In-network caching in information-centric networks for different applications: A survey

Abstract

Information-Centric Networks (ICNs) have become a promising paradigm for the future Internet instead of host-based communication. In this network, content-oriented data and in-network caching are two key properties, which have brought tremendous benefits to many challenging networks by improving the distribution and retrieval of content. The content is cached along the reversed delivery path, on-path caching and off-path caching. Generally, the existing caching approaches in Information-Centric Networks mainly consider the key characteristics of the content (to decide which content is to be cached), the network (to determine the appropriate time for caching), the nodes (to determine where content should reside), etc. This paper presents a comprehensive survey of in-network caching of ICN schemes (also known as ICN-based caching) in many network applications. Thus, a detailed analysis of the existing caching approaches in several network applications, such as MANET, VANET, IoT, and WSN, is given. This survey also provides a taxonomy for caching policies according to some feature-based approaches and highlights the trends and evaluation problems in these areas. Finally, the challenging research directions in different applications are also pointed out in this subject.

Keywords:

• ICN
• CCN
• NDN
• SCN
• MANET
• VANET
• WSN
• IoT

1. Introduction

Since the Internet was formed and designed in the 1960s, it has played a more prominent role in people’s life. So far, the vast majority of content usages in the current Internet is distributed from a source to several/many users (like multicast) both disseminating multi-media file from producers (e.g., Netflix, IPTV, Web, etc.) and sharing generated data from users (e.g., social networking Facebook, Weibo, YouTube, Youku, etc.). Along with the rapidly growing the number of users and demand service providers, the backbone network is improved by deploying a large number of routers, high-speed transmission, and private network systems for more efficient data delivery (Gill et al., 2008).

Conventionally, the content delivery techniques use the client-server model. However, moving the content from the original server to the edge of the Internet (known as local replica server) is a key solution, which has better performance in terms of lower access latency, higher data transfer rate, and less cost than the client-server model. Content distribution network (CDN) through IP multicast is a typical example to address the primary challenge of the Internet (Mosko, 2015). CDN distributes the content from the original server to the end-users through the replica servers that aim to solve the backbone network bottleneck and provide a better quality of service. The contents that are stored and served at replica servers are carefully selected so that the hit rate can approach 100% in some cases. That is to say, CDN can lead to short access delay, increase content distribution rate, and reduce network bandwidth usage significantly (Abu et al., 2014; Halloush et al., 2017; Urueña et al., 2017).

However, with the growing traffic volumes and the increasing challenges of quality of service (QoS) requirements, the CDN performance requires a high cost for deployment of a huge number of edge servers as well as maintaining its operations (Elkotob & Andersson, 2012). Moreover, the agility in server deployment is restricted, which takes quite a long time to install the servers and select the right locations with necessary capacities. The lack of information about the actual network conditions is also a disadvantage of CDN. The characteristics of the paths between the CDN servers and the end clients are not adequately measured and reported when a user assignment is implemented by the CDN, which causes additional load on existing network bottlenecks (Cheng et al., 2016; Mangili et al., 2016). With the explosive growth of network traffic in the past few years, the existing CDN systems have faced IT infrastructure limitations and insufficient storage space.

In addition, cloud computing provides elastic infrastructure and pay-as-you-go, which have great scalability and ability to process massive data (Li et al., 2017; Modi et al., 2013; Smara et al., 2017). Hence, cloud computing becomes a suitable solution for high scalability and load balancing of the CDN network. However, the latency distribution of the end-user and data congestion at cloud egress problem due to highly concentrated cloud computing resources in the cloud data center has not been given enough attention (Ling et al., 2013). In contrast to conventional approaches, peer-to-peer (P2P) solutions have illustrated the major benefits of content delivery and sharing applications, in which each peer contributes its resources to the streaming section (Liben-Nowell et al., 2002; Magharei et al., 2013; Rodrigues & Druschel, 2010) 16]. However, there are significant limitations when the number of users requires data at the same time due to the bottleneck link, peer selection, and high QoS requirement. Therefore, these solutions are only temporary works and, of course, the existing Internet architecture still faces great challenges.

ICN is emerging and promising as a future Internet model that has shifted from host-based communication to the user-driven data retrieval (Borgia et al., 2022). The foundation of the ICN can be started from TRIAC (Cheriton & Gritter, 2000). The next development is Data-Oriented Network Architecture (DONA) (Koponen et al., 0000), and the closest related technologies are Content-Centric Networking (CCN) () or Named Data Networking (NDN) (). In ICN, users are only interested in what content is instead of where the content comes from (G. Zhang et al., 2013b). Inspired by the fact that the current Internet’ demand is increasing for data dissemination instead of using host-to-host communication, ICN aims to reflect the urgency of a new alternative model to use better than the existing architecture. The interest-oriented networking uses the dynamic content caching to enable reliable and scalable content delivery by naming information at the network layer, therefore it accelerates the efficient and timely delivery of information to the end-users. However, there is more than one approach to the distribution of ICN initiatives related to research using the perception of information, such as the means to solve a series of additional restrictions on the Internet architecture (e.g. mobility management and security enforcement) (Xylomenos et al., 2014). A key feature of ICN is to take advantage of the feature of in-network caching to improve the efficiency of content dissemination and information retrieval from the network entities. From caching point of view, ICN cache has several new characteristics, such as cache transparency to applications, cache ubiquity, and fine-granularity of cached content, which are different from Web caching and CDN caching (Caldeira et al., 2015; Chai et al., 2013; H. Lee & Nakao, 2013). Beside the opportunities to the development of ICN caching technology, this feature pose some challenges involving caching dimensioning, caching sharing mechanism, and caching decision policy.

