A Survey of Quality-of-Service and Quality-of-Experience Provisioning in Information-Centric Networks

Abstract

Information-centric networking (ICN) is a promising approach to address the limitations of current host-centric IP-based networking. ICN models feature ubiquitous in-network caching to provide faster and more reliable content delivery, name-based routing to provide better scalability, and self-certifying contents to ensure better security. Due to the differences in the core architecture of ICN compared to existing IP-based networks, it requires special considerations to provide quality-of-service (QoS) or quality-of-experience (QoE) support for applications based on ICNs. This paper discusses the latest advances in QoS and QoE research for ICNs. First, an overview of ICN architectures is given, followed by a summary of different factors that influence QoS and QoE. Approaches for improving QoS and QoE in ICNs are then discussed in five main categories: in-network caching, name resolution and routing, transmission and flow control, software-defined networking, and media-streaming-based strategies. Finally, open research questions for providing QoS and QoE support in ICNs are outlined for future research.

1. Introduction

The original intent behind the development of the Internet was to facilitate peer-to-peer communication and the sharing of data and resources. However, over time, there has been a significant shift towards content dissemination, driven by the exponential growth in networked devices, the increasing demand for content sharing, and the widespread usage of multimedia applications. Cisco’s annual Internet report [1] projected that by 2023, the number of devices connected to IP networks would exceed three times the global population. Despite the Internet’s remarkable success in enabling universal connectivity, its current architecture is not well equipped to handle the immense volume of network traffic generated by contemporary usage patterns, which heavily emphasize multimedia content distribution. Activities such as video streaming, imaging applications, surveillance, social networking, and cloud services dominate Internet usage, placing significant strain on an infrastructure originally designed for end-to-end communication.

Another critical obstacle the current Internet architecture faces is the widespread use of mobile devices, wireless connections, and resource-constrained edge nodes like sensors and Internet of Things (IoT) devices. Traditional IP-based protocols, built for stable, fixed networks, struggle with dynamic wireless environments where mobility, link failures, and power constraints are common. The dual role of IP—as both a locator and an identifier—binds the network and transport layers, making seamless mobility difficult. When nodes move, sessions often break and must be re-established. Additionally, IP’s role as the protocol stack’s “narrow waist” limits content delivery efficiency, as it offers only best-effort service without built-in reliability, ordering, or state management.

To overcome the current Internet’s limitations, researchers have proposed clean-slate designs like information-centric networking (ICN), which has gained global traction [2]. ICN addresses key challenges such as data scalability, bandwidth efficiency, and security. Networking in ICN operates on an interest-driven model, meaning content is only transmitted when requested by a consumer. Conversely, publishers advertise their content to the network without specific knowledge of potential consumers, relying on the network to match interests with available content. Unlike traditional networking paradigms, ICN does not rely on identifying the locations of publishers or consumers; instead, it focuses on matching interests with content. This approach, combined with ICN’s in-network caching capabilities, promises to deliver content with performance comparable to Content Delivery Networks (CDNs) without requiring dedicated proprietary resources.

Furthermore, today’s Internet landscape is dominated by data-intensive services like multimedia streaming, online gaming, surveillance, wireless imaging, and teleconferencing. In addition, cutting-edge technologies such as the metaverse, virtual and augmented reality (VR and AR), holographic displays, and digital twins demand high-resolution, low-latency video content to deliver seamless interactions and foster a sense of presence within virtual environments. Enhanced video quality improves realism, user engagement, and comprehension, essential for AR/VR training simulations and collaborative digital twin environments [3]. In the metaverse, it enables expressive social interaction through clear gestures and facial cues [4].

All these multimedia-focused applications require satisfactory quality of service (QoS) and quality of experience (QoE), which ICN is well suited to support, leveraging features like in-network caching and multi-path transmission. It also effectively handles the massive, continuous data flow from IoT devices using its naming scheme and built-in data routing mechanisms [5].

While many survey papers explore the Future Internet domain, as described in the next section, none has examined the QoS and QoE dimensions within the realm of ICN. Our objective is to thoroughly analyze and compare the methodologies deployed in information-centric networks that specifically focus on enhancing QoS or QoE performance.

This article is structured as follows. The relevant existing works are discussed in Section 2. We describe the key similarities and differences among different ICN architectures in Section 3. The various factors that contribute to the QoS and QoE are explained in Section 4. Different ICN approaches that target QoS and QoE improvement are categorized and analyzed in Section 5. Section 6 lists open research questions and concludes the paper.

2. Related Works

Some good surveys have covered different aspects of ICN research. Ahlgren et al. [2] and Xylomenos et al. [6] outlined the advantages of ICN over traditional networking and also described and compared the features of different proposed ICN architectures. Bari et al. [7] analyzed and compared different ICN naming and routing mechanisms. They also provided detailed descriptions of the rendezvous mechanism, topology, and forwarding strategy of the various ICN frameworks.

Zhang et al. [8] explored ICN caching strategies aimed at minimizing redundancy and maximizing content availability. Abdullahi et al. [9] studied caching for selected ICN architectures and suggested caching methods to minimize total bandwidth consumption, enhance delivery of service, and reduce upward and downward streaming. Zhang et al. [10] discussed recent ICN caching techniques in detail, categorized them, and delineated the possible benefits and drawbacks of each method. Ioannou and Weber [11] surveyed caching challenges in ICN, with a focus on on-path caching. They outlined existing caching and forwarding strategies and discussed current trends and challenges. Zhang et al. [12] and Pruthvi et al. [13] also surveyed caching strategies in ICN; however, they focused on how ICN caching mechanisms impact the Internet of Things (IoT) domain.

Tyson et al. [14] studied key ICN technologies and their individual approaches to producer, consumer, or content mobility. In particular, the authors focused on two types of mobile networks: cellular networks and Mobile Ad Hoc Networks (MANETs). They presented the advantages of ICN in terms of mobility support and illustrated the new ICN-specific challenges, including source mobility, pairwise path routing, and handling outdated requests. Amadeo et al. [15] examined how content-centric networking can be applied to wireless environments in general, as well as in specific networks such as MANETs, Vehicular Ad Hoc Networks (VANETs), and Wireless Mesh Networks (WMNs).

Majeed et al. [16] surveyed the recent advancements in the field of ICN and the existing literature concerning multimedia communication in ICN. They described and categorized the different ICN video-streaming solutions and highlighted future research challenges for video communication in ICN. Aldaoud et al. [17] discussed the existing literature on integrating software-defined networking (SDN) with ICN, associated benefits and challenges, as well as open research questions. Conti et al. [18] and Benseny et al. [19] have examined the feasibility of deploying ICN over existing IP-based networks and listed the potential complications that may arise during implementation and operation.

In this paper, we study research works that focus on enhancing the QoS and QoE in ICN. We categorize the relevant works according to their strategy to improve QoS and QoE, analyze and compare the advantages and limitations of the methods, examine the evaluation metrics and experimental environments, determine which ICN architectures and quality measurements are most frequently studied, and list the open research challenges in the domain.

3. Overview of ICN Architectures

There are four major ICN approaches: Data-Oriented Network Architecture [20], Named Data Networking [21], Publish Subscribe Internet Routing Paradigm [22], and Network of Information [23]. While these projects vary in architectural specifics, they converge on several common assumptions, objectives, and structural properties as they approach ICN.

Among the architectures listed above, only the Named Data Networking (NDN) project [21] (shown in Figure 1) is in active development [24]. NDN expands upon the content-centric networking (CCN) [25] paradigm, offering a topology-independent naming scheme and exploring greedy routing for enhanced router scalability. Please note that although CCN is now known as NDN (both are part of the same research initiative), this manuscript uses the original terminology as presented in each respective article when discussing the existing literature.

