Contents Introduction
The growth in global connectivity has increased the traffic flowing through the network. The era of the Internet of Things (IoT) and Internet of Everthing (IoE) will be remembered for the significant growth in devices connecting to the global networks and for the challenges related to provisioning and maintaining the applications and services consumed by devices and end-users. The network is evolving to support the growth in connectivity and data consumption and to support applications and services that require low latency and high throughput. The IoE integrates objects, people, data, and processes using developing technologies and intelligent services [1]. Emerging IoE services include autonomous vehicles, linked vehicles, extended reality, brain-computer interfaces, flying cars, haptic technology, etc. [2]. The Fifth Generation (5G) mobile cellular specifications promise to greatly enhance the performance of computation, storage, and networks for various application use cases. Mobile cellular is transforming business and industry by connecting vehicles, factories, and farms to the network. The 5G mobile networks and upcoming 5G Beyond networking (5GB) facilitate various leading-edge services, which are categorised into three usage conditions: enhanced Mobile Broadband (eMBB), providing up to 20 Gbit per second data-rate connections; massive Machine-Type Communication (mMTC), enabling around one million devices per square kilometer connection density; and Ultra-Reliable Low-Latency Communication (URLLC), ensuring dependable connectivity and communication with a transmission delay of no more than 1 ms [3], [4]. While URLLC is considered a significant 5G innovation, it poses new challenges that need to be addressed to fully gain the benefits of this technology. URLLC services are characterised by a high level of responsiveness and an availability of 99.999 percent. URLLC aims to effectively facilitate real-time applications, e.g., tactile internet, remote surgery, and Vehicle-to-Everything (V2X). The Sixth Generation (6G) mobile cellular specifications are likely to push the boundaries of what can be achieved, to provide novel applications and services and to enhance network intelligence and edge computing capabilities [5].
Multi-access Edge Computing (MEC), standardized by the European Telecommunications Standards Institute (ETSI), is an essential system that provides compute and storage at the network edge to facilitate URLLC services [7]. The goal of MEC is to provide a low latency, context aware compute and storage capability close to end-user devices. MEC act as cloud systems at the network edge and include Network Function Virtualization (NFV) [7] to deliver network services through the sequential connection of Virtual Network Functions (VNFs) on virtualised infrastructure, serving as the fundamental basis for the implementation of network slicing within the 5GB technology. Network slicing over a 5G network is depicted in Fig. 1. Services with diverse demands (eMBB, mMTC, and URLLC) provided over a common infrastructure are delivered using network slicing in 5G. NFV and network slicing enable enhanced flexibility in resource management. Providing URLLC services is a challenge that involves effectively handling abstracted resources across several interconnected networks while ensuring the fulfilment of service needs [8], [9].
![FIGURE 1. - Network slicing over 5G network [6].](/mediastore/IEEE/content/media/6287639/10380310/10792437/grego1-3515077-large.gif)
This article presents a comprehensive review of the literature related to MEC and network slicing within the framework of resource allocation in 5GB and 6G networks. Here, an overview of MEC slicing and its associated technologies with a particular focus on the difficulties faced in achieving optimal resource allocation are presented.
The rest of the paper is organized as follows. Section II briefly reviews the existing works related to resource allocation, network slicing and MEC. Section III provides a description of MEC. Section IV presents network slicing; definition, framework, use cases, and Software-Defined Networking (SDN) and NFV. Section V investigates the integration of network slicing and MEC. Section VI identifies resource allocation in 5GB and 6G with MEC slicing. Section VII identifies research challenges. The conclusion is provided in Section VIII. The list of acronyms used in this paper is provided in Table. 1.
Related Work
The 5G network facilitates connectivity between end-user devices and service provider networks and facilities that are often referred to as the Cloud. A wide range of resources sustain the functionality of the numerous applications available to end-users today. Resource management is a critical activity carried out at mobile base stations. Optimally allocating resources such as radio spectrum, power, storage, and computing can enhance the system’s performance. This survey investigates and compares the current resource allocation techniques that integrate with MEC and network slicing in 5GB.
The authors of [10] provided an optimisation classification of several resource allocating elements and conducted a comprehensive evaluation of resource allocation strategies in a Cloud Radio Access Network (CRAN). The key components of effective resource allocation and administration in CRAN were highlighted, and include optimizing throughput, user allocation, spectrum utilization, selecting Remote Radio Heads (RRH), network services, and power distribution. The authors also introduced innovative applications, such as virtualized CRAN, Non-Orthogonal Multiple Access (NOMA)-based CRAN, heterogeneous CRAN, non- and full-duplex-enabled CRAN, and millimetre-wave CRAN to illustrate the potential enhancements in system performance with CRAN technology.
The authors in [11] discussed the potential issues associated with upcoming 5G systems and highlighted their importance. They discussed a survey strategy and the different methodologies used in recent surveys that categorize radio resource management (RRM) programmes. The study investigated the HetNet RRM methodologies, with a particular focus on enhancing allocation of radio resources, coupled with several additional approaches. The RRM strategies were classified based on their performance evaluations and subsequently subjected to qualitative analysis and comparison. The challenges of RRM schemes was noted by the authors in relation to their implementation and computational aspects.
The researchers in [12] published a comprehensive evaluation of resource allocation methodologies in distributed 5G networks. The initial discussion revolved around HetNet and other network scenarios. Resource allocation models were also investigated. The authors presented a classification approach for evaluating resource allocation systems found in the literature. Several challenging unresolved concerns and potential study paths were provided. The authors introduced two methods that address the radio access challenges in HetNet design for 6G communications: a control theory-based approach and a learning-based approach.