Looking at the ICN concept, many approaches have been investigated, such as TRIAD (Cheriton & Gritter, 2000), Data-Oriented Network Architecture (DONA) (Koponen et al., 0000), Content-Centric Networking (CCN) which is also known as Named Data Networking (NDN) (Jacobson et al., 2009), the Publish Subscribe Internet Technology (PURSUIT) (), the Publish Subscribe Internet Routing Paradigm (PSIRP) (), and Service-Centric Networking (SCN).

By improving the distribution and retrieval of contents, ICN has brought tremendous benefits for many challenging networks such as MANET, VANET, IoT, and WSN. In these network integrations, caching approaches play an important role in the efficiency of content distribution. Generally, the efficiency of different caching approaches depends on the peculiarities of the applied networks, the features of the required contents, and characteristics of the intermediate nodes. This survey gives a picture of the current ICN in-network caching in different networks applications. In contrast to some ICN working surveys that focus on an overview of ICN architecture (Ahlgren et al., 2012; Braun et al., 2013), energy-efficiency solution (M. Zhang et al., 2015), and caching policies (caching model and caching level of operation) and forwarding mechanisms (Fang et al., 2014), our work provides different survey on caching policies from the perspective of the application. In particular, we make the following contributions:

Firstly, we show an overview of the recently highlighted caching approaches, including the fundamental ideas, their advantages, and disadvantages. Secondly, we provide a taxonomy of the existing caching methods based on the caching criteria used. We provide a taxonomy of the existing caching policies according to the caching criteria used. In general, in-network caching can be classified into several groups based on their properties related to the content, the node, and the network that mainly affect the caching performance. Finally, a discussion of the challenging caching issues in different applications is used to highlight the current trends of caching approaches, and the future research directions for each network application.

The rest of the paper is organized as follows. Section 2 presents an overview of ICN that includes the architecture communication model and workflow of the typical information-centric networking approach. Section 3 shows the ICN-based caching for different network applications. Accordingly, the in-network caching problems in different environments are described in Section 4. Section 5 highlights the current trends of caching approaches and future research directions for each network application. Finally, the paper is summarized in Section 6.

2. Information centric network: an overview

In this section, we briefly introduce the components of ICN and present the architecture and workflow of the typical ICN approach to lay the foundation for the in-depth discussion of this paper.

The ICN showed a sufficient system model. The ICN is refined to execute the productive reuse of smart caching of popular content near the demanding users. In regular ICN, there are three primary data formats inside each node: the forwarding information base (FIB), pending interest table (PIT), and content store (CS) (Koponen et al., 0000).

Content Store (CS): CS is an organized buffer memory that is used to retrieve content by prefix matching lookup on names. ICN Content Object messages have many advantages, such as self-identifying and self-authenticating. Therefore, each packet is potentially useful to many users through reuse activity.

Pending Interest Table (PIT): PIT is a source table containing the unsatisfied forwarding information of interest packets. Each entry (e.g., data packet or Interest packet) is queried and processed in PIT. Each entry perhaps points to a source list and has to be timed out instead of being held indefinitely.

Forwarding Information Base (FIB): FIB is an outbound table dace for interest packet that is organized to retrieve by looking up the longest prefix match in names. When the content is not hit in the content store and no entry in PIT, the interest packet will be forwarded upstream to the potential sources by using FIB. Each prefix entry perhaps points to a list of faces instead of just one.

In ICN, the packets determine the content under two types, respectively, called Interest packet (IntPk) and Data packet (DataPk). When an ICN node receives an IntPk, it tracks down the CS. The DataPk will be sent as a request if an appropriate content-id is found; otherwise, the IntPk will be checked on the PIT (Figure 2a). After PIT creates a new entry for an ungratified IntPk, the IntPk is forwarded upstream towards a potential content source that is based on the FIB. A returned DataPk will be sent downstream and stored on the CS. When CS gets full or the content is invalid, the caching strategies, and caching policies (e.g., Least Recent Used (LRU), Least Frequently Used (LFU), First In First Out (FIFO), etc.) will be used to leave the space for new content to leave space for the new content. In summary, ICN-based caching operations and the ICN follow diagram (Doan Van & Qingsong, 2018; Ioannou & Weber, 2016) are shown in Figures 1 and 2.

Figure 1. ICN-based caching.

Figure 2. ICN follow diagram (Doan Van & Qingsong, 2018).

3. ICN-based caching to different network applications

Due to its great benefits, ICN is one of the promising future Internet architectures, which has been actively investigated in many applications. In this section, we demonstrate some ICN-based caching for different network applications (e.g., MANET, VANET, IoT, and WSN) that face many challenges in data content delivery.

3.1. ICN-based caching to mobile ad-hoc network

MANET is characterized by a non-infrastructured, autonomous, heterogeneous, self-configured, and unstable network topology. This network uses mobile nodes (MNs) as routers, and communications occur in a multi-hop fashion. These characteristics of the network lead to many challenges for data delivery to consumers. Also, all MNs have limited resources and are connected to exchange information through bandwidth-constrained variable capacity links (Ahmed et al., 2016). Therefore, there are a lot of barriers when applying the traditional IP-based ad-hoc environment.

Recently, ICN applications are mostly used in Internet-based name-based routing and in-network caching. In particular, the location- independent naming can revoke the demand for reassigning the host identifier, the location-independent naming can revoke the demand for reassigning the host identifier to the moving node/device. Besides, in-network caching can reduce redundant requests and decrease content retrieval delay since each node can exploit the full potential of the broadcast nature of wireless channels. As consistent with the requirements of ad hoc networking, many researchers have proven the advances in ICN for MANET in developing the routing protocols for improving the MNs connectivity (Amadeo & Molinaro, 2011; Amadeo, Campolo, & Molinaro, 2015; Meisel et al., 2010a; Priyanshu & Maurya, 2014; Siyu Yao et al., 2013; Varvello et al., 2011). Following is a brief recent work on ICN-based schemes for MANETs.