The details of all four major ICN architectures, including their working principle and differences, can be found in [2,6]. To provide a general understanding for the subsequent sections, particularly to clarify the specific contributions of ICN-focused studies in Section 5, the following discussion outlines the fundamental principles of the NDN architecture.

Table 1. List of abbreviations used frequently in this article.

Abbreviation	Description
CR	Content router
CS	Content store
FIB	Forwarding Information Base
LAN	Local Area Network
LRU	Least Recently Used
MANET	Mobile Ad Hoc NETwork
NDO	Named Data Object
PIT	Pending Interest Table
VANET	Vehicular Ad Hoc Network
WAN	Wide Area Network
WLAN	Wireless Local Area Network
WSN	Wireless Sensor Network

lists some frequently used abbreviations in the following sections.

Named Data Networking (NDN)

In NDN, Named Data Objects (NDOs) are published at designated nodes, facilitated by routing protocols to disseminate information on NDO locations. Routing within NDN can capitalize on aggregation through a hierarchical naming convention. Requests for NDOs—referred to as “interest” packets—are forwarded hop-by-hop through the network by content routers (CRs). Each CR relies on three core data structures to manage this process: the content store (CS), the Pending Interest Table (PIT), and the Forwarding Information Base (FIB), as illustrated in Figure 1.

The CS serves as a local cache, temporarily holding data objects that have passed through the router. When a CR receives an interest packet (steps 1–3 in Figure 1), it first checks the CS for a matching content object. If found, the data are sent along the incoming interface, and the interest is dropped.

If the requested content is not in the CS, the router consults the FIB, which maps name prefixes to one or more outgoing interfaces. Each FIB entry consists of a name prefix and an ordered list of next-hop interfaces, enabling multi-path forwarding and supporting path-performance evaluation [26]. The router uses the longest prefix match to select the appropriate forwarding path and sends the interest packet toward the next CR.

Meanwhile, the router records the interest in the PIT. Each PIT entry is indexed by the content name and includes a list of interfaces from which the interest has arrived. This ensures that when the corresponding data object is retrieved, it can be forwarded back to all requesting interfaces. Once the requested content is located—either at a publisher or cached in a downstream router—it is returned in a “data” message. The interest is then discarded, and the data travel back to the requester(s) along the reverse path, guided by the PIT entries at each hop (steps 4–6).

CCN natively supports on-path caching; any data message passing through the router is cached by default, allowing future requests for the same content to be served directly from the local CS without forwarding the interest further. From the perspective of a CCN node, a balance exists between requests and responses, with each request generating either a single response or none at all. Depending on local configurations or observed network performance, CCN nodes can adopt various strategies for request transmission pacing and interface selection.

Security for NDOs is ensured through public key cryptography, while trust in cryptographic keys can be achieved through multiple approaches. These include deploying certificate chains modeled after a hierarchical naming system, akin to public key infrastructure, or utilizing authenticated data from a trusted authority.

4. QoS and QoE

4.1. Traditional QoS and QoE Metrics

Quality of service (QoS) measures a computer network’s overall performance, with a particular focus on the experience of network users. In packet-switched communication networks, QoS refers to traffic prioritization and resource reservation rather than the achieved service quality. To quantitatively measure QoS in traditional networks, several metrics, such as latency, throughput, packet loss, bit rate, bandwidth, and jitter, are generally considered [27].

Quality of experience (QoE) measures the satisfaction or annoyance of a customer’s experience with a service, with the evaluation results typically presented as a quality score or rating referenced to a specific scale. The mean opinion score (MOS) is the standard metric used to measure QoE, obtained from subjective human tests. QoE in multimedia applications is a multidimensional construct that extends beyond mere technical metrics to capture the holistic user perception of media delivery. It incorporates various factors, such as initial playback delay [28,29], multimedia synchronization [30], audio/video encoding rate, frame rate [31], stalling or buffering [32,33], video presentation quality, viewing device, and human factors [34,35,36].

4.2. QoS and QoE in ICN

In addition to the traditional metrics, other QoS and QoE factors need to be considered in ICN to account for the new features it introduces. For instance, cache hit ratio, server hit rate, cache replacement rate, and cache utilization are some of the QoS and QoE metrics associated with ICN’s in-network caching. Similarly, the metrics associated with ICN’s name resolution, forwarding, and routing include data delivery rate, path discovery time, and routing success rate. Furthermore, ICN media-streaming performance for distributed content can be assessed by source switching delay. The commonly used QoS and QoE metrics in traditional networks and ICN are shown in Figure 2.

4.3. QoS/QoE Mapping Models

QoS and QoE mapping models bridge the technical parameters of a network (i.e., QoS) and the subjective perception of users (i.e., QoE). These models vary in complexity, inputs, accuracy, working principle, and applications. The mapping models can be divided into the following three classes: black-box media-based models, glass-box parameter-based models, and gray-box parameter-based models [37,38].

Black-box media-based quality models assess media captured at the system’s input and output, relying entirely on empirical data for analysis. They are characterized by their lack of interpretability, meaning that the internal workings of these systems are not easily understandable to users [39]. These models can be further categorized into full-reference and no-reference quality models, depending on the availability of feedback.

Glass-box models, also known as white-box models, consider the underlying mechanisms influencing QoE based on detailed knowledge of the system’s internal workings. These models typically involve looking inside the multimedia transmission pipeline and understanding how individual components contribute to user perception. For example, ITU-T G.1070, a glass-box model, considers factors such as video encoding parameters, packet loss rates, and network latency to predict QoE.

Hybrid or gray-box models [38] combine the benefits of black-box and glass-box mapping models. They examine essential characterization factors at the system’s output, along with some information about the clean stimulus. In other words, they leverage both detailed knowledge of the system internals and empirical observations of user behavior [40]. Because of their higher accuracy compared to glass-box models and easier implementation process, this class of models is gaining popularity.

4.4. QoS and QoE Guarantee in Traditional Networks vs. ICN

The Internet Engineering Task Force (IETF) has defined two models to support quality of service in IP-based networks: Integrated Services (IntServ) and Differentiated Services (DiffServ). IntServ follows the signaled QoS model, where the end hosts signal their QoS needs to the network. On the other hand, DiffServ operates on a provisioned QoS model, configuring network elements to handle multiple traffic classes with different QoS requirements. However, these QoS models face significant limitations. IntServ, for example, reserves the necessary resources for specific traffic flows but is hampered by the high cost of signaling, resource reservation, and the maintenance required for each flow, which greatly reduces its scalability. Conversely, DiffServ is better suited for access networks, requiring a mapping of flows to classes at each edge router. This approach may lead to substantial overhead, even though the burden is confined to edge routers to improve scalability in the core network [41].

MPLS traffic engineering (MPLS-TE) enhances QoS by using DiffServ Traffic Engineering (DS-TE) models to manage aggregated traffic in core networks [42]. MPLS-TE allows for the creation of label-switched paths (LSPs) based on specific performance requirements, such as latency, bandwidth, or reliability. When integrated with DiffServ, this enables differentiated handling of traffic classes, so critical or high-priority data can be routed along paths that meet strict QoS criteria. While DS-TE offers better scalability than IP-based DiffServ and IntServ, it is still limited by overhead [43]. Additionally, hierarchical networks require complex DiffServ mappings, increasing edge node processing demands [41].

