and reduces costs. Despite its advantages, fog computing also brings new security challenges, especially with the interconnectedness of IoT devices, raising concerns like personal data theft and exposure. Du details a novel differential privacy-based query model for fog data centers, focusing on enhancing data privacy in heterogeneous fog computing environments. The model integrates Laplacian noise to safeguard data, balancing privacy protection, and data utility. This approach effectively resists privacy attacks, crucial for large-scale datasets in fog computing. Both fog and MEC deal with data processing closer to the source, where privacy and security are significant. Applying this model could significantly improve data privacy and security in MEC mirroring the benefits observed in fog computing [1].

Wireless human sensing is a crucial aspect of human-computer interaction, as it allows computers to recognize and understand human activities and even emotions. Numerous endeavors have been undertaken to tackle this issue through various wireless technologies, including WiFi, RFID, Bluetooth, Radar, and Zigbee. Each of these sensing technologies possesses distinct characteristics and advantages, rendering them appropriate for particular application contexts. For instance, WiFi-based systems are adept at non-intrusive human sensing, whereas RFID-based systems excel in identifying individuals in environments with multiple persons. A comprehensive review is provided to assist users in comprehending the range of available wireless human sensing technologies, thereby facilitating an informed decision regarding the most fitting solution for their requirements. This review specifically delves into promising human sensing applications and classifies them into three categories: vital sign monitoring, gesture recognition, and activity recognition [2]. The advent of wireless human sensing applications has yielded a multitude of profound benefits for society, particularly in realms such as healthcare, gesture recognition, and the recognition of human activities. Within a Smart Wireless Human Sensing (SWHS) system, the dynamics of human movement apply a significant influence on the propagation of wireless signals. Using the resulting changes in signal characteristics, human activity can be identified by non-invasive methods.

The importance of healthcare has become increasingly apparent in recent years, especially in the context of an aging global population, which poses a major challenge to existing healthcare infrastructure. Increased attention has been given to routine healthcare as it is directly linked to the serious health problems prevalent among older persons. One such health issue is sleep apnea, a serious medical condition characterized by interrupted breathing during sleep. More than half of the world’s population aged 65 and above suffer from a sleep disorder. The prevalence of sleep apnea syndrome is estimated to be between 20 and 40%, a proportion that increases with age. The manifestations of sleep apnea can be observed through a variety of physical symptoms, including episodes of shortness of breath or apnea, increased blood pressure, potential heart complications, skin discoloration, and even loss of consciousness.

Traditional diagnostic methods, such as polysomnography (PSG), while widely used in clinical settings, come with significant inconvenience and financial burden, especially for older patients who often require care and supervision at home. To address these challenges, developing and implementing respiratory monitoring systems using ubiquitous wireless signals has emerged as a promising alternative. These systems are non-invasive and provide a more convenient and accessible way to monitor respiratory health.

3. Resource management and optimization

Resource management and optimization in edge computing involve efficiently allocating and utilizing computational resources (such as processing power, storage, and bandwidth) at the edge of the network, close to the data sources. This process aims to enhance application performance, reduce latency, and manage the constraints of edge devices, which often have limited resources compared to centralized cloud computing resources. The objective is to optimize the use of limited and distributed resources in edge computing environments to meet the demands of applications and services, ensuring they run effectively and efficiently. Section 3 will discuss the feature extraction, performance, and cost in heterogeneous edge systems. The advantages of the following methods can be concluded. The feature extraction problem in edge systems benefits for network limitations [3, 4], the performance problem helps the tradeoff between energy consumption and latency [5], and the load balance problem is conducive to latency-sensitive applications [6]. However, these methods only focus on partial performance elements, network power consumption, latency, etc. In most cases, resource allocation and optimization problems need to consider multiple factors.

In the rapid expansion of the Internet of Everything, network bandwidth constraints are a major challenge in transferring large amounts of data to cloud or edge servers. To alleviate this problem, a common strategy is to pre-process image data on devices before uploading it to edge servers, thereby reducing network traffic and bandwidth pressure. A key aspect of the process is the extraction of discriminant features from images, which is crucial for recognition tasks. However, due to the lack of image label information necessary for feature extraction and the fluctuating availability of resources on these devices, this task is challenging on mobile devices.