The authors in (Priyanshu & Maurya, 2014) proposed to listen first, broadcast later (LFBL) forwarding scheme for MANETs that is based on the named-data networking (NDN)-based.

This work in (Amadeo, Molinaro, et al., 2013) develops analytical models for a content-centric MANET (CCM) by evaluating the suitability and effectiveness of various existing designs. These models are identified to retrieve content in the MANET, which includes reactive flooding, proactive flooding, and the design uses a geographic hash table (GHT).

An architecture for oriented content in the MANET environment is content-centric fashion MANET (CHANET) was proposed in (Amadeo & Molinaro, 2011). This approach is built on a connectionless layer designed on top of the legacy IEEE 802.11 standard to provide routing based on the content and transport functionality. This approach is used to cope with wireless link impairments and issues arising from dynamic topology. The authors in (Amadeo, Molinaro, et al., 2013) proposed E-CHANET by extending the preliminary work of CHANET (Amadeo & Molinaro, 2011). This approach thoroughly solves the problems concerning routing schemes, reliable transport functions, and the limitations of wireless distribution domains (e.g., unreliability channels, broadcast storms, dynamic and unpredictable topology, etc.)

Another content-centric design for MANET called SCALE was presented in (Varvello et al., 2013). The main idea of this scheme is to use routing based on content names to provide local by using the geographic hash table (GHT)-like resolution and improve efficient content retrieval by using the popularity of content as the basic criterion to switch between two schemes (named-based routing and GHT-like resolution). A MANET publish-subscribe system exploiting the ICN scheme was discussed in (Detti et al., 2015). This system uses the hierarchical naming scheme and in-network caching to achieve the efficiency and reliability of content dissemination. The authors in (Amadeo, Campolo, & Molinaro, 2015) proposed two forwarding strategies in named data wireless ad-hoc networks. The provider-blind forwarding strategy is intended only to keep packet redundancy in a wireless radio environment without the knowledge of neighbourhood and content source identification. A provider-aware forwarding strategy is used to facilitate content retrieval by leveraging soft state information (e.g., source content, piggybacked, IntPk, DataPk, and locally kept by the nodes).

Typically, the forwarding schemes of ICN nodes in multi-hop wireless communication rely on broadcasting. Unicasting is a potential scheme to eliminate the broadcast issues of packet redundancy and unreliability. Therefore, the authors in (Amadeo et al., 2017) designed a forwarding strategy for ICN in wireless ad-hoc networks, termed Ad-hoc dynamic unicast (ADU). ADU helps to reduce the latency and solve flooding request packets, improve the Quality of Service (QoS) and efficient energy consumption. The authors in (Liu et al., 2017) investigated the impact of correlated mobility on the delay and the throughput performance of MANET in the ICN environment. Two regimes (the cluster-dense regime and the cluster-sparse regime) are investigated based on the degree of correlation among nodes. The results showed two important things: i) the correlated mobility improves the delay performance at the cost of throughput performance under fast mobility; ii) the correlated mobility and mobility of the nodes have a negative impact on both the delay and throughput performance under slow mobility.

Most recent works in this network are concerned in developing the routing protocols by proposing new schemes as mentioned above. Meanwhile, in-network caching is also a very important aspect, and a useful trend in MANET for content dissemination, and reducing access latency. So far, the current works on caching has mainly concentrated on some simple caching decisions and replacement policies.

A novel location available content management caching approach (LACMA) in (S. -B. Lee et al., 2013) was proposed by leveraging the location information available to mobile devices through GPS. This location-based approach helps to blind data content to geographic location. Thus, it allows to separate content placement problems from the dynamic network topology.

The work in (Anastasiades et al., 2014) proposed agent-based caching for the opportunistic network where the users can delegate content retrieval to their neighboring nodes. Leveraging the broadcast nature of the radio channel to improve the cache gain of ICN ad-hoc networks by reducing the utilization of the network cache space was proposed in (Zhou et al., 2015), called broadcasting-based neighbourhood cooperative caching (BNC) strategy.

Some powerful nodes will be selected for the storage of the content to solve the storage limitation. Less space still faster (LF) strategy (L. Zhang et al., 2015) is to choose the appropriate nodes for caching on the transmission path that is based on some criteria, such as the available space of the node, distance from the forwarding node to requesting one, the data popularity, and the duplicate storage. Our previous work (Doan Van & Mau, 2016) gave a novel Multi-Source Content-Centric Networking (MS-CCN) by using the concept of Multi-Source Mobile Streaming (MS²) with CCN scheme. In the MS-CCN model, the caching of each content is not limited to a single server anymore; instead, each content is fragmented and distributed to multiple servers over a large scale network. After experiencing disjoint multi-paths, various content fragments are concatenated at the client-side.

To solve the limitations of the storage cache space of the nodes, some powerful nodes will be selected for the storage of the content. A less space still faster (LF) strategy (L. Zhang et al., 2015) is to choose appropriate nodes for caching on the transmission path that is based on some criteria, such as the available space of the node, the distance from the forwarding node to requesting one, the data popularity, and the duplicate storage. In our previous research work (Doan Van & Mau, 2016), content-centric networking model is combined with the advantage of multi-source mobile streaming. Each fragmented content is distributed in storage at multiple servers. The fragmented content will be reordered at the end users after passing through disjoint multiple paths. Better Experience of Service (EoS) is gained in case large numbers of users request popular content at the same time. In (Kalghoum et al., 2017), a caching replacement policy is given based on the data popularity by the Software-Defined Networking (SDN) architecture to reduce bandwidth consumption and enhance the management performance of caches in named-data networking (NDN). The SND architecture segregates the underlying infrastructure from the application that uses it. Therefore, this architecture is a great solution to apply for an efficient replacement management system in mobile networks.