The primary challenges associated with IP and MPLS-based QoS models stem from the network’s limited information awareness. However, ICN can largely mitigate these limitations, as the network identifies the requested content using a distinct addressing scheme. This native content identification feature of the ICN architectures simplifies the process of mapping information to QoS requirements, and the need for deep packet inspection is significantly reduced or even eliminated [41]. In addition, the QoS description can be added to content chunks using the content namespace with negligible overhead. However, various ICN models may adopt different QoS approaches, each with distinct features based on their architectural design.

For instance, the PURSUIT architecture provides routing capabilities that facilitate IntServ-like paths without the need for signaling overhead, which is useful for scaling networks [44]. The CCN and NDN architectures support hierarchical and human-readable naming of contents, which makes differentiating between different types of data, i.e., QoS classes, much easier than the current Internet architecture. Moreover, chunk-level load balancing in CCN and PURSUIT multi-path communication can lead to much better multimedia streaming and higher QoE compared to TCP/IP [45].

5. Enhancing QoS and QoE in ICN

In this section, we will summarize and categorize the existing literature on QoS and QoE management in ICN. Depending on the strategy utilized to optimize QoS or QoE metrics, we have divided the research works into the following categories: (i) in-network caching, (ii) name resolution and routing, (iii) software-defined networking (SDN)-based ICN solutions, (iv) transmission and flow control, and (v) media-streaming strategies.

5.1. In-Network Caching

In-network caching is a core feature of ICN architectures that directly enhances both QoS and QoE. By storing data at intermediate network nodes, in-network caching minimizes the need to retrieve content from the original source, which reduces latency, routing delays, server load, and overall network traffic [46,47]. In addition, content-aware caching enables the network to identify cached data independently of the application layer. A trace-driven analysis in [48] demonstrated that in-network caching can reduce three or more hops for 30% of packets, with 11% of requests handled within the requester’s domain—improving access speed and reliability for end-users. Moreover, in-network caching is also attractive to content providers since it can mitigate the capital expense of their content distribution network (CDN) servers while still maintaining satisfactory QoE levels.

ICN architectures use ubiquitous in-network caching, although the design objectives and details may vary depending on the architecture. Based on their operational characteristics, the existing caching mechanisms can be divided into the following categories: (1) Homogeneous and Heterogeneous Caching, (2) Cooperative and Non-Cooperative Caching, and (3) On-Path and Off-Path Caching [10]. In uniform caching systems (homogeneous), all content routers uniformly store passing data, each allocated identical storage capacity. Conversely, selective caching (heterogeneous) restricts data retention to specific routers positioned along the return transmission path. Collaborative caching models enable routers within a network to pool resources, collectively storing a broader array of content segments and sharing cached data to fulfill external requests. In contrast, isolated caching (non-cooperative) operates independently: routers autonomously decide which data to store without disclosing their cached content inventory to peers, eliminating coordination.

Caching in ICN is one of the most popular research areas, and many researchers have focused on utilizing this attribute to improve QoS and QoE. Wu et al. [49] presented a memory update strategy for the content store (CS) of NDN routers (Figure 1) considering data lifetime. Different types of data are assigned different lifetimes, which means instead of using cache update algorithms such as first-in, first-out (FIFO), or Least Recently Used (LRU), an IP packet will instead be discarded from the CS based on how long it has been in the network. Wu et al. [49] and Khelifi et al. [50] utilized traffic classification strategies to improve cache efficiency. In [49], a new column titled “lifetime” was added to the CS, which held the lifetime value for different types of data.

On the other hand, a QoS-aware cache replacement policy for vehicular NDN was presented in [50]. The authors categorized traffic into different classes and divided the CS into sub-cache stores; content was retained or removed based on its popularity density value. In [51], a class-based DiffServ model was proposed for NDN, where traffic classification and conditioning occurred at receiver-side edge networks. The model claimed to offer greater flexibility than traditional IP DiffServ, allowing users to request the same content with different service classes through modified interest aggregation.

Although in-network caching offers many benefits, it also creates some challenges to support Internet-scale ICN cache routers [52]. One major challenge is efficiently managing limited memory resources, as high-speed caching demands fast but costly and low-capacity memory chips. Second, integrating and optimizing multiple memory technologies with different read–write characters is challenging. To tackle these challenges, Li et al. [53] presented HCaching, a hierarchical caching method that uses SRAM (faster than DRAM but costly) as the secondary structure of DRAM to improve the overall throughput of ICN routers. Figure 3 shows a simplified architecture of HCaching, where a large Layer 2 (L2) cache (DRAM) is masked behind a small Layer 1 (L1) cache (SRAM). The SRAM contains three main components: (i) the L2 $A^2$ Index, which uses the $A^2$ buffering algorithm [54] to index L2 cache, (ii) the Reading Cache, managed by the LRU algorithm, and (iii) the Writing Cache, which also utilizes the LRU algorithm.

Simulation results showed that HCaching required much smaller SRAM than LRU and DRAM_SSD [55] for all DRAM sizes and OPC [56] for DRAM sizes more than 1 GB. In addition, HCaching achieved a throughput close to 100 Gb/s, LRU 10 Gb/s, DRAM_SSD 15 Gb/s, and OPC less than 10 Gb/s.

Wang et al. [57] investigated the financial implications of ICN, specifically evaluating its cost-effectiveness prior to widespread adoption within internet service provider (ISP) infrastructures. The authors introduced a caching strategy informed by cost considerations to analyze the trade-off between enhancing QoS and minimizing the economic overheads.

Content popularity, especially for multimedia objects, was studied in [58,59]. Li et al. [59] proposed DASCache, a framework that finds optimal video content placement in the transmission queue of the router. Figure 4 illustrates the queueing model for DASCache’s adaptive video streaming. In each round, the most popular video chunk is placed in the front of the queue, which minimizes the average access time per bit to all users and thereby increases their respective throughput.

Lv et al. [60] and Sellami et al. [61] proposed cooperative and distributed caching strategies for CCN. In [60], a heterogeneous cache allocation method distributed the cache capacity across different CRs. The importance of CRs was evaluated based on network topology and traffic characteristics. The proposed model demonstrated an improved cache hit ratio and utilization and lower routing delay. However, this framework had a high computational complexity, and the study was based on a single topology. On the other hand, a fog computing-based model was proposed in [61]. The authors integrated CCN with fog computing to improve content availability and energy efficiency by optimizing content placement within the network.

Another downside of in-network caching is that content producers do not know where their content is cached, which introduces a problem when the producer wants to update or change the content. Lal et al. [62] proposed a content update scheme for CCN that added extra fields to the interest packets to facilitate version records and backtracking. ndnSIM simulations demonstrated reduced network overhead, lower delay, and improved QoE.

Table 2. In-network caching-based strategies: architectures, experimental platforms, and QoS/QoE metrics.