Radio Interference and Resource Management (RIRM) technologies were considered by [13] and the research contribution involves analyzing, synthesizing, and summarizing established RIRM methodologies to tackle the challenges encountered by 5G Radio Access Network (RAN) systems. The study highlighted unresolved research questions related to recently proposed RIRM technologies.
The study participants in [14] focused their attention on resource allocation methods for 5G network slicing, covering the fundamentals and models. NFV and SDN concepts were presented, including their significance for 5G network slicing. The network slicing management and orchestration architecture was investigated, and highlighted as a foundation for resource allocation techniques. In addition, an analysis of resource classifications, with isolation factors was carried out for RAN slicing and core network slicing. Resource allocation algorithms were categorized based on their objectives and illustrated using real-life examples. This study also identified practical strategies to address open research challenges.
The study by the authors of [15] analyzed vehicular network resource allocation strategies by considering short-distance vehicle to vehicle communications and vehicle to cellular network communications. The authors examined challenges and opportunities related to resource allocation in existing vehicular networks, along with possible areas for future research. The resource allocation in 5G was addressed by the authors of [16], focusing on RAN and core network. They categorized existing works according to application scenarios, research objectives, and network architecture. The authors also identified several important challenges for future study.
Shah et al. in [17] provided a review of the literature, providing a comprehensive overview of the cloud-native approach by synthesizing findings from various studies. The paper explored the details of how MEC and SDN can be leveraged to enable efficient network slicing. The authors emphasized the significance of a cloud-native approach, specifically focusing on a microservice architecture, to facilitate dynamic and flexible network slicing tailored to diverse 5G use cases. By utilizing SDN for resource management and allocation, the integration of MEC and network slicing aims to enhance service delivery, optimize network performance, and meet the stringent requirements of emerging 5G applications. The survey provides valuable insights into the evolving landscape of cloud-native network slicing, shedding light on key challenges, opportunities, and future directions in this rapidly evolving field.
Table. 2 provides a brief overview and analysis of recent survey papers on resource allocation in 5G networks integrated with MEC and network slicing.
Multi-Access Edge Computing
The goal of edge computing is to provide computation and storage capability at the network edge. By providing computation and storage resources closer to the RAN and User Equipment (UE) reduces the propagation delay that might otherwise occur if UE are required to communicate with cloud servers; therefore, applications can achieve real-time performance [29], [30]. Data processing at the edge of a network offers various benefits; including reducing the risk of data leakage by eliminating single points of failure, minimizing traffic congestion, and enhancing network scalability. Edge computing a significant facilitator of improved 5G network application and service provision [1]. ETSI has established and standardized edge computing as Multi-access Edge Computing (MEC) to highlight the multi-technology implementations. The process of integrating applications from various service providers, suppliers, and third parties in an effortless and effective way is enabled by the ETSI framework [31], [32]. MEC offers a platform for executing applications and services, aggregating and storing localized data, at the network edge. This enables the development of context-aware applications [33], [34]. 5G applications are anticipated that will utilize MEC, including:
Video analysis. One significant advantage of edge computing is the ability to deal with video data directly from UE, such as cameras, mobile phones, and cars. MEC possess sufficient processing capabilities, improves the efficiency of transitioning, greatly decreases network bandwidth pressure, and enables end users to make quicker decisions in crucial scenarios [35].
Smart home and city. Edge computing has proven highly beneficial in managing and coordinating devices for smart homes and cities. This is achieved by executing computations and caching the data locally so it significantly reduces response times for these devices [36].
Healthcare. Edge computing in hospitals is designed to efficiently prioritize critical traffic, expediting computationally intensive tasks like compressing and decompressing medical surgery images, reducing the performance burdens, and enhancing security, all while capturing, analyzing, and combining essential information [37].
MEC provides functionality that can reduce latency and operational expenses, and enhance end user experience. However, it also introduces novel security and energy consumption issues. The massive amount of data collected in real-time from mobile devices and challenges with storing and processing this data on edge servers may result in data privacy and security risks. The ongoing collection, aggregation and communication of data between edge servers and mobile devices results in significant energy consumption, posing an energy efficiency challenge. The challenges related to constructing edge-sliced network solutions [38], [39] provide future research opportunities.
MEC facilitates the incorporation of different technological frameworks, including network slicing, NFV, and SDN, to enhance the features and capabilities available to be consumed by connected UE. The adoption of the NFV architecture enables the implementation of an MEC system through the association of MEC components with VNFs. The infrastructure facilitated by MEC-NFV enables the delivery of network slice requests. The network slicing paradigm is based on the concept of NFV, which involves virtualizing an abstract of the network, thereby providing flexible infrastructure. The SDN paradigm facilitates programmable coordination of resources to enhance the dynamic configuration of network resources for the provision of MEC-enabled services [40].
ETSI established a comprehensive structure and architecture for MEC [28]. Fig. 2 illustrates the high-level MEC framework. The MEC system is composed of two tiers. The first tier operates at the MEC host level, which encompasses MEC hosts, MEC Platform Manager (MEPM), and Virtualization Infrastructure Manager (VIM). The second tier operates at the MEC system level, which includes the MEC Orchestrator (MEO). Resources can be located at MEC hosts. Nevertheless, MEO coordinates and manages these resources. MEC host is a physical unit with a Network Function Virtualization Infrastructure (NFVI) offering computational, storage, and memory resources and networking capabilities. The MEC platform will employ MEC hosts to assist MEC applications that use the virtualized resources provided and supervised by the NFVI and the VIM, respectively. The MEPM oversees the many parts of the MEC host, such as the lifecycle of Mobile Edge Computing Platform (MEPs) and MEC applications. A single MEC host is responsible for managing a single MEP, although a single MEPM may supervise several MEC hosts within the MEC host-level administration. The MEO integrates multiple MEPMs to facilitate system-level management, offering a comprehensive overview of the entire MEC system. The primary duty of MEO is to deploy MEC hosts to conduct status assessment of the available resources, services, and existing topology. The MEO has the capability to initiate and conclude the process of creating or moving applications [41].