3.2. ICN-based caching to vehicular ad-hoc network

With the increased demand for vehicles, the driving and traveling experience is very necessary to improve. Vehicles need special networks to connect each other and share essential information while moving. Vehicular Ad-hoc Network (VANET) is one of them that commences from a traditional MANET model. Like mobiles, each vehicle works as a sender, receiver, and wireless router. Some particular connection technologies in VANET are desirable to fulfill the increasing content of users on wheels including Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2I), Vehicle-to-Roadside units (V2R), and Vehicle-to-Everything (V2×) (Amadeo et al., 2016).

The vehicular network is a specific version of the Ad-hoc model. This network has some peculiarities like dynamic topology, short-lived and unreliable connections, and high-speed moving vehicles. This poses many challenges for the existing IP-based networking solutions in supporting a wide range of emerging vehicle applications. Vehicle applications are typically strict safety applications. However, the connection environment is poor and intermittent due to high-speed mobility and lack of road side units. Therefore, the existing IP-based network is not well suited for VANET.

Recently, new research has focused on networked communication to limit the negative impact of the above network-specific properties. The efficiency of distribution content of ICN in the VANET domain has been intensively researched, ranging from pre-fetching to cooperative caching strategies. The popularity-driven caching approach was shown to dynamically place the replicas of the caches at selected routers along the path to minimize the inter-Internet Service Provider traffic and reduce latency (Jun et al., 2012).

Similarly, popular content is selected in the most popular content (MPC) caching strategy that allows to gain high performance, high-hitting ratio, release node’s memory, and reduce offloading routers (Bernardini et al., 2013). Manage cache in ICN by combining cache development strategy and cache replacement policy called Sate-value-and-Cache-rate-Method (SCMethod) strategy in (Man et al., 2020). Selecting the popular content, the pre-filter queues are set up in front of the cache queue to reduce latency and effectively mitigate data redundancy.

A probabilistic in-network caching in ICN is proposed, called ProbCache (Psaras et al., 2012). The ProbCache algorithm approximates the capability of the paths to cache the content that is accounted based on path lengths and multiplexes content follow accordingly. The goal of this algorithm is to utilize resources efficiently and reduce network redundancy.

Moreover, a fixed probability is defined as the threshold level for the storage decision (Deng et al., 2016; Laoutaris et al., 2006). When a data packet comes at each router, it randomly creates a probability. The content can only be stored if its calculated probability is lower than the defined probability.

The authors presented the Storage Planning and Replica Assignment (SPRA) approach to utilize a trai-explicit cooperative caching policy (Sourlas et al., 2011). The content in SPRA is digested based on three factors when minimizing the client’s response delay, their popularity, and the overall traffic of the network. This approach helps to reduce the request time and balance the network traffic. However, the communication strategy has not been shown clearly. On the other hand, a cluster-based mechanism for vehicular networks in the Scale-Free ICN score network is proposed (Hasan & Jeong, 2021). This mechanism ensures to address the interest message flooding problem and improve the content caching.

Although the methods given above have shown the effectiveness of network performance, it is not really suitable for the wireless network environment because it is affected by many environmental factors. In particular, they do not consider the characteristics of the VANET domain. Usually, the default caching strategy in VANET is also known as ALWAYS caching, where all incoming packets are cached by all nodes. This caching strategy is entirely dependent on caching capabilities. However, this simple caching strategy has shown inefficiencies due to increased costs and reduced data diversity in the network. To date, several advanced caching strategies in VANET have been studied to minimize redundancy and maximize cache hit rates. In (Mauri et al., 2017), the authors proposed an integer linear programming formula for the problem of optimally distributing content to nodes while taking into account available storage capacity and link capacity availability. This is intended to maximize the probability of users retrieving the desired content in a vehicle-to-infrastructure scenario. However, vehicles may not be able to connect directly to roadside devices (RSUs) using poor-quality wireless links and rapidly changing topologies. Therefore, implementing V2V with caching strategies is an effective way to address this challenge. This work in (Tian et al., 2016) proposed a Leave Copy Everywhere (LCE) strategy according to traffic features for the dissemination of neighborhood marketing files in an urban environment. In this case, each vehicle is considered as either a subscriber to request data or a cache node to speed up file transmission effectively. In addition, data delivery is greatly improved as the caching function in Vehicle to Vehicle (V2V) communication instead of a large part of the vehicle-to-infrastructure (V2I) communication. Although this approach is suitable for dynamic vehicular networks, they were not specifically considered for all the key attributes of the peculiarities of VANET applications for caching decisions.

Some typical caching schemes are applied in the VANET environments (e.g. fist in fist out (FIFO) in (Doan Van et al., 2016)). The proactive caching (PeRCeIVE) scheme is investigated in (Grewe et al., 2016). Before a request is sent from a consumer, PeRCeIVE includes some actions like pre-caching content near the consumer, and decreasing caching resource utilization. In (Quan et al., 2014), an ICN-based collaborative caching solution (specifically ICoC) was proposed based on two new social collaboration schemes (partner-supported and courier-assisted) to enhance information-centric caching by improving the quality of experience (QoE) of media streaming services.

3.3. ICN-based caching to vehicular internet of thing

The Internet of things (IoT) is a network of physical objects (called things) which have billions of devices embedded with sensors, actuation, and software for the purpose of connecting and exchanging data with each other over the Internet (Correia et al., 2017; Lin et al., 2017; Whitmore et al., 2015). When billions of devices exchange data at the same time, some issues arise from resource-constrained devices (e.g. low power, limited memory, slow processing capabilities) and heterogeneous access technologies. Several approaches and standardization activities have been proposed to temporarily solve these challenges for the IoT environment using IP-based networking functionalities (e.g. CoRE, RoLL, 6LoWPAN approach) (Sheng et al., 2013). In order to simultaneously service a huge amount of processing under the stringent requirements of IoT and multicast in the case of large-scale deployment, it requires a comprehensive solution. Beside technology solutions, a connectivity model is the first necessity to deploy. ICN technology is considered as a fully promising solution in the future. The advantages of ICN are also utilized to provide content-based security, native multicast support, as well as in-network caching by leveraging unique, persistent, and location-independent content names, which can be specifically useful in the IoT (Amadeo, Campolo, Iera, et al., 2015).