Mobile devices often run multiple applications at the same time, and frequent startup and shutdown of applications can affect resource availability. Therefore, it is important to develop discriminative image feature extraction methods, based on current mobile device resources, to reduce network traffic while maintaining high recognition accuracy. Ding proposes a method to solve this problem [3]. The detailed steps are as follows: after uploading the pre-processed image data to the edge server, the server processes the data and returns the label information of the most similar image to the mobile user. The main challenge is to create an effective feature extractor, called Extractor E, which can identify key discriminant features. These features are critical because they significantly affect the quality of service (QoS) of moving image recognition, including recognition accuracy and response time. To solve this problem, the authors introduce the discriminant feature extraction (DFE) algorithm. The DFE algorithm is designed to generate an extractor E that can extract a minimal but efficient discriminant feature set from image data, improving recognition accuracy and shortening response time by reducing network traffic and the number of matched features. This is achieved by building a new similarity function that preserves the intra and inter class structure of the image dataset and by introducing a tradeoff parameter that balances these structures [3].

The second major challenge is dynamically selecting extractors based on a mobile device’s fluctuating resources. To address this issue, Ding also introduces the Nested Discriminative Feature Extraction (NestDFE) algorithm. This algorithm segments the extractor E into several sub-extractors, collectively constituting a multi-capacity extractor. Notably, this multi-capacity extractor occupies the identical memory space as the original extractor E. This efficiency is achieved as each sub-extractor of lesser capacity shares parameters with and is embedded within, a sub-extractor of greater capacity, thereby obviating the need for additional memory space. Consequently, the NestDFE algorithm permits mobile devices to dynamically choose the most suitable sub-extractor without necessitating substantial memory space consumption [3].

Traditionally, these processes have relied on mobile cloud computing, where users upload images to cloud servers for retrieval results. However, these methods often suffer from long network transmission delays. To solve this pr oblem, Wang proposes a method leveraging MEC, which allows mobile users to interact with edge servers that are closer to them than cloud servers, thereby reducing transmission latency [4]. By facilitating faster data processing and reduced response times, MEC plays a key role in various fields such as the Internet of Things, e-healthcare, and autonomous vehicles. One challenge with existing MEC-based image retrieval methods is the need to extract a large number of features from the image, which must then be uploaded to a cloud server. Since extracting features from image datasets stored in the cloud is isolated, the process can be inefficient and result in poor retrieval accuracy.

Wang introduces a new cloud-guided feature extraction method for mobile image retrieval. In this approach, a cloud server learns the projection matrix P from its dataset of labeled images. Subsequently, this matrix P is utilized to extract discriminant features from images, thereby creating a low-dimensional feature dataset. The edge server then applies matrix P to the image, and the resultant features are uploaded to the cloud server for label recognition. This technique notably diminishes network traffic and simultaneously enhances retrieval accuracy. A prototype system was implemented to validate this approach, and extensive experiments in a real MEC environment are conducted. The results show a remarkable reduction in network traffic by 93% and an improvement in retrieval accuracy by 6.9% compared to existing state-of-the-art image retrieval methods in MEC [4].

Another work focuses on optimizing energy efficiency in MEC while ensuring performance [5]. This work addresses the challenge of balancing energy consumption and performance in computation offloading tasks in MEC environments. The authors propose an energy-minimizing optimization problem and solve it using the Karush-Kuhn-Tucker (KKT) conditions. The solution involves a request offloading scheme based on energy consumption and bandwidth capacity. Numerical results demonstrate that the proposed offloading scheme offers better energy consumption and delay performance compared to local computing and complete offloading methods. The research contributes to the field by providing an effective method for computation offloading in MEC, which is crucial for mobile users who require lower energy consumption and better performance for their tasks. This research is particularly relevant for mobile users who seek to balance energy consumption with effective performance in their computational tasks within MEC environments [5].