![FIGURE 2. - ETSI MEC framework [28].](/mediastore/IEEE/content/media/6287639/10380310/10792437/grego2-3515077-large.gif)
By utilizing NFV architecture components, MEC systems can be implemented more effectively, facilitating the initial MEC deployment. It is possible to connect various MEC components, such as MEP, MEC App, and MEPM, with VNFs. The blueprint for implementing MEC within the NFV architecture is provided by ETSI, as outlined in [28]. Within the MEC-NFV framework, the role of the MEO is replaced by an MEC Application Orchestrator (MEAO) to facilitate communication with the NFV Orchestrator (NFVO). The orchestrators play a crucial role in coordinating and managing resources, working together to handle tasks like lifecycle management, transfer, and termination of MEC applications. The primary role of the MEAO is to coordinate the deployment VNFs and ensure the reliability of the MEC Platform for each instance of an MEC application. The NFVO is responsible for handling and controlling service lifecycle, and it interacts with the MEAO [41].
Conversely, the SDN paradigm facilitates the management of network resources and enables the MEC system to access a broader range of network information. SDN enables the segregation of the data and control plane, facilitating personalized traffic routing. To meet the low latency requirement, the SDN controller aids in the operation of MEAO by providing comprehensive network status information. SDN facilitates MEC mobility, service migration management, and bandwidth control inside the 5G and future mobile network slicing architecture [17]. OpenFlow is the predominant SDN networking protocol used to manage network traffic flows, and it is managed by the Open Networking Foundation (ONF) [42].
To further clarify the role of MEC within the network slicing framework, it is important to explore its involvement across the various phases of the network slicing lifecycle: preparation, commissioning, operation, and decommissioning.
Preparation: In this phase, MEC integrates with the NFV architecture to enable resource abstraction and virtualization. The MEO evaluates available resources and prepares MEC hosts for deployment by coordinating with the VIM and MEPM [41]. This process involves defining application requirements and aligning them with the available computational, storage, and networking resources within the MEC system [28].
Commissioning:During this phase, MEC components, such as applications and platforms, are deployed onto the virtualized infrastructure. The MEPM manages the lifecycle of MEC applications, ensuring correct instantiation on MEC hosts [28]. The MEC-NFV framework allows VNFs to be deployed at the edge, enabling services to meet latency and bandwidth requirements [41].
Operation: This phase involves the active coordination and management of MEC resources. The MEO monitors and dynamically adjusts resource allocation based on real-time network conditions and application demands [41]. MEC hosts facilitate low-latency, high-performance services by running applications closer to end users. Functions such as application migration, scaling, and optimization are performed to maintain service quality and meet performance metrics [17].
Decommissioning: In this phase, MEC applications and associated VNFs are gracefully terminated or migrated as service requirements change or resources need reallocation. The MEPM and MEO coordinate the shutdown of services, ensuring resources are freed for future use without impacting other network functionalities [41].
Network Slicing Definition
Network slicing is an aspect of network virtualization that facilitates the transition from a fixed to a more flexible network infrastructure. It allows for establishing multiple independent logical networks, referred to as network slices, using a shared physical infrastructure. The objective is to adapt the current networks based on the application’s requirements and transition from a uniform approach to a more logical solution, which involves creating separate slices with allocated resources, isolation, and specific applications [43]. Network slicing may be executed in various domains, including core networks, transport networks, and RAN, by leveraging cloud computing infrastructure or alternative network resources and functions. Each network slice has unique properties and is dedicated to serving a single service with distinct characteristics and requirements [44].
The service requirements have been categorized by the International Telecommunication Union (ITU) into three groups: URLLC, eMBB, and mMTC. Mission-critical applications, like remote surgery and autonomous driving, fall under the first group. These applications require extremely low latency, measured in sub-milliseconds, with an error rate of less than one packet loss in every 105 packets [45]. The second category encompasses applications that require a significant volume of data, including virtual and augmented reality, as well as extensive video streaming. The third category refers to providing connectivity for several devices that experience sporadic traffic, such as smart metering, sensing, and monitoring applications, where the significance of latency and throughput is not vital.
By controlling the duration of network slices, SDN and NFV facilitate network slicing. Network slicing allows for the dynamic generation, modification, and termination of network slices through the use of the SDN or NFV orchestration architecture. A slice comprises a series of interconnected VNFs that enable the specific service offered to the end-user of the slice [46], [47]. In addition, there are certain crucial prerequisites that must be fulfilled while implementing network slicing. The requirements can be summarized as isolation among slices, customization, and efficient resource utilization [48].
Robust network slicing isolation is a crucial prerequisite for effectively running several simultaneous operations on a common, shared foundation. The concept of isolation must be comprehended in relation to:
Performance: Each segment is precisely determined to fulfill certain service demands, typically indicated by key performance indicators. Performance isolation is a comprehensive concern that must guarantee the fulfillment of service-specific performance criteria on every slice, irrespective of congestion and performance levels on other slices [49], [50].
Security and privacy: It is imperative that any attacks or flaws that arise in one slice do not have any negative effects on the others. Additionally, it is necessary for each slice to possess autonomous security mechanisms that prohibit unauthorized entities from obtaining the ability to read or write configuration, management, or accounting data that is specific to a given slice. Furthermore, these mechanisms should be capable of documenting any such attempts, regardless of their authorization status.