#	Proposal and Publication Year	Architecture	Experimental Environment	Optimized QoS/QoE Metric
1.	Wu et al. [49], 2014	NDN	Matlab	Service latency, router load/search time
2.	Kim al. [51], 2014	NDN	CCNx [63]	Throughput
3.	Li et al. [53], 2017	ICN	CCNHCaching [64]	SRAM size, throughput, hit ratio
4.	Wang et al. [57], 2017	ICN	Numerical analysis	Cost, service delay
5.	Khelifi et al. [50], 2019	NDN	Python, VANET	Number of cache replacement operations
6.	Lal et al. [62], 2019	CCN	ndnSIM, WAN	Network overhead, average delay, consumer satisfaction score
7.	Yokotani et al. [65], 2019	CCN	CCNx, ESXi virtual environment	Bandwidth, content download time, cache space
8.	Lv et al. [60], 2021	CCN	NS3, WAN	Cache hit ratio, cache utilization, routing delay, load balance degree
9.	Li et al. [66], 2021	CCN	Icarus [67], WAN	Cache hit ratio, latency
10.	Gupta et al. [68], 2022	ICN	Icarus	Cache hit ratio, latency, hop count
11.	Nguyen et al. [58], 2022	NDN	ndnSIM	Cache hit rate, server hit rate
12.	Sellami et al. [61], 2022	CCN	CCNx Contiki framework [69], WSN	Energy consumption, stretch, interest satisfaction rate, cache hit ratio, replacement rate, and delay
13.	Mishra et al. [70], 2023	NDN	ndnSIM, IoT (mesh topology)	Cache hit rate, delay
14.	Hubballi and Chaudhary [71], 2024	NDN	Icarus	Cache hit ratio, content access time
15.	Chaudhary and Hubballi [71], 2025	NDN	Icarus, WAN	Cache hit ratio, content access time, hit distance, cache diversity

lists various ICN frameworks that utilize new caching and allocation strategies to enhance QoS and QoE metrics, the experimental platform they used, the type of network (if details available), and the QoS or QoE metrics they optimized.

Comparative Analysis of Caching-Based Solutions

Based on the references listed in Table 2, we can categorize the strategies into a few general subgroups: (i) popularity-based, (ii) cooperative, (iii) topology-aware, (iv) allocation-based, (v) energy-efficient, and (vi) hybrid.

Popularity-based caching techniques prioritize cache content based on its access frequency. Caching frequently accessed content increases the likelihood of satisfying future requests from the local cache, increasing the cache hit ratio and reducing delay compared to popularity-agnostic methods [72]. However, accurately estimating content popularity (local or global) is a challenge. Local estimation might miss broader trends, while global estimation can introduce communication overhead. In addition, pure popularity-based approaches might overlook factors like content freshness, content size, network topology, or the location of the consumer.

Cooperation can lead to a wider variety of content being cached within a neighborhood, increasing the chance of a cache hit and decreasing redundancy [60,61]. However, cooperation requires exchanging additional messages, which can increase network overhead, especially for larger networks. Also, implementing and managing cooperative caching systems can be more complex than non-cooperative ones.

Topology-aware techniques consider the network topology (e.g., node centrality, distance to consumers/sources, router connectivity) when making caching decisions. Caching at central or well-connected nodes can serve a larger number of potential requests, improving the cache hit ratio and reducing network load. The downside is that determining the “central” or important nodes can be computationally intensive [60], and caching heavily on a few “important” nodes might lead to these nodes becoming congested.

Allocation-based strategies focus on how the total available cache capacity is distributed among the network nodes (homogeneous vs. heterogeneous). Heterogeneous allocation can help distribute the caching load more evenly across the network and can optimize the deployment of larger ICN caches. However, deciding how much cache to allocate to each node based on various factors (topology, traffic, cost) is a complex optimization problem [60].

Primarily relevant in resource-constrained environments like vehicular networks and IoT, energy-efficient caching strategies aim to minimize energy consumption related to caching operations. A challenge faced by these techniques is increasing metrics such as cache hit ratio and retrieval latency [68].

Hybrid strategies combine elements from different caching approaches to leverage their individual strengths [66]. On the other hand, combining different strategies can lead to more complex designs and implementations.

5.2. Name Resolution and Routing

Name resolution in ICN involves matching an object name to a server or source that can supply that object to a requester. This can be achieved through hierarchical namespaces, resembling domain names in DNS, or through flat namespaces, where each content name is globally unique. Hierarchical namespaces enable scalable and efficient name resolution by utilizing hierarchical routing tables, while flat namespaces often rely on distributed lookup mechanisms like distributed hash tables (DHTs). Efficient name resolution ensures that content requests are quickly directed to the correct network locations, contributing to the scalability and decentralization of ICN architectures.

Routing means constructing a path to transfer the object from the server to the client. In ICN, routers route packets based on content names instead of destination IP addresses. Content-centric routing and interest-based forwarding prioritize delivering content efficiently by caching and forwarding content toward the nearest or most suitable locations. Plus, content-based routing decisions optimize content delivery by considering factors such as content popularity, proximity, and network conditions. One aspect that sets apart various ICN approaches is their treatment of name resolution and routing. These functions may either be interlinked or operate separately from each other.

Kerrouche et al. [73] and McCarthy et al. [74] presented QoS-aware forwarding strategies for NDN environments. In [73], an ant colony optimization-based forwarding strategy was proposed, which utilized both forward and backward ants (interest and data packets) to rank interfaces and then selected the best interface to forward incoming Interests. To improve data transmission efficiency, this framework incorporated computed probability metrics from all available interfaces into a roulette wheel selection mechanism. This methodology enabled the inclusion of interfaces exhibiting reduced probability metrics (pheromone levels) in the decision process, ensuring a more equitable dispersion of network traffic. On the other hand, McCarthy et al. [74] proposed a QoS-aware algorithm that extended the NDN model to embed QoS information into request and corresponding data packets and utilized multi-hop and multi-path forwarding techniques to meet the QoS constraints.

Thomas et al. [75] proposed a multi-path selection strategy for the PSIRP architecture, considering bandwidth and error rate constraints. This solution facilitates QoS-aware routing decisions within individual network domains. The strategy specifically targets the requirements of on-demand video-streaming applications by prioritizing increased bandwidth allocation and minimizing packet loss while also including configurable limits on permissible path counts. The authors also studied the interaction between their model and a video-on-demand streaming service to deliver high-definition content. Scalable video technology (H.264/SVC) [76] was used to provide a layered representation of the content (three SNR-scalable layers, 1080p). To evaluate QoE, the Pseudo-Subjective Quality Assessment (PSQA) [77] approach was used, and the proposed method was shown to improve QoE and path computation.

Hou et al. [78] presented an optimization technique leveraging particle swarm intelligence, which analyzes historical data from content object transmissions to dynamically adjust routing likelihoods in forwarding tables (FIB). This method enables intelligent path selection by directing user requests toward the most efficient content source (publisher or server) among redundant providers of identical data, balancing QoS parameters during decision-making. Another swarm optimization technique was employed by Cheng et al. [79]. By mapping user QoE to specific QoS metrics, they calculated satisfaction levels and determined forwarding probabilities for router interfaces. This method allowed for dual-path content discovery, significantly improving routing success.

Rani et al. [80] introduced a fuzzy-based congestion-aware routing method to address congestion from flooding attacks in ICN. Their approach builds on the OSPF algorithm, evaluating paths using satisfaction ratio and packet loss metrics. These metrics, along with trust values indicating link reliability during high traffic, are shared across the network via link state packet vectors, enabling the selection of secure, congestion-free routes that improve QoS.

Kuang and Yu [81] proposed a QoS-aware architecture for multimedia applications in wireless ICN. Their architecture unifies QoS management across path selection, transmission policies, resource reservation, source selection, and bandwidth allocation to optimize performance in decentralized settings. Another QoS-aware forwarding strategy was proposed by Abdelaal et al. [82]. This framework utilizes a cooperative forwarding approach where routers share data names and interfaces to estimate optimal paths toward cached versions of the requested data in NDN.

Mizunaga and Kobayashi [83] proposed a flat naming scheme for CCN instead of the standard hierarchical naming scheme. By utilizing multiple keywords, the flat naming scheme simplifies content retrieval for users, making it more intuitive and user-friendly. Unlike traditional hierarchical naming schemes that require precise hierarchical addresses, the flat naming scheme reduces the burden on content requesters by offering a more flexible and accessible way to identify and retrieve content, thus enhancing QoE in CCN.