Edge computing has emerged as the foremost method for delay-sensitive applications, strategically placing compute and storage resources at the network’s edge. Its fundamental function is to efficiently manage data transmission between the cloud downlink and the terminal uplink and to organize these data effectively at the edge, laying the groundwork for subsequent data analysis and processing. This raises a critical question: How can data be efficiently organized, stored, and retrieved at the edge? To address this, a variety of methods have been devised for establishing data storage and retrieval services at the edge, encompassing structured, unstructured, and hybrid approaches. However, research in heterogeneous edge environments regarding data storage and retrieval services is still lacking, particularly in load balancing of edge data stores. The principal aim of edge computing is to optimally utilize edge node resources, including storage, bandwidth, and CPU, to satisfy user requirements. Given the limited resources of individual edge nodes, cooperation among nodes becomes essential. A common strategy is to distribute the workload evenly across edge nodes, thereby optimizing resource utilization. Chen introduces the w-strategy, a novel load-balancing technique that uses weighted Voronoi graphs to achieve optimal distribution across heterogeneous edge nodes [6]. Using the software-defined networking (SDN) paradigm, the proposed approach supports data distribution based on virtual space-distributed hash tables (DHTS). Such heterogeneity presents additional complexities in load distribution. The w-strategy comprises two key components. Firstly, utilizing the SDN paradigm, this work maps nodes and data within the edge network to virtual-space coordinates in the SDN control plane. The w-strategy then divides the load area of each node to achieve balanced distribution, based on the virtual-space location of nodes and data. Secondly, w-strategy accounts for edge node heterogeneity by assigning weights to nodes according to their storage and computational capabilities and subsequently distributing loads based on these weights. The main challenge of the w-strategy lies in integrating these two aspects: achieving even load distribution while considering node capabilities. Evaluation results demonstrate that the w-strategy outperforms existing methods, such as greedy routing for edge data (GRED) and Chord, by enhancing resource utilization efficiency by an average of 20% [6].

4. Economic models and pricing strategies

Compared with cloud computing data center systems, the resources in edge computing systems are very limited, which leads to increased competition for resources among users who expect high-quality services. Inefficient resource allocation decisions may lead to low service quality, high energy consumption, and high operating costs. At present, most of the work is focused on the design of resource allocation algorithms in edge computing systems. The pricing mechanism has a significant effect in the competitive use of resources. The decentralized distribution of edge nodes, the heterogeneity of resource requirements, and the competition among users for high-quality services make resource allocation and pricing in edge computing systems a challenging problem.

4.1 Pricing mechanisms in edge computing

The distributed nature of edge nodes, the heterogeneity of resource demands, and the competition among users for high-quality services render the allocation and pricing of resources in edge computing systems a challenging issue. Price mechanisms have some advantages with this challenge including:

1. In the edge environment, providers often aim to maximize profits while satisfying the needs of edge users, making the protection of providers’ interests the primary goal of edge resource management. Pricing models, such as profit maximization or cost minimization, can effectively achieve this goal.

2. The edge environment encompasses a variety of roles belonging to different groups, such as end-users, edge service providers, network service providers, and edge brokers, each with their distinct objectives and demands, such as cost, profit, and utility, and facing various constraints like budget limits and resource capacities. Pricing mechanisms can quantify these different goals and attempt to solve them.

3. More and more new edge applications and computing modes, such as big data analysis and distributed machine learning frameworks, require a significant amount of computing and bandwidth resources. Pricing mechanisms, such as bandwidth congestion perception pricing, can regulate users’ demand for resources and maximize resource utilization.

Often, these roles in the edge systems have conflicting goals, making economic models and pricing strategies highly effective for the allocation and management of edge resources. More specifically, through negotiation and game mechanisms among these roles, under given constraints, incentive and pricing mechanisms can provide the best solutions for self-interested parties. These works mainly consider the edge and node pricing model and seldom take the chain of cloud-edge-node situation.

4.2 Dynamic pricing strategies for edge computing systems

For an individual Edge Service Provider (ESP), its primary objective is to obtain revenue through the provision of services to users. A prerequisite for its participation in an edge federation is that it can reduce the costs associated with constructing and maintaining edge nodes. In the motivation shown in Figure 2, each edge node is associated with a base station, enabling users to access the network via these base stations. It is assumed that a user has agreed with ESP 1 in edge 1, meaning the user can only access the network through ESP 1’s base station. Initially, the user’s service is located on edge 1. As a consequence, the user is required to pay a fee to ESP 1. Upon leaving the service area of edge 1 and entering the combined service areas of edge 2 and edge 3, the system must decide whether to migrate the service in response to the user’s movement. Due to edge 2 being at full capacity, it cannot provide efficient service, necessitating that the user’s service remain on edge 1. However, the increasing distance from the user to edge 1, where their service is hosted, would inevitably lead to higher communication delays, significantly impacting the user’s Quality of Service (QoS). Thus, the optimal solution is for ESP 1 to form a federation with ESP 2, granting ESP 1 the right to utilize a portion of ESP 2’s resources. Consequently, the system can migrate the user’s service to edge 3, which is closer to the user with some additional migration delay. To maintain the stability of the federation, ESP 1 compensates ESP 2 for servicing the customer’s request. As the user moves beyond the service area of edge 3 to that of edge 4, the system weighs the migration delay against the communication delay to decide whether to migrate the service. If the service is not migrated, ESP 1 continues to compensate ESP 2. This analysis demonstrates that the edge federation expands the service range and capabilities of an ESP, enabling it to leverage idle resources for greater profit. Essentially, the process involves redirecting user service requests through the federation to the most cost-effective edge node for processing. The dynamic cooperative optimization of service placement and pricing by network edge nodes can enhance both the user experience and ESP revenue, achieving a win-win situation.