Management: Every slice needs to be autonomously administered as an independent network [51].
To accomplish isolation, it is necessary to establish suitable and consistent policies and methods at every level of virtualization in accordance with the previously mentioned recursion principle. The policies outline sets of regulations that specify the necessary measures for effectively segregating various controllable entities without providing specific details on the methods to accomplish this. The mechanisms refer to the specific procedures and methods that are put into action in order to ensure compliance with the established policies. To achieve the necessary level of isolation, it is essential to combine both virtualization and orchestration [47].
A. Software-Defined Networking
SDN plays a crucial role in facilitating network virtualization. SDN separates the control plane from the data plane, allowing for centralized network control through a centralized controller, or a hierarchy of distributed controllers. The data plane entities are simplified packet forwarding devices that receive forwarding choices from the centralized controller and store flow control rules in flow tables. Network programmability is facilitated by separating the control plane from the data plane [52], [53]. In other words, the software applications, also known as network apps, that operate on the SDN Controller [54] determine the way data is forwarded. The forwarding behavior is converted into forwarding rules, which are then distributed to and implemented on the packet forwarding devices. The SDN controller communicates with the packet forwarding devices via the Southbound API, also known as the resource control interface. The ONF created the OpenFlow protocol, also known as the OpenFlow API, which the controller frequently uses to establish communication with the data plane devices [55], [56], [57].
The initial vision and specification of the SDN architecture included features and functions, specifically focusing on the orchestration capability [58]. The data plane is commonly denoted as consisting of “resources” within such circumstances.
B. Network Functions Virtualization
Network slice orchestration necessitates the virtualization of network resources, infrastructure, and network functions, which are then assigned to individual slices [46]. The implementation of network resource virtualization, infrastructure virtualization, and network function virtualization necessitates new management paradigms that can effectively handle tasks such as lifecycle management of VNFs, management of virtualized infrastructure, and monitoring of infrastructure operations. The ETSI has established a reference architecture for NFV to facilitate the widespread implementation of virtualized systems and the seamless collaboration between various functional entities inside these systems [59], [60].
The NFV framework provided by ETSI is anticipated to significantly impact the deployment of network slices in real-world scenarios. The system has the ability to coordinate, control, and carry out the functions of the virtualized network. VNFs are software-based network functions that can be executed on the NFV infrastructure. The NFV infrastructure is responsible for abstracting physical resources into virtualized resources, such as computing, storage, and network resources. The virtualized resources are then provided to the VNFs for execution and operation. The coordination and control of hardware and software resources fall within the responsibility of the NFV control and Orchestration (NFV-MANO) component. The NFV-MANO is responsible for the management of VNFs. The NFV architecture facilitates functionalities for the dynamic coordination and efficient administration of virtualized networks. Fig. 3 provides a depiction of the functional level of the NFV framework [61], which comprises three primary elements:
Network Function Virtualization Infrastructure Hardware resources and virtual instantiations constitute the NFV foundation platform, establishing the infrastructure for deploying, managing, and executing VNFs. Cloud providers use the NFVI as part of the Infrastructure-as-a-Service (IaaS) cloud to set up Virtual Data Centres (VDCs), which function like a data center and include virtualized computing, storage, and networking [63]. NFV providers receive the VDCs to provide customers with network services. It is important to keep NFV providers’ resources inside a VDC separate from one another. This Isolation allows NFV vendors to securely utilize the shared cloud infrastructure. VNFs are set up among virtual machines in a VDC using NFV services. There are three main domains to the NFVI: compute domain, infrastructure networking domain, and hypervisor domain. The computing domain provides the physical computing and storage resources required to support the deployment of VNFs by providing processing and storing capabilities. The physical and virtual networking components, such as switching and routing, are included in the infrastructure networking domain. The components are responsible for connecting the compute and storage resources within the NFVI. The virtualization layer is located above the hardware resource layer and comprises the hypervisor, which is known as the Virtual Machine Monitor (VMM). Simulating hardware resources, assuring isolation between virtual machines, and separating virtual resources from physical resources are the three primary functions of the VMM [63], [64].
Virtual Network Functions VNFs are software packages that can be placed on the NFVI and facilitate the functioning of traditional, non-virtual network services. An individual VNF has the capacity to encompass multiple internal components, including residential gateways, packet data network gateways (PGW), firewalls, and other similar elements, which reduces management and deployment complexity. A VNF may consist of a single component to enhance scalability, reusability, and responsiveness. It should be noted that multiple VMs could share a single VNF. Typically, telecommunications service providers offer virtual network services consisting of multiple Virtual Network Functions tailored to meet users’ requirements. Each VNF is directly linked to an Element Management System (EMS) responsible for executing standard management tasks for the attached VNF [65], [66].
NFV Management and Orchestration NFV management and orchestration handle the administration and orchestration of virtualization specific duties needed for the entire lifespan of the VNF, from bundling several services into a single VNF package to linking this service to consumers when requested. MANO manages errors in VNFs and maintains state information for service VNF. In addition, the MANO establishes communications between several VNFs that are forming a network graph. A collection of VNF services forms a network graph that works together to deliver a selected service to the consumer by creating a virtual network using the existing VNFs. It consists of three management units: (i) VMI, which supervises and controls the VNFs’ connection with the NFVI resources; (ii) VNFM, which manages and monitors the VNFs using the EMS and provides resources and the state of the data traversing between the orchestrator and the VIM; and (iii) The orchestrator that supervises service lifespans in the network, and consists of functions including policy administration, performance management and the instantiation process [65].