With regards to in-network caching, an ICN-IoT integration network helps to shorten data retrieval times, reduces multi-path, and improves energy efficiency. However, IoT is a challenging environment as mentioned before, characterized by generated data contents such as small size, “transient” short life-time to device capabilities, such as constrained energy and limited storage size. So far, there is no comprehensive methodology for caching decisions. It just focuses on a few single metrics, such as data refresh, request rate, cache size, and even communication costs, which lead to incorrect caching decisions.

In ICN, an important issue in on-path caching is how each node makes a caching decision to increase the cache hit rate of content delivery, improve network resource utilization, and speed up content distribution (Fang et al., 2015). Several caching decision and replacement strategies have been applied to solve this problem (Feng et al., 2014; G. Zhang et al., 2013a; Kim & Yeom, 2013; Yan et al., 2015). These traditional strategies are mainly applied to static data items (e.g., multimedia files), popular contents, and a sheer number of users as a feature of content delivery and request forwarding systems. However, they are inefficient and quite expensive due to the peculiarities of the IoT environment. Only a few preliminary studies focus on caching in IoT systems but do not propose a different and comprehensive caching policy (Baccelli et al., 2014; Hua et al., 2020; Quevedo et al., 2014; Song et al., 2013).

The feasibility of deploying ICN in-network caching for IoT has started gaining momentum by the research community. The Information-Centric Networking Research Group (ICNRG) (Baccelli et al., 2014) has been proposed as a network paradigm that enables native content awareness by an underlying network including content searching/resolution and caching. In (Baccelli et al., 2014), we evaluate the performance of content retrieval from different consumers using standard NDN in-network caching. In this work, caching is illustrated to be highly beneficial even with constrained nodes (e.g., small storage capacity) by comparing the implementation of data retrieval from various consumers in the presence of standard NDN in-network caching and when caching is disabled. However, the cache size of resource-constrained nodes (used in the experiments) is 1 Kbyte, and the information is ephemeral (short-lived, transient). Also, the authors in (Quevedo et al., 2014; Song et al., 2013) also built a scheme for constrained-resource devices based on the key idea and fundamental CCN model for more efficient edge networks, in which weak network devices with constrained resources map the overcapacity task in terms of storing, publishing, and retrieving supper routers/strong devices. This improves the accuracy of the data received by consumers. The challenge here is how this caching approach is sustained in the presence of a large number of consumers with different freshness requirements.

Since IoT content is transient, information freshness is a very important parameter that highly affects the performance of caching strategies existing in the ICN approach. In (Quevedo et al., 2014; Song et al., 2013) a new consumer-driven freshness approach for ICN was analyzed. In this scheme, content can be considered as valid in the content store (CS) by establishing a certain period of time. Old contents are dynamically discarded from a CS that honors the freshness declared by the producer. As a result, a freshness value included in both Interest and Data packets were explored to carry out the accuracy of caching and retrieval operations. Similar to (Khedher et al., 2017), the optimal placement algorithm (OPA) is proposed to select the optimal placement location for ICN nodes in the IoT environment based on some parameters (e.g., the required consumer end-to-end response time, memory, and CPU resources of the nodes, and migration cost).

On the other hand, the work in (Vural et al., 2014) proposed a distributed probabilistic caching algorithm for more efficient usage of the available caching space, alleviating the load of nodes and bandwidth consumption, where nodes automatically update their caching probability by using information about the network topology, the freshness value, and the rate of incoming requests. However, this scheme was mainly designed for the case of large storage capacity; it was not well supported for constrained devices and low power networks. To solve this limited storage, different caching and replacement policies were evaluated in the ICN-IoT wireless network in (Hail et al., 2015). In particular, probabilistic caching coupled with LRU gives the highest performance in terms of retrieval delay and Interest retransmission. Conversely, “always” caching strategy leads to a high level of content redundancy and poor utilization of available cache resources. Considering the challenging features of IoT (resource-constrained devices, transient data), our previous work in (Doan Van & Qingsong, 2018) has proposed an efficient in-network caching decision algorithm that considers a set of key attributes of the IoT environment. Fog caching scheme enabled by ICN for IoT is cluster-based scheme, in which in-network and end-user devices are utilized to cache content close to the edge network by leveraging cache nodes, the content near the path, and popularity of the content (Hua et al., 2020). However, in the case of unpopular content or popular content being scattered in the network, the near-path attribute will take a long time to search at each node until a cache hit. This leads to an increased latency that is due to the low quality of service.

So far, most ICN in-network caching in IoT proposals is still in the early stages of implementation and a very limited number of studies have only focused on this domain, both theoretical research and experimental implementation. Meanwhile, the existing caching decision and replacement policies normally focus on the simple caching strategies (AlwaysCache-FIFO, AlwaysCache-LRU, etc.) with some key features of the content separately. Therefore, there are several other issues related to caching approaches that must be addressed when adopting ICN in IoT.

3.4. ICN-based caching to wireless sensor network

As a key part of IoTs, wireless-sensor networks (WSNs) are the set of a huge number of small battery-operated sensing devices that have communication capabilities. Real-life applications of this network have been widely used in many areas, including environmental monitoring, emergency response, agriculture, industrial automation, and aeronautics. The goal of WSN is to oversee the field for a lengthy period of time to gather sensor information for monitoring, processing, and operation analysis. The dynamic network architecture of WSN frequently changes due to node failures. Therefore, the nodes in this network must self-organize and self-configure for a longer lifetime (Kafi et al., 2014; Mohamed et al., 2017).