Figure 2.

Motivation for dynamics pricing for edge systems.

Dynamic pricing usually sets the price of resources within a small time frame and can quickly respond to changes in resource demand. Especially in situations where user information is unknown, the system can adjust prices based on the real-time scarcity of resources to pursue high returns from resource providers and efficient allocation of resources. Han [7] studied the dynamic pricing scheme of idle vehicle resources under the condition that the arrival of vehicles is random and unknown and minimized the cost of the entire edge computing system under the premise of guaranteeing the quality of service. Xue [8] utilized D2D technology to fully utilize the computing resources of idle edge users to cope with network congestion caused by some hot spots. The author defines the ratio of computing resource prices to the service capabilities of candidate auxiliary users as the dynamic transaction price. The decision-making problems of resource allocation, task offloading, and candidate auxiliary user selection are defined as minimizing total energy consumption and total working user expenses. The original problem is decomposed using the alternating direction multiplier method to obtain the resource allocation and dynamic pricing scheme. Siew [9] also studied the problem of uneven resource allocation in multi access wireless networks based on this idea, and proposed two dynamic pricing mechanisms to achieve the supply of computing resources, which, respectively, improved social welfare and platform benefits. Wang [10] studied the dynamic pricing problem of drone resource supply in hotspot areas. Considering the differences in drone hover time and service capacity, the authors designed multiple drone deployment schemes for heterogeneous hotspot areas. By balancing the occurrence rate of users in hotspot areas and the flight distance of drones, the overall revenue of drones was improved. Although dynamic pricing can also address unknown user requests, it almost always considers short-term optimization goals and cannot provide overall service assurance. Therefore, Chen et al. proposed a real-time dynamic pricing strategy to optimize ESP’s long-term goals [11, 12].

The primary goal in multiple-edge systems is to minimize Internet data transfer costs for small ESPs in the IIoT. Chen’s research introduces the Sublessor framework [12], which reduces WAN data transmission costs by using nonprofit ESPs as neutral brokers who charge only for technology and administration. This framework leverages the ability of MEC providers to share infrastructure and exchange network data locally, significantly lowering costs within the same region [12]. Chen’s another work highlights challenges in edge computing, emphasizing the need for effective load evaluation on edge nodes due to the unpredictability of requests and timing [11]. This necessitates real-time, adaptive pricing decisions. Additionally, aligning pricing with resource allocation is complex, given user sensitivity to price fluctuations, which can impact customer satisfaction. Therefore, Chen proposes separating these processes to operate on different timescales. Furthermore, current dynamic pricing strategies often overlook long-term profit maximization, underscoring the importance of strategic, long-term planning despite the randomness of network requests.

4.3 Auction-based pricing strategies for edge computing systems

Auction is a popular form of transaction that efficiently allocates a seller’s resources to a buyer at a competitive price in the marketplace. Many works study resource auctions in edge scenarios [13]. Hung [14] investigated live video streaming services bidding on resources of edge servers to improve the smoothness of live video streaming. The edge system allocates cache resources utilizing combinatorial clock auctions and uses dynamic planning to finalize the price and quota of cache resources, which improves the utility of the entire edge system. Jiao [15] considered a mobile blockchain scenario in which miners competed for edge resources to implement blockchain services and proposed a resource allocation method based on combinatorial auctions to improve social welfare. There is another work [16] aiming at the heterogeneous nature of edge users’ demands and the competition for high service quality among multiple users. Two different resource allocation mechanisms are proposed, namely, an auction-based individually rational and envy-free allocation mechanism and a linear programming-based approximation algorithm. The algorithm does not guarantee the absence of jealousy, but the approximate solution can ensure that the difference with the optimal solution is controlled within a definite range.