![FIGURE 3. - NFV reference architecture framework - functional level [61].](/mediastore/IEEE/content/media/6287639/10380310/10792437/grego3-3515077-large.gif)
Understanding the lifecycle of network slicing phases: preparation, commissioning, operation, and decommissioning provides valuable context for its implementation.
Preparation: This phase begins with designing and defining slice templates based on service-specific requirements, such as URLLC, eMBB, or mMTC. SDN and NFV architectures enable the virtualization of physical network resources into isolated slices. Key performance metrics, such as latency and throughput, are identified, and the orchestration framework ensures the necessary infrastructure is in place for slice instantiation [43], [46].
Commissioning: Slices are instantiated using the NFV Management and Orchestration (NFV-MANO) framework. VNFs are deployed to form slices tailored to specific service demands. SDN controllers coordinate resource allocation, ensuring proper configuration and isolation of slices. This phase involves initial testing and validation to ensure the slice is ready for operational use [46], [47].
Operation: This phase focuses on maintaining slices and ensuring their performance meets predefined service-level agreements (SLAs). Dynamic management, such as scaling, load balancing, and fault handling, is performed by orchestrators (e.g., VIM, VNFM, and SDN controller). The system adapts to real-time traffic demands and ensures efficient resource utilization while maintaining isolation between slices [47], [65].
Decommissioning: Slices are dynamically modified or terminated as service needs evolve. SDN and NFV frameworks handle the termination of VNFs and the release of associated resources. This process ensures minimal disruption to ongoing services and prepares the network infrastructure for new slice deployments or reconfigurations [47], [65].
Integration of Network Slicing and MEC
A network slice refers to a VNF Forwarding Graph which can be created by a collection of VNFs. A network slice is created to fulfill specific network attributes, such as extremely low latency and high reliability. The MEC system enhances the network slice by incorporating various VNFs, including other MEC capabilities and functions into a single service. From an enterprise perspective, a Service-Level Agreement (SLA) can be used as a commercial contract that could define what constitutes a network slice. The SLA establishes a contractual agreement between two parties, typically the tenant and the service provider, by specifying availability, service quality, and other conditions. A network slice is a fundamental component of the 5G network. Incorporating MEC into the 5G network provides added advantages for implementing network slicing, including the assurance of end-to-end network latency [41], [67]. Within the 5G architecture, the mobile equipment provider is referred to as a third-party application known as the 5G Application Function. Network operators authorize the MEC platform to access the operator’s core network as an application function. This can be done through direct interaction or by employing another application interface. Kekki et al. [31] provide an example of MEC integration in 5G networks.
The 3GPP provided clarification on the deployment of MEC and its seamless integration into 5G based on certain factors and the established functional enablers. Fig. 4 depicted the network functions specified in the 5G architecture, and their respective duties can be succinctly described as follows [61]:
Access and Mobility Management Function (AMF): implements protocols and methods for managing connectivity and accessibility, such as handling connections, managing reachability, notifying mobility events, terminating the control plane of the RAN, and ensuring authentication and authorization.
Session Management Function (SMF): executes operations associated with session management, such as establishing sessions, terminating connections using policy control features, and notifying about downlink data.
Network Slice Selection Function (NSSF): performs the distribution of slicing resources and assigns an AMF to provide services to users.
Network Repository Function (NRF): facilitates the identification of network functions and services.
Unified Data Management (UDM): manages user subscriptions and service identification.
Policy Control Function (PCF): integrates network policies and offers policy rules for controlling plane functions.
Network Exposure Function (NEF): functions as a service-aware border gateway to ensure secure connectivity with the network functions’ supported services.
Authentication Server Function (AUSF): executes authentication protocols.
User Plane Function (UPF): offers features that improve user plane operations, including packet routeing and forwarding, data buffering, and IP address allocation.
![FIGURE 4. - 5G architecture-network functions [62].](/mediastore/IEEE/content/media/6287639/10380310/10792437/grego4-3515077-large.gif)
The SMF establishes a connection with the 5G UPF, which may be accessible at an aggregation point. Alternatively, the AMF establishes a connection with the 5G access network Centralized Unit (CU). The CU subsequently engages with the Distributed Unit (DU), which caters to the UE. The 3GPP defines the 5G core system, and ETSI specifies how it interacts with the MEC system [31]. Data centers are comprised of a cluster of computers and storage that offer a substantial quantity of resources for processing and data storage. The remote cloud bears comparable responsibilities to data centers. The primary distinction between distant cloud and data centers is evident from a commercial standpoint. Organizations maintain data centers privately, whereas remote cloud infrastructure can be supported by cloud solution providers, e.g., Amazon Web Services and Microsoft Azure. The SDN controller and NFVO can be effectively incorporated into the core network to establish logical connections with the MEC systems. The SDN controller controls the transmission of data packets and the observation of the network, enabling it to retrieve the current state of the entire system [68]. The NFVO controls the VNF instance lifespan and considers MEC applications within the MEC systems to be VNF instances. An MEC system can consist of one or more MEC hosts, and the MEO controls the MEC hosts. MEC systems are interconnected by the MEO. In the context of 5G, the MEO is seen as a 5G application function responsible for overseeing and controlling MEC operations in a centralized manner. The MEO, functioning as an access function, subscribes to the 5G NEF and obtains network-related data for monitoring and managing the MEC system. The MEO has the capability to make resource allocation decisions by utilizing information obtained from the 5G NEF. The MEO can be deployed within a cloud network as a fundamental component of the core network or in conjunction with network aggregating points and 5G UPF. In 5G, the network aggregation point is responsible for monitoring the access and core networks.