Table 3. Name resolution and routing-based strategies: architectures, experimental platforms, and QoS/QoE metrics.

#	Proposal and Publication Year	Architecture	Experimental Environment	Optimized QoS/QoE Metric
1.	Thomas et al. [75], 2015	PSIRP	Simulation, PSQA [77]	QoE (mean opinion score)
2.	Hou et al. [78], 2017	NDN	NS-2 and Matlab	Delivery rate, average cost
3.	Kerrouche et al. [73], 2017	NDN	ndnSIM 2.0 [84], WAN	Data delivery time, cost, hop count, dropped packets, and mean hit ratio
4.	Rani et al. [80], 2018	NDN	NS-3 based NDN simulator	Packet loss, throughput
5.	Kuang and Yu [81], 2018	ICN	OMNeT++, wireless network	Throughput, delay
6.	Cheng et al. [79], 2018	ICN	CERNET, CERNET2, WAN	Hop count, routing success rate, cache hit rate
7.	McCarthy et al. [74], 2019	NDN	ndnSIM and SUMO, VANET	Packet loss, retransmission rate
8.	Mizunaga and Kobayashi [83], 2020	CCN	Cefore (Python)	Processing time
9.	Abdelaal et al. [82], 2020	NDN	Mini-NDN [85]	Round-trip time, bandwidth
10.	Tsai et al. [86], 2022	CCN	Python programming, WAN	Cache hit ratio, delay, forwarded interest count, timeout interest count, throughput, jitter, packet loss
11.	Li and Li [87], 2023	ICN	Python programming	Delay
12.	Ma et al. [88], 2024	ICN	Icarus, WAN	Delay, link load

lists various ICN frameworks that utilize new name resolution and routing strategies to enhance QoS and QoE metrics.

Comparative Analysis of Name Resolution and Routing-Based Solutions

To compare the strategies listed in Table 3, we can divide them into the following subgroups: (i) name-based forwarding, (ii) name-resolution architecture, (iii) FIB management, and (iv) cooperative forwarding schemes.

Name-based forwarding techniques select interfaces or paths that promise improved QoS or QoE metrics, e.g., better bandwidth, lower delay, or reduced loss. Their advantage lies in their direct integration with ICN’s fundamental name-based forwarding, allowing for fine-grained control at each hop. However, they often rely on accurate and timely QoS parameter estimation, which can introduce overhead through probing or information exchange. Also, per-hop optimizations might not always guarantee end-to-end QoS, and these strategies need to balance exploration of better paths with exploitation of known good ones.

The second category focuses on name resolution architectures, which are crucial for scalability and efficient content discovery, especially in ICN with flat naming. These solutions improve network performance by providing a structured way to map content names to network locations, reducing the need for extensive flooding and potentially offering deterministic resolution latency. Note that the scalability of the name resolution system depends heavily on its design (e.g., DHT-based vs. hierarchical), and issues such as single points of failure or challenges in maintaining mapping consistency need to be addressed.

Forwarding Information Base (FIB) management solutions proactively discover content availability and maintain accurate FIBs. By making routers more aware of content locations, these strategies can reduce network congestion and improve response times. However, maintaining an accurate and up-to-date FIB can be resource-intensive, especially in highly dynamic networks.

Cooperative forwarding strategies utilize collaboration between ICN routers to find cached content closer to the consumer, and the content may be delivered from either single or multiple sources. By distributing the load across multiple providers, cooperative forwarding can lead to reduced congestion. However, the effectiveness of these strategies relies on the network’s up-to-date knowledge of available providers and accurate bandwidth estimation, and stale or incorrect information can lead to suboptimal decisions.

5.3. SDN-Based ICN Solutions

Software-defined networking (SDN) [89] is essential for enhancing and managing QoS and QoE in modern networks due to its centralized control and traffic engineering capabilities. SDN separates the control plane from the data plane [90], allowing efficient resource allocation, load-balancing congestion control, and traffic prioritization, all of which are essential to maintaining high QoS and QoE.

SDN plays a pivotal role in advancing ICN by enabling dynamic control and programmatically efficient management of network resources. SDN enhances ICN’s content distribution, caching, and retrieval capabilities by providing a centralized platform for managing content routing, caching policies, and QoS requirements. SDN controllers leverage a variety of protocols, such as OpenFlow, to communicate with forwarding devices, enabling real-time adjustments to network policies and traffic flows. In addition, SDN allows for the seamless integration of ICN with existing network infrastructures, facilitating incremental deployment and migration strategies.

One strategy used by some research works in this domain is utilizing the traffic engineering ability and independent control layer of SDN. Zhang et al. [91] proposed an information-centric wireless network virtualization technique that transmits time-sensitive multimedia data with the delay-bounded QoS guaranteed in SDN environments. The authors combined the inherent advantages of SDN with the in-network caching feature of ICN to provide maximum effective capacity and minimum transmission delay. Adnan et al. [92] utilized the SDN controller to keep track of the routing information for mobile users in an NDN environment and showed improvement in several QoS metrics.

Another strategy adopted by some SDN-based approaches is implementing an intermediate layer to translate names into IP addresses. This layer, often functioning as a proxy and overlay, is prevalent in modern networks where proxy servers are common. Several works [93,94,95] have leveraged proxy servers to manage ICN networks efficiently. Nguyen et al. [94] utilized a proxy to envelop and hash messages between the CCNx daemon and the OpenFlow switch. This proxy’s wrapper module hashed the content name prefix in the interest packet and embedded it within the existing IP packet. The authors considered off-path caching and addressed cache replication issues to enhance hit ratios. However, the wrapper added a slight delay and marginally reduced forwarding efficiency by 5%. Trajano et al. [95] employed multiple proxies, each including a shared and distributed hash table that served as the content index. Upon receiving interests, the controller directed them to the nearest proxy for data retrieval. If the proxy located the requested data in its hash table, it sent the interest to the cache node holding the data; otherwise, it forwarded the interest to the closest proxy server. However, this method risks overloading proxies as they handle all interests by resolving and forwarding them to the controller.

Liu et al. [96] proposed an SDN-based NDN architecture to support IP-compatible ICN packets. The Multi-Protocol Label Switching (MPLS) technique was used to encapsulate and label the NDN payload within IP packets. Flores et al. [97] presented an OpenFlow-compatible key-based routing solution to enable content-centric networks to be deployed on the current TCP/IP-based architecture. End hosts were distinguished by virtual identifiers utilized for network communication. This SDN-over-ICN model demonstrated improved transmission efficiency and high QoE.

Table 4. SDN-based strategies: architectures, experimental platforms, and QoS/QoE metrics.

#	Proposal and Publication Year	Architecture	Experimental Environment	Optimized QoS/QoE Metric
1.	Nguyen et al. [94], 2013	CCN	CCNx, Pronto 3290 OpenFlow switch	Throughput
2.	Zhang and Zhu [91], 2016	ICN	Simulation (platform details not included), wireless network	Statistical QoS, power consumption
3.	Trajano and Fernandez [95], 2016	ICN	Mininet [98]	QoE, delay, throughput
4.	Flores et al. [97], 2019	ICN	Mininet, wired network	Throughput, QoE
5.	Liu et al. [96], 2021	NDN	Mininet	Download time (delay)
6.	Adnan et al. [92], 2023,	NDN	Mininet, mobile network	Delay, packet loss, jitter, throughput
7.	Tavasoli et al. [99], 2024	ICN	Python programming	Cache hit ratio, cost, power consumption

lists various ICN frameworks that utilize SDN to enhance QoS and QoE metrics.