Recently, ICN communication was investigated in WSN through efficient architecture (Abidy et al., 2014; Amadeo, Campolo, et al., 2013; Bernardini et al., 2015; Doan Van et al., 2017; Mohamed et al., 2017; Singh & Al-Turjman, 2016; Tilak Singh & Al-Turjman, 2016). The ICN scheme is applied mainly for many networks to efficiently distribute the content (e.g., multimedia files that are required by a huge user) as shown in the aforementioned sections. However, from a name-based point of view, ICN has brought many benefits to WSN. Since the exchanged information is named and hence recognizable at the network layer, the different network individuals identify the type of traffic and utilize different caching/routing policies to guarantee optimal network performance. This feature is a key factor in approaching ICN to WSN scenarios where it is important to deliver different treatments to various types of information (e.g., urgent event alerts in the healthcare domain). On the other hand, another issue appears in WSN when considering the important features of the content, such as the freshness (where the consumers are mainly interested in the latest information), the delay time of sensor nodes, the energy of the node, etc. In this regard, the cached data does not seem to be a value with the classical caching methods. It would require more effort to propose more content distribution efficient ICN-based solutions for WSNs.

In general, the cache size of a storing node is expected to be infinite. Hence, the ALWAYS caching (Amadeo, Campolo, et al., 2013) is commonly utilized as a default caching approach, where all incoming packets are stored without any selection. However, the cache size of a node is restricted in a practical scenario. A caching strategy is crucial to preserve valuable data packets in a cache with a restricted size. However, the current replacement strategies, i.e., least recently used (LRU), first in first out (FIFO), etc., have been applied mostly to IP-based networks and data centers (Doan Van et al., 2017), which are not appropriate for WSN. Furthermore, in some sensing events, the gathered data have to be provided in a timely manner for quick action from a specialized department of the application domain. In addition, the current caching strategies may not store valuable data in caches. These factors will result in the low utilization efficiency of the stored contents for WSN data traffics. Generally, an appropriate caching strategy for WSN depends on metrics, namely network delay, data rate, and the popular content (Hajimirsadeghi et al., 2017; Lim et al., 2014; Vural et al., 2017; Wang et al., 2012). In (F. Al-Turjman, 2017) proposed a cognitive caching strategy for the future fog (CCFF) that considers the value of the exchanged data in information-centric sensor networks. These main factors, which are the age of the data, popularity of on-demand requests, delay to receive the requested information and data fidelity are computed to assign a value for caching decisions while maintaining the most valuable one in the cache for lengthy periods. Similarly, the study in (F. M. Al-Turjman et al., 2017) considered a set of parameters for cache replacement to remove the invalid content from the cache. Nevertheless, this approach does not consider if the cache has enough space to store the value of an incoming data item. Also, the priority level of the selected elements in each type of application is not described in detail, thus affecting the precision of the caching strategy.

The ICN-based caching strategies for different network applications discussed previously in section 3.1, 3.2, 3.3, and 3.4 are summarized in Table 1.

Table 1. ICN-based caching for different network applications

Short Form	Caching Scheme/Architectures	Characteristics	App.	Ref.
LFBL	Listen first broadcast later	Reliable data forwarding	MANET	(Priyanshu & Maurya, 2014)
CCM	Content centric MANET	Reactive, proactive flooding; and the design uses a geographic hash table (GHT)	MANET	(Varvello et al., 2011)
CHANET	Content centric fashion MANET	Hierarchical naming, chunks content, broadcast-nature interest and data message	MANET	(Amadeo & Molinaro, 2011)
E-CHANET	Enhancing CHANET	Routing, forwarding, and reliable transport functions	MANET	(Amadeo, Molinaro, et al., 2013)
SCALE	Content centric MANET	Named-based routing and GHT-like resolution	MANET	(Varvello et al., 2013)
MANET publish-subscribe	MANET publish-subscribe system	In-network caching and multicasting	MANET	(Detti et al., 2015)
LACMA	Location available content management caching approach	Leveraging the location-based information available to manage caching content	MANET	(S. -B. Lee et al., 2013)
-	Agent-based content caching	Delegate content retrieval to the neighbor nodes from customer	MANET	(Anastasiades et al., 2014)
BNC	Broadcasting based neighborhood cooperative caching	Improve cache gain of ICN-ad-hoc wireless network by the broadcast nature of radio channel	MANET	(Zhou et al., 2015)
LF	Less space still faster	Select the appropriate nodes for caching on the transmission path	MANET	(L. Zhang et al., 2015)
SDN	Software-defined networking	Cache replacement strategy based on the data popularity driven	MANET	(Kalghoum et al., 2017)
DPC	Distributed Probabilistic Caching	Caching probability based on users’ demands mined, vehicle’s importance, and relative movement	VANET	(Deng et al., 2016)
-	Cluster-based machanism	Solve the interest message flooding, efficiency the content caching strategy	VANET	(Hasan & Jeong, 2021)
-	Optimal content prefetching in NDN V2I	Integer linear programming formulation of the problem of optimally distributing content, account the available storage capacity and the available link capacity	VANET	(Mauri et al., 2017)
LCE	Leave copy everywhere	Leave copy everywhere strategy for disseminating proximity marketing files	VANET	(Tian et al., 2016)
FIFO	First in first out	Address the storage content constraint	VANET	(Doan Van et al., 2016)
PeRCeIVE	PRoactive Caching	Calculates an output containing the list of chunks and an optimal distribution strategy of the Chunks along the available RSUs	VANET	(Grewe et al., 2016)
ICoC	ICN-based cooperative caching	Novel social cooperation schemes (partner-assisted and courier-assisted)	VANET	(Quan et al., 2014)
-	On-path caching	Storage content on the delivery path	IoT	(Fang et al., 2015)
-	Resource-constrained devices	CCN scheme for weak network nodes based on task mapping	IoT	(Song et al., 2013)
OPA	Optimal placement algorithm	Select the optimal placement location no.des for caching based on the response time, memory, and migration cost	IoT	(Khedher et al., 2017)
-	In-network caching IoT data	Trade-off between multihop communication costs and the freshness of a transient data item	IoT	(Vural et al., 2014)
-	In-network caching ICN-IoT	Probabilistic (LRU) outperforms Probabilistic (RAND) and Always (LRU; RAND) in ICN-IoT	IoT	(Hail et al., 2015)
-	New consumer-driven freshness	Valid content is stored in CS during a certain period of time	IoT	(Doan Van & Qingsong, 2018)
-	Frog cluster-based	Cache content close to the edge network by leveraging cache nodes the content near the path and popularity of the content	IoT	(Hua et al., 2020)
-	Caching policies ICN-WSN	Caching depends on a few metrics such as network delay, data rate, and the popular content	WSN	[94- 97]
CCFF	Cognitive caching strategy for the future fog	Caching decision based on the age of the data, popularity of on-demand requests, delay and data fidelity	WSN	(F. Al-Turjman, 2017) (F. M. Al-Turjman et al., 2017)