The communication links within the edge network can be categorized into two components: user access and inter-access points within the access network or the network edge. The user initiates a connection with the access point via the user access link, typically established by radio or wireless technology. The inter-access point link establishes connectivity between various access points, encompassing both mid-haul and front-haul connections. The inter-access point can be established using several mediums, e.g., copper cables (Ethernet or HFC), fibre optics or wireless.
The 5G access network consists of gNodeBs (gNBs) serving as access points with extensive coverage and occasionally incorporates Road Site Units serving as access points with narrower coverage. The 5G central unit and distributed unit are components of the next-generation base stations gNBs. ETSI has established multiple options for integrating MEC hosts into the 5G network [31]. In this regard, the MEC host may be positioned either alongside the gNB (base station) or at the network aggregation point where the core and access networks are connected. The radio network information service provides the radio network status along with information that an MEC platform within an MEC host can use for subscription. The MEC platform has the capability to observe the condition of UE and MEC apps and provide the latest information to the mobile edge operator.
A task is a service request that is typically initiated by a UE. It typically comprises three characteristics: the amount of data that needs to be processed, the quantity of CPU processing cycles required for computation, and occasionally the highest delay permissible. Users are linked to distinct tasks. If the service currently exists in the system, the task assignment will be immediately sent to the service that is available. If the service is currently unavailable, a MANO unit, such as MEO, NFVO, or any orchestrator, will assemble the service components. The NFVI virtualizes the hardware resources by abstracting them into virtualized resources, VNFs, and interconnected to produce a chained network function. The service function chaining example in Fig. 5 consists of a sequence of VNFs, specifically VNF1-VNF2-VNF3-VNF4, which collectively represent a distinct service [41].
![FIGURE 5. - 5G MEC, NFV-and-SDN enabled in resource allocation scenarios. [41].](/mediastore/IEEE/content/media/6287639/10380310/10792437/grego5-3515077-large.gif)
Resource Allocation
Resource allocation in 5G networks with MEC and network slicing is critical for performance and efficiency optimization. In the context of 5G, resource allocation involves efficiently distributing resources such as spectrum, time allocation, and power to support the diverse requirements and objectives of different applications and services. It is particularly important in MEC and network slicing, where resources must be allocated dynamically to different network slices and edge computing instances based on their specific needs and performance metrics. Research is being carried out to address the challenges and complexities of resource allocation in 5G networks [60], [69]. The studies have examined existing state-of-the-art challenges, the significance of resource allocation, techniques, existing rules, and algorithms employed for resource allocation, as well as the parameters and metrics taken into account when resource allocation occurs in 5G networks [23].
The integration of MEC and network slicing in 5G introduces new considerations for resource allocation, such as dynamic allocation of computing and storage resources at the network edge and isolation and management of resources for different network slices. As 5G continues to evolve, ongoing research and development efforts are focused on designing advanced resource allocation schemes that can effectively cater to MEC-integrated 5G networks’ diverse requirements, ultimately enabling enhanced service quality and improved overall network performance [70].
Network slicing is an essential technology for 5G that plays a crucial role in fulfilling diverse service requirements. Ensuring service isolation while dynamically allocating RAN bandwidth resources to guarantee service satisfaction presents issues [71]. Experimental RAN slicing scenarios have been examined in [46], [72], and [73] to facilitate the flexible distribution of control functions between the centralized controller and the decentralized agents. The systems aim to provide a cost-effective approach for operating the next-generation radio access network as a VNF. Nevertheless, although progress has been made in the realm of RAN virtualization, the idea is still deficient in terms of a slicing architecture. The authors of [46] have also suggested that a RAN slicing solution implemented on the FlexRAN platform ensures functional separation between slices. For a comprehensive analysis of Functional Split (FS) optimization, refer to the works of Wizhaul [74] and FluidRAN [75]. The studies propose a combined approach to routing and FS optimization in order to maximize the network’s centralization, which refers to the placement of NFs at the CU based on the available network resources. Similarly, FluidRAN [75] adopts a similar approach but with the specific goal of minimizing monetary costs. Nevertheless, despite their perceptive results, both papers disregard the RAN slicing option. The authors of [76] and [77] introduce a combined approach to routing and FS optimization, taking into account several slices. Previous research has not included dynamic RAN slicing with MEC, which takes into account the effects of MEC placement inside the O-RAN design on both throughput and network costs. Additionally, these studies have not assessed the costs associated with creating slices or the influence on UE quality of experience.
The introduction of a share-constrained proportional resource allocation mechanism in [78] aimed to achieve functional network slicing. This technique allowed users to profit from sharing while still being able to customize the resource allocation in order to maximize usefulness. In [79], the authors formulated the issue of resource allocation for network slices into a biconvex issue, taking into consideration for slice coordination and RAN vertical slicing concerns. In their study, Jia et al. [80] introduced a network slice resource allocation method that utilizes a bankruptcy game approach. This algorithm aimed to optimize the use of the limited radio resources and guarantee fair allocation. Research has been carried out to examine the issue of resource distribution and pricing for network slices [80], [81], [82]. The issue of allocating resources in network slicing was addressed by using a generalized Kelly mechanism (GKM) and utilizing Karush-Kuhn-Tucker (KKT) conditions [80]. Wang et al. [81] introduced a distributed algorithm that tackled the issue of resource allocation in network slicing by considering the benefits of optimization and resource efficiency together. The solutions were suggested without taking into account the entire network slicing lifespan. To optimize the use of system resources, it is possible to foresee the requirements of slices, such as resource needs, in order to effectively handle Services that are experiencing dynamic changes [83], [84]. In their study, Bega et al. [83] introduced a sophisticated neural network structure that provided a precise estimation of future capacity needs, taking into account the associated costs. This forecast may be readily utilized by operators to make informed decisions regarding the reallocation of resources, both in the short and long term, to maximize their revenues.