Comparative Analysis of SDN-Based Strategies

To compare the strategies listed in Table 4, we can categorize them into the following subgroups: (i) optimized centralized management and (ii) NFV-based solutions.

The first category leverages the global view and programmability of SDN to enhance core ICN functions. An advantage of this strategy is the ability to make informed decisions regarding caching, routing, and load balancing across the entire network. For instance, Nguyen et al. [94] uses a central controller to determine popular content and optimal caching locations to reduce retrieval time and bandwidth consumption. This centralized control directly improves QoS and QoE by reducing delays and increasing network efficiency. However, a potential limitation is the scalability of the central controller in very large and dynamic networks, as well as the risk of a single point of failure.

The second category focuses on SDN and NFV for flexible and coexistent ICN deployments, which emphasizes the practical aspects of deploying ICN incrementally. Architectures like ContentSDN [95] utilize NFV to deploy caching functionalities dynamically, allowing for flexible adjustments based on demand and policies. SDNDN, proposed in [96], directly addresses the coexistence of ICN (specifically NDN) with existing IP networks by extending OpenFlow switches to handle both types of packets. The primary advantage here is the flexibility and ease of incremental deployment without requiring a complete overhaul of the network infrastructure. This approach improves QoE by bringing content closer to users in a manageable and scalable way and by enabling the use of ICN benefits for specific traffic types.

5.4. Transmission and Flow Control

Transmission and flow control mechanisms play fundamental roles in enabling efficient and reliable content delivery. Transmission in ICNs involves the dissemination of content packets from producers to consumers across the network. To ensure efficient transmission, ICNs often employ caching strategies at intermediary nodes to store frequently accessed content, reducing the need for long-distance transmissions and improving content delivery latency. Additionally, transmission in ICNs may involve multicast or anycast mechanisms to efficiently transmit content to multiple consumers or replicate content across distributed nodes, enhancing scalability and resilience.

Flow control mechanisms in ICNs regulate the rate of content transmission to prevent congestion and ensure fair resource utilization. This is particularly important in scenarios where multiple consumers are competing for the same content or where network resources are limited. Flow control strategies typically employ congestion detection and avoidance algorithms, such as Explicit Congestion Notification (ECN) and window-based flow control. ECN allows routers to notify senders of congestion by setting a specific bit in packet headers, enabling senders to adapt their transmission rates accordingly. Window-based flow control, on the other hand, adjusts the transmission window size based on network conditions to avoid overwhelming the network with excessive traffic. However, the traditional congestion flow control mechanisms designed for the Transmission Control Protocol (TCP) may not be directly employed in some ICN architectures (e.g., CCN or NDN) because the RTT estimation could be inaccurate due to how NDN interest (request packet) and content objects (response packet) work [100,101]. Furthermore, managing data flow in ICN systems is challenging due to unpredictable user requests, dynamic network conditions, and in-network caching [5,102].

Dynamic Adaptive Streaming over HTTP (DASH) has become a standard for many streaming services and platforms due to its ability to deliver high-quality media streaming over the Internet despite network congestion or bandwidth fluctuations. DASH dynamically adjusts the quality of the video stream based on the available network bandwidth and the capabilities of the receiving device. However, in ICN, this client-driven bit rate adaptation mechanism may inaccurately gauge available bandwidth, as cached content fragments along transmission paths reduce data retrieval latency. This mismatch between perceived and actual end-to-end bandwidth triggers unstable oscillations in streaming quality, where the system alternates between high and low bit rates [103].

Figure 5 illustrates the bandwidth overestimation problem in DASH over CCN implementation. Suppose the video is available in several different qualities, and the entire file is divided into four segments. Router $R_1$ has the first segment cached in its local content store, while the entire video is available at the server (producer). When the consumer requests segment #1, router $R_1$ quickly sends the content using its stored copy. Since the request is filled fast from a nearby cache, the user might assume the communication channel has a high bandwidth and then request a higher-quality version of segment #2. But the links between routers $R_1$ – $R_2$ and $R_2$ – $R_3$ are incapable of supporting the higher bandwidth needed for that better-quality video. This will result in client-side buffer depletion and stalling in video playback and, consequently, a poor QoE.

Liu and Wei [104] addressed this throughput estimation problem by proposing a hop-by-hop quality adjustment strategy named HAVS-CCN. The authors introduced a new scheduling window (SDQ) for CCN routers, which helps manage data flow by controlling how fast packets are sent at the router level. To adapt video quality, routers organize content in queues based on how they affect video playback—prioritizing critical data to avoid buffering. When cached content or new data need to be sent out, routers use SDQ rules to decide what goes first. Also, instead of relying on standard DASH with AVC encoding, their system uses scalable video coding (SVC) [76], which allows smoother quality adjustments.

Zhu and Zhang [105] introduced a negotiation-based gaming framework to manage the complexities of multiple users simultaneously accessing popular content from distributed network caches while meeting statistically defined QoS criteria. Their multilateral negotiation framework involves bidirectional interactions between players—content repositories and users—who engage in price-based bargaining with all relevant counterparts to maximize their respective utilities. For users, utility is defined as the net benefit of achieved QoS improvements minus incurred expenses, while content stores evaluate utility as revenue from service agreements offset by operational costs. The strategy was evaluated based on the utility outcomes for all participants and the mathematical modeling of consensus likelihood across bargaining scenarios.

Van and Mau [106] studied a multi-source CCN architecture in which data are segmented and dispersed across multiple servers, deviating from traditional single-server storage models. The authors claimed that this distributed caching could help tackle the limited cache size constraint of CCN routers, especially in the case of multimedia streaming.

Han et al. [107] and Kuang et al. [108] designed solutions for wireless ICN architectures. In [107], an adaptive retransmission algorithm was proposed for wireless CCN that addresses packet loss and round-trip time (RTT) fluctuations caused by in-network caching. This approach dynamically adjusts retransmission timeouts via sequential hypothesis testing while accounting for packet loss causes (e.g., congestion or signal issues). NS-2 simulations showed the scheme outperformed default CCN and RTO methods [109], achieving better loss recovery and video quality. On the other hand, a real-time content streaming framework for wireless ICN was presented in [108]. This strategy considers the different QoS requirements of media consumers and selects the optimal content provider. Experimental results from the OMNeT++ simulator for an IEEE 802.11g network demonstrated lower delay and jitter and increased throughput.

In [110], a DiffServ-based congestion control method was proposed for ICN that adjusts the traffic flow at the router’s forwarding interface to improve QoS. The framework classifies ICN traffic into distinct service categories and applies service-category-specific traffic regulation at each network node along the transmission path. Additionally, it uses a resource-availability signaling system to communicate current bandwidth conditions to content requesters. Simulation results from ndnSIM demonstrated higher throughput, lower latency, and minimal packet loss.

Table 5. Transmission and flow control-based strategies: architectures, experimental platforms, and QoS/QoE metrics.