4. Caching challenges in different applications

Once ICN has been created and developed, so far, there remains a lot of internal challenges that have not been resolved effectively.

The main challenges of ICN are issues in naming, name resolution, routing, security, energy consumption, and caching. Among them, cache sharing is an aspect that so far has obtained relatively little attention. This is partly because of the large number of open challenges (Ghodsi et al., 2011).

The caching aspect will be discussed in detail in this section to actualize the goal of ICN in challenging applications. Study on in-network caching strategies is essential in many networks due to its potential to enhance the delivery of service (DoS), minimizes total bandwidth usage, and reduces latency in a scenario where most of the content requirements originate from the wireless networks including node-constrained resources like IoT environment. Generally, in-network caching includes two key parts: i) Caching decision, which decides whether a data item should or not to be stored, and where will the data item to be stored, and ii) Replacement policy, which is used to retain the valuable data item in the cache (placed) and remove the unnecessary or invalid items (replaced) and to determine when is the most appropriate time for caching and for how long. The aim of in-network caching is to answer the questions including which the entire content to be selected for caching when content is cached and for how long?

Based on these motivations, different caching strategies have been proposed as in the aforementioned analyses in sections 2 and 3. The top sensitive issues that influence the active deployment of caches in ICN are the nature of the content, the characteristics of network applications, location of the nodes, cost, energy consumption, etc. These features are investigated independently and separately in most of the existing caching strategies, which aim to optimize the cache space and efficient use of the stored contents. Therefore, the characterized challenges of each specific network application are not fully accessed through caching strategies. As a result, the benefits of in-network caching are limited. Figure 3 presents a taxonomy of the existing caching policies according to the features-based caching as criteria used.

Figure 3. A taxonomy of the existing caching policies.

5. Open issues and future research directions

To open new research avenues toward a more intensive and comprehensive understanding of caching, this section highlights major issues in ICN when applying in-network caching of ICN into MANETs, VANETs, IoTs, and WSN environments. It raises many challenging issues not only due to the ICN architecture but also caused by the peculiarities of network applications, the features of the data items, and the end user’s requirements, etc. The following issues that influence caching performance in different network applications should be further studied.

Splitting content in chunks: The ICN content is fragmented into many small-sized ones that are called chunks. Different chunks of the same object can be cached at different nodes, and they are concatenated to complete content at the client-side. This phenomenon may affect data retrieval, especially in the case of a single-path forwarding strategy where it maintains sending the IntPk to the first found content source (Meisel et al., 2010b). In addition, the ICN requests for incessant chunks of the same object are correlated; hence, the independent reference caching model no longer holds (Rossini & Rossi, 2013).

Caching redundancy: The caching redundancy is a metric defined as the number of duplicate/redundant content stored in a network. It is a criterion used in the evaluation of the caching policies. Many existing solutions have made considerable achievements in this aspect to minimize caching redundancy in-network caching. In fact, many remaining challenges are mainly caused by two cases. In the first case, when many users require cached content at the same time from different nodes in a network, it is easy to increase the redundant cached content. In the second case, the content is fragmented into many chunks that are stored as scattered segments on the different nodes in the network. Hence, the redundancy of chunks in the same or different content is further aggravated. So far, several solutions have been proposed, such as cache the popular content, cooperative caching among the nodes, etc. However, it does not completely solve this problem, especially in highly mobile environments.

Dynamicity of wireless networks: Network topologies are not stable due to node mobility. This dynamic change depends on the specific network applications (low-to-medium in MANET, WSN, and IoT for high mobility in VANET). Topology changes lead to unreliable broadcast channels, short-lived, and intermittent connectivity, which have negative effects on the caching performance. In addition, the ICN mobility feature may reduce the information monopoly and the privacy rights of users.

Resource-constrained devices: A significant feature of ICN against IP-based models is in-network memory. Data is mainly stored in the core network. Therefore, the embedded computing of ICN nodes/devices in network applications is expected to be resource-constrained. This resource constraint is caused by small energy, low processing power, and limited storage capabilities. Concretely, while on-board units in VANET are not limited by energy and storage capability, the sensor nodes and RSU are constrained because their batteries are not rechargeable or renewable for the long term. Meanwhile, wireless nodes in WSN are battery-powered devices with limited storage capacities and processing power. These constraints are especially critical in the IoT environment.