Double Auction is designed for multi-edge and multi-user auction scenarios. Sun [17] proposes a credible and efficient algorithm rooted in a two-way auction model. It guarantees no-loss and employs dynamic bidding to match “mobile device-edge server” pairs while meeting local constraints. Multi-Round Auction is a flexible and efficient auction method, that facilitates online buyers to adopt different bidding strategies. Wang [18] firstly designs a set of non-competitive incentive mechanisms to encourage edge cloud to sell resources, while ensuring the QoS of users and the interests of edge cloud; then designed a multi-round auction mechanism for competitive environments to achieve matching of edge terminals and resources and maximize the profit of the edge cloud. Zhang [19] first proposed to fully exploit the vehicle computing resources in the parking lot as a supplement to the edge computing environment to provide low-latency services to the mobile users, and the ESP can shut down the idle edge servers to save energy and operation cost. The authors propose a multi-round auction to maximize the utility of resources, such as vehicle computing, which optimizes the service provisioning structure at the edge and achieves a win-win situation for both vehicles and ESPs. There is another work [20] recruiting edge server owners to carry the computing tasks of edge users for the cloud. Zhou’s method uses the reverse auction method to select appropriate edge nodes to provide computing resources for edge users and minimize the cost of cloud service centers. Xu [21] studies the scenario of mobile users choosing trusted edge nodes to cache resources in a campus environment and proposes a reverse auction-based method to select the most suitable edge node for users among the nodes that satisfy user requirements.

5. Service provisioning and networking

Addressing service provisioning and networking challenges by focusing on service quality, social welfare, and availability involves a subtle and integrated approach. Edge computing improves service quality by reducing latency and optimizing bandwidth, making it ideal for time-sensitive applications and efficient data transmission. However, maintaining consistent service quality across diverse edge environments poses challenges, including ensuring low energy consumption and managing the variability in computational resources [22, 23]. Edge systems can extend high-quality digital services to remote and underserved areas, contributing to greater accessibility and inclusivity. The challenge lies in ensuring that these benefits are equitably distributed and that edge systems are designed with social welfare in mind [24]. Edge computing also enhances system resilience and fault tolerance, ensuring service availability of virtualized network functions (VNF) even in the face of disruptions. The scalability of the service chain is another advantage, allowing the network to accommodate growing demands. However, ensuring high availability in diverse and potentially harsh operational environments requires sophisticated resource management and infrastructure resilience strategies [25, 26].

5.1 Service quality

Service quality in MEC is critical as it directly affects user satisfaction and the overall effectiveness of edge computing services. Ensuring low energy consumption and high reliability of data processing and transmission is the key to providing high-quality services. This is achieved by strategically placing services on edge nodes closest to users and optimizing network routing to reduce latency. In addition, consistent performance is critical even when demand or network conditions fluctuate. QoS metrics such as throughput, latency, and energy consumption are continuously monitored and managed to ensure that services meet expected standards.

Tao’s work involves optimizing the placement of virtualized network functions (VNF) in multi-region mobile edge systems [22]. This work emphasizes balancing performance metrics such as latency and energy consumption. The authors propose a cost-driven VNF placement Edge (VPE) strategy, which effectively solves the VNF placement problem using a graph-cutting method. The VPE approach considers three cost types: communication, location, and energy consumption. This approach is superior to other approaches in terms of latency and energy consumption, providing a balanced solution for VNF placement that takes into account both user and system perspectives. This research contributes to improving edge system performance and cost efficiency, which is critical to supporting high-quality mobile user experiences in multi-regional edge environments. The proposed VPE solution uses the graph-cutting method to efficiently determine the location of low-cost and high-performance VNF. The study demonstrated the effectiveness of VPE in improving latency and energy consumption, with adjustable cost weights to meet the needs of different users [22].