#	Proposal and Publication Year	Architecture	Experimental Environment	Optimized QoS/QoE Metric
1.	Han et al. [107], 2013	CCN	NS-2, WLAN	Loss recovery rate, video quality
2.	Liu and Wei [104], 2016	CCN	MiniCCNx [111] (based on CCNx 0.8.2)	Playback quality, average delay
3.	Zhu and Zhang [105], 2016	ICN	Numerical analysis, wireless network	Statistical QoS
4.	Van and Mau [106], 2016	CCN	OPNET, wireless network	Cache hit ratio, bandwidth utilization, server load, content delivery delay, QoE
5.	Kuang et al. [108], 2019	ICN	OMNeT++, MANET	Delay, throughput, jitter
6.	Bai et al. [110], 2022	ICN	ndnSIM	Throughput, latency, packet loss rate
7.	Gao et al. [112], 2024	CCN	OMNet++ and ccnSim	Throughput, latency
8.	Nakagawa et al. [113], 2024	ICN	amus-ndnSIM [114]	QoE, delay, bit rate, throughput

lists various ICN frameworks that enhance QoS and QoE metrics by optimizing transmission, flow, and congestion control.

Comparative Analysis of Transmission and Flow Control-Based Solutions

To compare the strategies listed in Table 5, we can divide them into the following subgroups: (i) transmission fairness, (ii) hop-by-hop flow control, and (iii) client-driven adaptations.

The fairness-focused flow control strategies primarily aim to improve the end-user experience by ensuring equitable resource sharing and preventing highly popular content from monopolizing bandwidth. A potential drawback is that strict fairness might sometimes lead to suboptimal overall network utilization if resources are not allocated based purely on demand.

Hop-by-hop flow control techniques involve intermediate routers along the forwarding path making local decisions to manage network traffic and congestion. The localized control and class-based interest shaping allow better management of Differentiated Services and improved QoS and QoE support. The hop-by-hop control approaches add state management overhead on routers; plus, they require implementing new scheduler components.

In client-driven adaptation methods, consumers actively participate in optimizing their content retrieval. Clients fetch content fragments from multiple sources over different paths and dynamically adjust retransmission timeouts and window size. The negotiation-based gaming approach enables clients to negotiate with content caches based on price and desired QoS levels. Note that these methods rely on accurate client-side measurements (as shown in Figure 5).

5.5. Media-Streaming Strategies

Traditionally, adaptive streaming has been widely used to improve the consumer QoE for multimedia streaming. Adaptive streaming schemes such as DASH enable video quality adjustment during playback to accommodate the current network conditions, which allows for avoiding bigger issues like stalling.

While DASH has been very effective in IP-based networks, new challenges associated with media streaming in ICN make it less effective to implement DASH without modification. For instance, DASH utilizes end-to-end bandwidth estimation, which is difficult in ICN due to content retrieval from multiple sources [115,116]. In addition, DASH could also suffer from bandwidth overestimation issues illustrated in Figure 5. Furthermore, DASH cannot utilize the content cached at nearby ICN routers’ content stores (CSs) as it is designed for a client–server architecture, while ICN focuses on one-to-many content dissemination. Some studies investigating QoS and QoE in ICN from the multimedia-specific perspective are described in this section and listed in Table 6. These media-focused studies pay particular attention to QoE due to its importance in multimedia consumption.

Ramakrishnan et al. [115] proposed a CCN adaptive video-streaming framework based on network coding to enable a CCN client to fetch a video from multiple sources using multiple interfaces of a CCN router. The model was tested on the CCNx 0.8.2 platform in conjunction with the CCN-VLC plugin made available by the ITEC-DASH project [117]. Concurrent streaming from multiple sources (two servers with network coding enabled) produced better QoE and higher throughput than streaming from a single server.

Nunome [118] studied the access time of contents within the content store of a CCN router and how users could exploit it. The author considered a scenario where multiple consumers access the same video or audio content simultaneously. If a terminal deliberately delays access time, it opens up the possibility of exploiting cached data in the CS.

Sadat et al. [116] investigated the impact of distributed content caching on video streaming in CCN. The authors quantified the number of source switchings during video streaming and the impact on users’ QoE. Similar to [118], simultaneous retrieval of the same multimedia content was considered in [116]. A new adaptive video-streaming framework was developed utilizing MOS from human subject tests to guarantee satisfactory QoE. Experimental results derived from the ccnSim simulator and CCNx emulator demonstrated the efficiency of the proposed streaming framework compared to standard DASH and SVC-based streaming in terms of QoE, delay, and bit rate/request.

Takada et al. [119] and Kobayashi et al. [120] proposed solutions for CCN audio and video transmissions and investigated the impacts of different caching strategies on QoE and application-layer QoS. In [121], authors proposed a strategy to improve QoS and QoE for repeatedly accessed multimedia content (“multi-view”) through coordinated scheduling of content retrieval initiation times across users.

Table 6. Media-streaming strategies: architectures, experimental platforms, and QoS/QoE metrics.

#	Proposal andPublication Year	Architecture	Experimental Environment	Optimized QoS/QoE Metric
1.	Ramakrishnan et al. [115], 2016	CCN	CCNx 0.8.2, LAN	QoE, throughput
2.	Sadat et al. [116], 2019	CCN	ccnSim, CCNx [63], LAN and 802.11n WLAN	QoE (MOS from human subject test), delay, bit rate/request
3.	Toshiro Nunome [118], 2022	CCN	CeforeNet, LAN	MOS, video media unit (MU) loss ratio, video error concealment ratio, audio media unit loss ratio
4.	Nunome and Takada [121], 2023	CCN	Cefore [122], LAN	Audio media unit loss ratio, video media unit loss ratio, viewpoint change delay, MOS
5.	Takada and Nunome [119], 2024	CCN	Cefore, LAN	MOS, cache utilization ratio, video media unit loss ratio
6.	Kobayashi and Nunome [120], 2024	CCN	NS-3 and ccns3sim [123]	Cache hit ratio, audio media unit loss ratio, video error concealment ratio

Table 6. Media-streaming strategies: architectures, experimental platforms, and QoS/QoE metrics.

#	Proposal andPublication Year	Architecture	Experimental Environment	Optimized QoS/QoE Metric
1.	Ramakrishnan et al. [115], 2016	CCN	CCNx 0.8.2, LAN	QoE, throughput
2.	Sadat et al. [116], 2019	CCN	ccnSim, CCNx [63], LAN and 802.11n WLAN	QoE (MOS from human subject test), delay, bit rate/request
3.	Toshiro Nunome [118], 2022	CCN	CeforeNet, LAN	MOS, video media unit (MU) loss ratio, video error concealment ratio, audio media unit loss ratio
4.	Nunome and Takada [121], 2023	CCN	Cefore [122], LAN	Audio media unit loss ratio, video media unit loss ratio, viewpoint change delay, MOS
5.	Takada and Nunome [119], 2024	CCN	Cefore, LAN	MOS, cache utilization ratio, video media unit loss ratio
6.	Kobayashi and Nunome [120], 2024	CCN	NS-3 and ccns3sim [123]	Cache hit ratio, audio media unit loss ratio, video error concealment ratio

Comparative Analysis of Media-Streaming Strategies

To compare the strategies listed in Table 6, we can divide them into the following subgroups: (i) adaptive and multi-source streaming and (ii) media-focused cache optimization.

The first category of strategies improves QoE and QoS through adaptive mechanisms and by leveraging multi-source streaming or temporal availability of content. These approaches employ adaptive streaming algorithms that switch quality based on network conditions and potential source switches, aiming to provide a higher-quality viewing experience despite network variations or the distributed nature of content in ICN/CCN. Some techniques also delay access to already-cached content to improve the chances of retrieving content from closer, faster sources [118]. A potential drawback is the added complexity in implementing these dynamic adaptations and the overhead of continuous monitoring of cache availability.