Nevertheless, the precision of forecasting relies on numerous control parameters, such as the scale factor. Reinforcement learning is utilized in research to address resource allocation challenges, as evidenced by [85] and [86]. Aijaz et al. [85] proposed a new framework for dividing radio resources, using haptic communication and an efficient slicing mechanism that utilizes a reinforcement learning technique. In [87], a study of resource distribution methodology using a deep reinforcement learning-based autonomous slicing refinement algorithm is provided. Also, a shape-based heuristic approach is developed to optimize resource consumption for user resource customization. Deep reinforcement learning is considered an interesting approach for addressing the management issue of resources that are demand-aware in network slicing. This is because it enables controllers to adapt their operations based on changing demands.
The authors in [32] introduced a two-phase algorithm to achieve optimized resource allocation and slice pricing in MEC-integrated network slicing environments. In the first phase, They introduce the Resource Pre-Allocation Algorithm (RPAA), which decomposes the resource allocation problem into sub-problems for different types of resources: communication, computing, and caching. This stage aims to minimize costs while ensuring Quality of Service (QoS) for each user device. Using methods like Branch-and-Bound for uplink communication and computing resource allocation and the Interior Point Method for downlink communication, the RPAA iteratively solves for optimal allocations within the constraints of latency and bandwidth requirements. This process ultimately provides an optimized baseline resource allocation cost for each slice. Then, in the second phase, they involve the Network Slices Pricing Algorithm (NSPA), which models a Stackelberg game between the MEC Network Service Provider (NSP) and user devices. The MEC-NSP sets slice prices to maximize its profit, while the user devices respond by selecting slices that balance price and QoS. The NSPA uses the Optimal Response Dynamic Method (ORDM) to iteratively adjust slice prices and user choices until reaching a Nash equilibrium, representing a stable state where neither the MEC network service provider (MEC-NSP) nor the user devices have an incentive to change strategies. This equilibrium ensures that resource costs and user satisfaction are balanced, providing an efficient and profitable network slicing model.
In another study [4], the authors present a cost-minimization algorithm designed for a Slice Broker (SB) in a 5G MEC environment, where the SB operates between Infrastructure Providers (InPs) and Slice Tenants (STs). The algorithm addresses the challenge of resource allocation by using a Mixed Integer Programming (MIP) formulation to minimize the costs associated with leasing resources from multiple InPs while meeting the service requirements of the STs. The SB procures both computing and network resources from InPs and allocates these resources to create virtual network slices tailored to ST needs. Each slice is modeled as a Service Function Chain (SFC), which consists of interconnected VNFs requiring specific amounts of computational power, storage, and network bandwidth. By solving the MIP problem, the SB selects the optimal configuration for resources and allocates VNFs to minimize costs while respecting QoS requirements. Also, it introduces predefined resource configurations that allow the SB to select from various combinations of vCPU, storage, data rate, and delay levels, each associated with a cost. This approach avoids the need for negotiation, simplifying resource allocation in the multi-domain 5G MEC environment. Ma et al. in [39] introduced an Intelligent Scheduling and Management System (ISMS) to address joint MEC selection and wireless resource allocation within a 5G RAN environment. The approach frames the problem as a Mixed-Integer Nonlinear Programming (MINLP) challenge and applies a modified Deep Deterministic Policy Gradient (mDDPG) algorithm. The ISMS uses reinforcement learning to optimize MEC selection and UE power allocation, balancing system energy consumption, latency, and user costs. The state space includes user channel gain and task information, while actions involve task offloading choices, power allocation, and offloading ratios. By reaching a Markov decision process-based equilibrium, the algorithm achieves efficient resource allocation across the network. Simulation results show that ISMS outperforms traditional local computing and greedy partial offloading schemes, providing more stable and effective performance by reducing energy consumption and latency, thus offering a reliable solution for dynamic 5G MEC environments.
There has been considerable interest in the issue of network slicing in multi-tenant cellular networks. The study in [89] investigates two tiers of dynamic network slicing within a heterogeneous C-RAN. The higher level is responsible for RRH association, the control of user acceptance, and baseband unit (BBU) capacity allocation. Additionally, at the lower level, power and physical resource block are distributed. RAN slicing which is implemented by deep reinforcement learning in [91], is examined for the fog RAN (F-RAN) system and it is implemented by deep reinforcement learning. The authors of [92] and [93] discuss the deployment of RAN slicing across several operators using dedicated physical resource infrastructure. The authors introduce a framework named O-RANFed in [90] to support 5G slicing services by executing and enhancing federated learning tasks on O-RAN devices. The authors of [94] provide a federated deep reinforcement learning technique for implementing network slicing in O-RAN.