For network issues, Deng addresses the critical need for efficient resource management, specifically in the context of sampling node selection and sampling duration allocation [23]. This work underscores the necessity of managing the constrained resources in MEC systems with precision. The importance of meticulously determining the number of sampling nodes and judiciously allocating the duration of sampling is considered, especially considering the bandwidth constraints of packet sampling collectors. Subsequently, this paper delves into the application of Deep Reinforcement Learning (DRL) in resource allocation within wireless environments, referencing relevant studies. DRL, an amalgamation of Reinforcement Learning (RL) and deep learning, has demonstrated its efficacy in autonomously managing resource allocation and adapting to evolving time-based demands. For instance, a particular study developed a resource allocation strategy that accounts for the computational prowess of MEC servers and the cache capacities of base stations (BSs). Aiming to enhance network-wide flow-level sampling precision, the authors advocate for a strategy that involves adaptively choosing sampling nodes and dynamically formulating sampling techniques for each interval. This is achieved using the Simple Probabilistic Sampling (SPS) method. This approach aims to optimize network monitoring within the constraints of MEC systems’ resources, addressing the challenges of achieving high accuracy in flow-level sampling [23].

5.2 Social welfare

Social welfare in the context of MEC involves making computing resources more accessible and benefiting a wider segment of society. This includes deploying edge computing solutions in a way that addresses the digital divide, ensuring that underserved communities also benefit from technological advances. It extends to creating fair resource allocation algorithms that distribute computing power and storage fairly between users and applications. By reducing the centralization of computing resources, MEC can contribute to a more balanced and inclusive digital ecosystem.

The existing research assumes that all edge nodes in a network are owned by a single Edge Internet Provider (EIP). However, the reality of the MEC environment is often characterized by the presence of multiple EIPs, such as AT&T in the United States, Mobile in China, and Bells in Canada. In such a diverse MEC landscape, two primary challenges arise, prompting our investigation into edge federation networks with multiple EIPs to realize the concept of edge federation. First, each EIP tends to build its own private edge computing environment, serving only the customers it contracts with its federation, while this approach limits each EIP to managing its edge nodes and serving only its customers. For EIP participating in edge federation, it is important to ensure that participants reduce costs when building and maintaining edge nodes while enhancing customer service capabilities. Determining how EIP in edge federated networks can gain greater social welfare is another major challenge. Chen solves the problem of service placement in a MEC network consisting of multiple EIP [24]. These EIPs are coordinated in a joint manner to maximize social welfare, a concept known as a federation. While seemingly simple, the task is complex due to several factors: the EIP must decide whether to service the request internally or outsource the request to another EIP, and determine the appropriate edge node to handle these tasks [24].

In addition, Edge Infrastructure Providers (EIPs) must decide on the number of requests to accept from fellow EIPs and allocate appropriate edge nodes for these requests. A critical question emerges: how can EIPs effectively schedule various user service requests (internal, incoming, and outsourced) to improve service performance? Furthermore, to augment the overall profit, each EIP should adopt strategic pricing for services offered to both customers and those from other EIPs within the federation. This involves adjusting prices in response to user QoS and request capacity, aiming to optimize social welfare. Addressing these challenges, this paper introduces a model for horizontal collaboration among edge federations, encompassing edges from all EIPs. Initially, the model treats the service placement dilemma as a programming issue, to maximize social welfare. It then presents two dynamic pricing strategies for EIPs: one for setting standard prices for customers and another for determining insourcing prices for services from other EIPs [24].

5.3 Availability

Availability is another key factor in MEC service delivery. It refers to consistent and reliable access to computing resources and services. In an edge computing environment, where resources are distributed and can be subject to a variety of local conditions, maintaining high availability is challenging but essential. This includes designing robust systems that can withstand hardware failures, network outages, and fluctuations in user demand. Redundant strategies, such as replicating critical services across multiple edge nodes, and dynamic resource allocation, which can quickly respond to changes in demand or network conditions, are key to ensuring high availability.

With the rise of network function virtualization (NFV) and MEC, network service providers (NSPS) are increasingly outsourcing network functions (NFs) to MEC, as it provides greater scalability and flexibility for NFV deployment and maintenance. In this process, each user request is processed through a service function chain (SFC) consisting of multiple VNFs. These VNFs are software alternatives to traditional hardware-based middleware that are arranged in a specific order to respond to requests. However, unlike hardware, VNFs are less reliable due to potential software errors and host failures. To improve reliability, it is wise to incorporate redundancy in the primary VNF within the SFC, where the key issue is to determine the optimal MEC node to place each VNF and the number of backup instances required to meet the availability requirements of each SFC.