The second category focuses on optimizing cache management and policies to improve efficient media delivery. It includes approaches such as linking different viewpoints in multi-view video to improve cache utilization [121] and deleting duplicate cached content to obtain better hit rates when handling multiple streams at once [120]. This inherently improves network performance by reducing traffic on core network links and lowering content retrieval latency. The improved cache hit ratios and reduced loss ratios contribute to better QoS, which in turn positively impacts QoE.

6. Discussion and Conclusions

6.1. Observations

• QoS vs. QoE: The majority of the existing ICN literature has studied QoS metrics. Some of the most commonly optimized QoS metrics include delay, throughput, cache hit ratio, and packet loss. Compared to QoS, fewer studies have measured QoE metrics for performance comparison. Among other factors, this could be due to the difficulty in measuring QoE accurately without human subjective tests. Some studies [75] have used QoE prediction models to approximate human perceptual quality, while others have defined QoE as a function of satisfied interests [115]. Sadat et al. [116] have utilized both the QoE prediction model as well as mean opinion scores (MOSs) from human subject tests.
• Popular Strategies: In ICN, the focus on optimizing QoS or enhancing QoE seems to predominantly center around in-network caching. This is because ubiquitous in-network caching is one of the key differences between the current Internet structure and ICN, and this caching impacts many aspects of a network, including throughput, congestion, delay, load balancing, routing, security, and consumer QoE. Another frequently studied area in the existing literature is name resolution and routing. This focus is justified by ICN’s unique content naming and multi-path forwarding capabilities, which facilitate the development of novel QoS and QoE optimization techniques.
• Commonly Studied ICN Architecture: Among the literature we reviewed, CCN or NDN is the most commonly studied ICN architecture. In total, 32 out of the 48 studies listed in Table 2, Table 3, Table 4, Table 5 and Table 6 introduced solutions specifically for either CCN or NDN. This is consistent with the fact that NDN is considered the most prevalent among the ICN architectures [13].
• Commonly Optimized QoS/QoE Metrics: Among the most frequently optimized QoS and QoE metrics across the various strategies we examined are latency, cache hit rate, throughput, and mean opinion score (MOS). Figure 6 shows the number of research studies where each of these metrics was assessed. Aside from cache hit rate, the low counts of other ICN-specific QoS and QoE metrics in Figure 6 are due to the use of different evaluation metrics across studies, depending on the specific strategy employed.

6.2. Future Research Directions

6.2.1. Scalability

One of the major challenges in ICN is ensuring scalability and performance. Maintaining efficient and effective content distribution becomes increasingly complex as the content volume and the number of users grow. This includes addressing issues related to large-scale content naming, routing, and caching mechanisms [124,125]. The dynamic nature of content creation and consumption patterns further complicates the development of scalable solutions that can adapt in real time while maintaining high throughput and minimal packet loss, which are essential for a good QoE.

6.2.2. Security

In contrast to the current Internet, which secures the “path” through which traffic flows, ICN secures each “content chunk”. This is equivalent to securing a road vs. the vehicles. Common attacks in ICN include content poisoning, interest flooding, forwarder watchlist, timing analysis, cache flooding, cache replay, and cache pollution [126,127]. To secure data, ICN uses per-packet encryption, and this additional feature may potentially result in larger packets/more overload. For example, encryption can introduce latency and affect overall user satisfaction [128]. Therefore, security mechanisms designed for ICN must be lightweight and efficient to avoid degrading QoS and, consequently, the QoE.

A critical issue in existing literature is the predominant focus on evaluating network performance (QoS or QoE) based solely on the intrinsic security features of ICN, which is insufficient for real-world network environments and the growing landscape of cyber threats [127]. For instance, implementing encryption—including at the ICN packet header level—makes caching non-trivial [129,130]. Therefore, further research is essential to examine how additional security measures may influence QoS and QoE within ICN frameworks.

6.2.3. Naming and Namespace Management

Hierarchical naming schemes in ICN, as adopted by NDN, offer the advantage of being more intuitively perceived by humans due to their similarity to DNS names and file paths. They also allow for aggregation in the routing table, leading to a smaller FIB size [83]. However, ensuring global uniqueness requires strict constraints during name creation, increasing the load on content creators, and longer or non-aggregable names can increase the FIB size and cause routing overhead. Flat naming schemes have been proposed as an alternative; however, they require a name resolution system to map these names or identifiers to network addresses for routing. Plus, as content names become more complex, the overhead associated with name resolution and routing increases, negatively affecting content retrieval times and user experience [124]. Balancing between detailed, expressive names and accurate and fast content retrieval is crucial.

6.2.4. Caching

Effective caching strategies are integral to achieving high QoS and QoE in ICN. Optimal cache placement, size, and replacement policies can significantly reduce content retrieval times and network congestion, improving overall network performance and user satisfaction. Plus, understanding content popularity and access patterns is crucial for developing adaptive caching mechanisms that enhance QoE. In addition, preventing cache pollution and ensuring fair content distribution are also essential for maximizing the benefits of in-network caching. Furthermore, ICN-specific QoS/QoE metrics, such as SRAM size, interest satisfaction rate, content store search time, etc., should be more thoroughly investigated.

6.2.5. Interoperability and Integration with Existing and Emerging Networks

The integration of ICN’s new components with existing TCP/IP-based network architectures and protocols while maintaining backward compatibility is crucial for a smooth transition. Issues related to protocol translation, inter-domain routing, and unified management frameworks must be addressed to ensure consistent QoS and the overall user experience during the integration process. ICN implementation solutions often rely on clean slate deployment [131,132] due to its straightforward design, but practical implementation is challenging due to ICN being newer than TCP/IP. Therefore, integrating ICN with existing IP infrastructure through overlay/IP [133], underlay/IP [134], and coexistence/IP [135] modes should be explored despite potential drawbacks, such as IP-related issues and the additional overhead from ICN/IP duality [136]. Additionally, while the pull communication model is commonly used, ref. [136] suggests further investigation into push and publish–subscribe models [22,137] for their valuable applications in IoT.

Several studies have investigated the interaction between ICN and emerging networking paradigms, highlighting its natural alignment with the shift toward non-IP-based networking, super convergence, and data-centric communication in 6G. ICN’s fundamental focus on named data, along with its in-network caching, improves 6G performance by accelerating content delivery, particularly when combined with Multi-access Edge Computing (MEC), which brings computation and storage closer to users [138]. Additionally, ICN’s inherent support for mobility complements the dynamic nature of future networks, while semantic routing can enable more intelligent and efficient content delivery. The convergence of ICN with in-network computing further optimizes resource utilization and reduces latency for demanding applications like extended reality (XR) and holographic communication [139]. Furthermore, dynamically created cloud instances and vehicular cloud architectures can leverage ICN to enable efficient multimedia streaming and data retrieval [140]. However, several research challenges remain, including enabling ICN to provide contextual information about users, applications, and services to MEC for improved QoS in 6G use cases [138]. Additionally, understanding how ICN-MEC integration can effectively interact with key 6G technologies, such as terahertz (THz) waves, optical wireless technology, dynamic spectrum management, and intelligent reflecting surfaces, presents open challenges for future research.

6.3. Conclusions

In conclusion, the introduction of ICN presents a compelling solution to the limitations of the current internet architecture. ICN’s focus on delivering content based on user interests rather than specific node addresses aligns well with the evolving demands of modern communication and content sharing, particularly in a landscape dominated by data-intensive services and IoT. This research article analyzes, describes, and categorizes the current ICN literature, focusing on QoS and QoE metrics. Furthermore, the current research trends and existing challenges have been outlined, offering the research community insights to focus on under-explored areas.