Cache location (or different degrees of node mobility): Cache location in ICN is an influencing factor for bandwidth utilization, bottleneck links, round-trip time, as well as overall network delay. Therefore, this aspect has gained a lot of attention in the ICN research community. From the perspective of location, the caching strategy is classified into centrality-based caching and edge caching. Centrality-based solutions aim to select some nodes that have a higher probability of getting a cache hit for storage by leveraging the node betweenness centrality in a network topology (Caldeira et al., 2015; Rossini & Rossi, 2012). The disadvantage of centrality-based solutions is limited scalability due to their centralized features. Consequently, they can be easily applied only to the mesh network topology, and are unsuitable to be applied to other large-scale networks and mobility scenarios where the concept of node centrality is ambiguous.

Concerning the edge caching, the authors in (Eum et al., 2015) explored a scheme for the location of cache-enabled routers in the transmission path. In this scheme, to reduce the data retrieval delay and improve the content dissemination in networks, the closer the nodes to the consumers, the probability to cache the data items will be higher. In fact, deployment of the edge caching is more effective than the centrality-based caching. However, in cases where the content is unpopular (e.g., emergent signals in WSN) and transient (e.g., the stringent requirements on the content freshness in IoT), and the users are always with high speed (e.g., vehicles in VANET), the cache location near the users may cause waste cache resources that need to use a lot of nodes cache at the edge network. In addition, it cannot afford the storing of content for the long term.

Selecting some important nodes for more storage size than the others is investigated in (T. Zhang et al., 2019). Although this scheme derives the minimum usage of cache size capacity while improving the retrieval content of users, the problem here is how to determine the important nodes for each challenging application network, which is dominated by many other features in caching performance. Concretely, the important nodes in IoT and WSN may be the resource-constrained nodes that have superiority in terms of energy, cache size, and computing capacity. Meanwhile, the nodes considering some features that afford the fetching content during high-speed mobility are more important than others in mobile networks and VANET environments.

On the other hand, the optimal strategy for provisioning the storage capacity is proposed by global coordinating the in-network storage capacity in (Li et al., 2015). Although these solutions have been deployed to select the right location for in-network caching, the effectiveness of caching performance is significant when considering the peculiarities of specific application network context.

The storage and computing capabilities of the nodes: The cache size determines the space available in an ICN node for storing temporal chunks of content that is used to decide how much the content can be stored. Along with the development of technologies, the cost of deploying storage devices in the application networks is not too expensive. Therefore, in general, the limited cache size will not be a serious problem. However, it still faces challenges in the trade-off between the benefits of caching strategy and the cost of resource consumption for some special application networks. For example, in IoT, billion nodes/devices (called resource-constrained devices) are interconnected with a limited resource, such as storage capacity. These resource-constrained devices usually do not have large enough storage to cache the contents (e.g., multimedia files) that they have produced. Similarly, the cache size of sensor nodes in WSN is bounded due to the size-limit of wireless sensor nodes.

Caching latency: for on-path caching, the caching performance takes place on the nodes when the content is traversed toward the consumers. The time required for the lookup of content, caching decision, and replacement operation is the caching latency, which is being added to the content retrieval time. If the content is transient and usually expires in a short period of time, the caching latency is limited to the time that the content traversed an intermediate node. It is to ensure that the consumer will receive the content freshness while satisfying the required lifetimes, which are defined and generated by producers or requested by consumers. For example, the consumers are interested in the latest content (e.g., the vital signal of a patient, the emergent information in natural disasters, the latest temperature from sensors in a specified location, etc.). In these cases, the content’s lifetime that the consumers require to be fetched is pre-defined and limited. The consumer may receive the invalid (or expired) content due to the long caching latency (as shown in Figure 4a) (Doan Van & Qingsong, 2018). This is caused by the fact that the required content expires at a time when traversing on the path before coming to the consumer. Vice versa, the required content is valid and non-expiring to use at the consumer as in Figure 4b. Therefore, the implementation of caching strategies is required in the fastest way (shortest caching latency) while ensuring its effectiveness is an important yet challenging problem in ICN network applications.

Figure 4. Data item freshness (a) Fresh (b) Non-Fresh (Doan Van & Qingsong, 2018).

6. Conclusion

In-network caching is the core function of ICN, and many mechanisms have been proposed to implement efficient content distribution. In general, this concept has been investigated in various names, such as PSIRP, DONA, CCN, NDN, and NetInf architectures. Expectations play as the future paradigm to meet the demand of today’s Internet, ICN has been discussed in many specific fields. However, it increased many challenges due to their peculiar characteristics of the network and will need high attention of the research community to bring the results of actualizing ICN. This study focused on an extensive survey about the caching aspect and analyzed the existing caching mechanisms when ICN is integrated into several network applications (MANET, VANET, IoT, and WSN). Hence, this study provided a taxonomy for caching policies according to a features-based approach and highlighted the problems in these areas. The challenging research directions to yield and suggest potential research directions for in-network caching are finally being held.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Notes on contributors

Dong Doan Van

Dong Doan Van received his M.S. and Ph.D. degrees in information and communication engineering from Huazhong University of Science and Technology and Wuhan University of Technology, respectively. He worked at Ho Chi Minh City University of Transport as the Dean of Electrical and Electronic Engineering from 2019 until recently. He is currently the Head of the Training Department at Ho Chi Minh City University of Transport. His research interests include content-centric networking, vehicle ad hoc networks, sensor networks, telecommunications, urban and environmental remote sensing, Internet of Things, and image processing.

Qingsong Ai

Qingsong Ai received the M.S. and Ph.D. degrees in information engineering from the Wuhan University of Technology, Wuhan, China, in 2006 and 2008, respectively. He is currently a Professor at Hubei University. He has authored over 50 technical publications and editorials. He has directed over 10 research projects. His research interests include signal processing, rehabilitation robots, and advanced manufacturing technology.