Consequently, Li proposes an availability-aware provisioning strategy for SFCs in the MEC environment, designated as APoS [25]. APoS main aim is to maximize the number of served requests while complying with the requirements and reliability standards of SFCs. In addition, APoS tackles two principal challenges: one, the effective mapping of both primary and backup Virtual Network Functions (VNFs) to fulfill the availability needs of SFCs, and two, the diminution of latency for user access to SFCs. The first challenge is conceptualized as an integer nonlinear programming (INLP) problem, considering the resource constraints of each MEC node. Addressing this NP-hard problem, Li employs a novel binary N-back search method to optimally position the primary and backup VNFs. Regarding the second challenge, the objective is to minimize the average delay for all requests within each time slot. To this end, this work introduces an online service switching (OSS) method. This method accounts for queuing, communication, and switching delays, providing an optimal solution underpinned by theoretical evidence. The evaluation of these methods using real-world datasets shows significant improvements over benchmarks. Specifically, the proposed method achieves an approximate 20% increase in request acceptance and up to a 30% reduction in delay on average [25].

Within NFV, SFC stands as a pivotal concept, denoting the sequence of VNFs that network traffic traverses from its source to its destination. The core challenge of NFV-enabled networks is SFC embedding, which requires optimally assigning VNFs to nodes and routing each stream so that it traverses the required SFC. Recent research has focused on various goals such as minimizing costs, ensuring availability, and reducing latency. This approach can significantly reduce total latency, which is an inherent limitation of VNFs sequential processing in traditional SFC. Another key issue with NFV is that when separated from dedicated hardware, the vulnerability of network functions increases due to potential software failures or physical node failures. To address this issue, the researchers advocate adding backup VNF instances to enhance availability. The authors discuss a mixed-deployment SFC using sequential and parallel VNFs.

However, the dynamic, virtualized, and multiplexing nature of NFV also increases the risk of security vulnerabilities, presenting a larger attack surface. Current validation tools fall short in addressing the complexities of this environment, particularly given that existing guidelines predominantly cater to traditional, sequential Service Function Chains (SFCs), which are merely a subset of hybrid SFCs. As a result, agile and comprehensive validation techniques for hybrid SFCs are still uncharted territory. To address this deficiency, the authors have developed verification scheme for hybrid service function chain (vHSFC), a novel validation framework specifically designed for hybrid SFCs [26]. This framework enables enterprises to authenticate SFC execution in real time accurately. The vHSFC framework employs a streamlined, verified routing protocol to identify various SFC breaches and threats, including packet alterations and deviations from compliant SFC forwarding paths. This work details the creation and deployment of vHSFC prototypes using multiple containers, alongside rigorous testing with actual network traffic. The findings demonstrate that vHSFC is adept at ensuring consistent enforcement and detecting unforeseen breaches, all while sustaining a manageable level of system overhead [26].

6. Conclusion

In conclusion, this chapter has provided a comprehensive exploration of the strategic placement and pricing of computing services at the network’s edge. The detailed analysis has demonstrated how optimal service placement in edge computing systems can significantly enhance social welfare and system performance. This involves careful consideration of demand, cost, and usability factors. Moreover, dynamic pricing frameworks have been investigated that support real-time pricing adjustments, ensuring efficient resource allocation and balancing economic objectives with quality of service. The findings suggest that these strategies not only are economically advantageous for service providers but also meet the stringent requirements of edge applications in terms of latency, computational power, and energy consumption. This chapter underlines the potential of edge computing to revolutionize service delivery in our increasingly connected world, offering a roadmap for service providers to optimize both their operational efficiency and user satisfaction.

Nomenclature

AI	artificial intelligence
MEC	mobile edge computing
IIOT	industrial Internet of Things
SWHS	smart wireless human sensing
PSG	polysomnography
QoS	quality of service
DFE	discriminative feature extraction
NestDFE	nested discriminative feature extraction
KKT	Karush-Kuhn-Tucker
VPF	virtualized network function
VPE	VNF placement edge
SDN	software defined networking
DHTs	distributed hash tables
MIP	mixed-integer programming
DRL	deep reinforcement learning
DREAM	dynamic pricing-based online control algorithm
RL	reinforcement learning
SPS	simple probabilistic sampling
BS	base station
EIP	edge internet provider
NSP	network service provider
NFV	network function virtualization
SFC	service function chain
INLP	integer nonlinear programming
OSS	online service switching