O-RAN is an ideal replacement for RAN because of its adaptability, intelligence, cost-effectiveness, and transparency. O-RAN was created to leverage the advantages of both C-RAN and virtual RAN simultaneously. Visualizing RANs enables operators to enhance flexibility, reduce capital expenditure and operating expenses, and swiftly incorporate new functionalities into their networks. The C-RAN architecture partitions the RAN into two primary components: the BBU and the RRH. A centralized BBU pool may be utilized to connect multiple RRHs [95]. O-RAN partitions the RAN into three separate entities: the Distributed Unit (O-DU), the Central Unit (O-CU), and the Radio Unit (O-RU), in contrast to C-RAN. The baseband processing which operates in the O-CU layer is real-time whereas in the O-CU layer, it is mostly non-real-time. Within the O-RAN framework, the physical layer is segmented into high and low physical components, which differs from the C-RAN approach. Fig. 6 illustrates the structure of an O-RU, which consists of a low physical layer and the RF. The RF module sends and receives signals via radio, whereas the low physical layer incorporates digital beamforming. The O-DU is usually a logical component with advanced capabilities regarding the RLC, physical layer, and MAC. It is a sub-function of the eNodeB and is placed in close proximity to an O-RU. It is linked to the O-RU using an open fronthaul interface. O-CU supports both the lower layers and the upper layers of the protocol stack [88]. The O-CU is comprised of two components: the O-CU control plane (O-CU-CP) and the O-CU user plane (O-CU-UP). the O-CU-CP contains the Radio Resource Control and the PDCP-Control Plane, while The O-CU-UP contains the Service Data Adaptation Protocol and the Packet Data Convergence Protocol (PDCP) User Plane. O-CU and O-DU are linked through an open interface, F1. The O-CU-UP is linked to the UPF by an O-backhaul connection. The O-RAN architecture includes additional key logical components known as RAN Intelligent Controller (RIC) - Near Real-Time, O-Cloud, Orchestration and Automation. Orchestration and Automation encompass features, e.g., RIC non-real-time. The RIC oversees the use of machine learning techniques and enhancing the system’s intelligence. The main characteristic of the O-RAN architecture is the separation of hardware and software, which enables NFV. Each component is built as a VNF, which is a functional unit in NFV and is able to be implemented on a container or a Virtual Machine [96]. Therefore, as depicted in Fig. 6, O-RAN elements, including UPF, O-DU, O-CU, RIC-near real-time, and UPF are virtualised and executed as VNFs [97], [98].
![FIGURE 6. - Network sliced O-RAN system [88].](/mediastore/IEEE/content/media/6287639/10380310/10792437/grego6-3515077-large.gif)
To enhance clarity and summarize the key contributions of the reviewed works on resource allocation in 5G/6G networks with MEC and network slicing, we have included Table. 3. This table provides a comprehensive overview, categorizing the studies based on their focus areas, the specific problems they address, the methodologies or algorithms employed, and their key contributions. It serves as a quick reference, highlighting major challenges such as cost minimization, fairness, and demand-aware slicing, along with the state-of-the-art techniques used to address them, including proportional resource allocation, reinforcement learning, and neural networks. Additionally, the table summarizes the significant outcomes and contributions of these studies, such as utility maximization, efficient resource management, and enhanced flexibility in multi-tenant environments.
Challenges
While significant progress has been made in resource allocation, unresolved challenges remain that necessitate exploration.
Reducing Latency Latency minimization is one of the most challenging issues in resource allocation. Increasing the number of base stations may raise the frequency of transmission delays. It is crucial to investigate the scheduling delay and transmission impact since these factors might significantly influence the suggested strategies for real-time processing capacity. Consider the trade-off between delay and performance is also crucial while coding across numerous fading blocks.
Energy conservation Energy efficiency is another important factor that needs to be taken into account. It is crucial to assess and analyze the balance between an application’s performance and the allocation of power as a power-saving feature on cellular devices. Furthermore, evaluating the efficiency of beamforming algorithms on a broad scale requires increased focus. Utilizing renewable resources for energy can enhance the energy efficiency of ultra-dense C-RANs. It is crucial to research effective RRH switching-off strategies to reduce energy usage in low-traffic circumstances.
Mobility Management Mobility management is another crucial element that should be considered. Providing consistent and robust connectivity across different cellular communication technologies is essential for advancing autos. It is important to analyze the effectiveness of operations and enhanced algorithm designs with minimal complexity according to the needs of network operators or users. Future research will need to focus on developing mobility-aware adaptive strategies to optimize extreme patterns of identical base stations in a coverage region due to mobile call correlation [5].
Network Virtualization Exploring wireless network virtualization is essential for enhancing end-to-end performance. Communication with only one person in a virtual cell is inappropriate. This will cause interference while approaching other users. It is crucial to explore dependable virtualization approaches in order to benefit from reduced interference through multi-user cooperative transmission. Network slicing solutions should be explored that provide 5G diverse services with highly reliable and low latency communication, improved mobile broadband and extensive machine-type communication [99].
Network Scalability In addition to the earlier challenges, there is another consideration that requires attention which is network scalability. There has always been a requirement for enhancing the Channel State Information (CSI). While the stochastic beamforming approach has been previously explored to reduce excessive CSI acquisition, it necessitates a more efficient algorithm for extensive networks. Furthermore, enhancing the uplink compression techniques can increase the overall capacity of the sum rate. Heuristic algorithms need to be created for efficient infrastructure deployment and layout planning. Heuristic methods are required to improve time efficiency with increased focus on reducing the intricate obstacles related to network scalability [100], [101].
Conclusion
This paper comprehensively reviews the recent advances in 5G/5GB and 6G with network slicing and MEC. Researchers propose numerous strategies, although they are still in the early stages of development. This paper discussed the MEC and network slicing framework and architecture, as well as current research and proposals for resource allocation in the RAN, which provide optimized and dynamic resource allocation. This paper explores the concept of merging MEC and network slicing in order to maximize their benefits. The integration of MEC and network slicing will bring a new era of profitability to resource allocation in the RAN and optimize the consumption of resources. Challenges have been identified that could be resolved by solutions based on network slicing and MEC depending on the availability of resources for custom network slices. MEC is a promising technology that can augment existing networks to address complicated scenarios and improve service provision thereby improving consumer quality of experience. Network slicing has led to the emergence of new services, each with its own distinct network requirements.