4.3.1 Classification of CCN Contents and their Mobility

Keeping in view the stability and mobility concerns in CCN, there are two types of contents named as data contents and user contents. Data content is significantly termed as access to the content itself. Persistent content providers are used to serve the data contents and disseminate their replication for the ease of access to content. So, locations are comparatively stable for applications like web pages, streaming services, and files. While user-contents, on the other hand, contain data having real-time characteristics, e.g. Internet phone, chatting, and e-gaming constitutes user-content. Creation and management of the content is performed by individual mobile devices. Content service components such as rendezvous points and control messages are used for serving data contents and user contents. A rendezvous point is basically a location manager used to keep track of locality information of all providers containing location information of peers. Moreover, a rendezvous point is used as mobility handler by intervening in the relay of all interest and response packets. It is also known as indirection point. While control messages address the issues related to changing of content provider's routable prefixes. However, a sender-driven control message is needed for source mobility.

4.3.2 User Mobility

Mobility for the buffered contents in CCN on the user side is supported by using re-transmission mechanism of request and response packets that first come to Internet service provider (ISP). On the caller's hand-off, it receives a sequence number of the contents from a caller. During communication, lost chunks of information data can be recovered by simple process known as retransmission of data packets. User-side mobility deals with persistent interests. Kim et al. sectioned the CCN services as three types: real-time documents, channels, and on-demand documents based on service characteristics [6]. There is difference in persistent interest and the regular interest. In case of persistent interest, after hand-off, the route remains persistent to the old location and responses are multi-casted to the old location. This concept of multicasting results in waste of network resources. To address this, the signaling message can be used if the frame router could not successfully deliver content corresponding to persistent interest. When a threshold is reached, the edge router generates “host unreachable” with the specific name, for example, (persistent interest name)/Host unreachable, and so on. This signaling message is then delivered to the content server. The gateway used to communicate that signaling message can abolish the related obsolete persistent request and responses from their PIT. While the hand-off occurs due to the mobile client, persistent interests are retransmitted to the new path. Process of retransmission includes a dedicated bit that eliminates the outdated PIT entry. If the interest received on content server, it sets it to “on” in the next responses. It requires additional clients to send interests in addition to preserving the valid path to the clients.

4.3.3 Server-side Mobility

For receiving interests smoothly after hand-off, the name of the mobile is required to be updated in the routing table for each individual content. For this purpose, binding service between changing names is required at both old and new locations. To change the content name, the application server, such as a rendezvous point or a proxy agent binding service, are useful to provide binding service. Rendezvous-based approaches are used for changing names and binding services; each is discussed in subsequent sections.

4.3.4 Direct Exchange for Location Update

The hand-off during direct exchange of location updates indicates data source mobility. Callee is then informed about the new classified pathname of the caller for smooth running of ongoing service. Callee can continue getting replies even during hand-off. Recovery of lost data portion by simple retransmission is impossible. In order to recover the lost data, application regenerates the interests with the new hierarchical pathname. With the factors of simplicity and effectiveness, the issues arise when hand-offs are performed simultaneously by the peers.

4.3.5 Query to the Rendezvous for Location Update

On initiation of caller service with the mobile callee, caller queries a rendezvous server for obtaining callee's location record. As location information of all clients is maintained with rendezvous server, so on occurrence of hand-off new path-name of the client is communicated to the rendezvous server. Therefore, using the rendezvous services, a peer can obtain the routable prefix even after simultaneous hand-offs.

4.3.6 Mobility with Indirection Point

Another technique for reducing the hand-off delay is the indirection scheme. To initiate a request, caller issues a request and response packet with the name (callee). On successful receiving of packet, it issues another interest to the callee with its new routable name. On callee's response, the indirection server issues a corresponding response packet that is referred toward the caller. During a caller hand-off, it notifies occurrence of the hand-off event toward the indirection server. However, on receiving interest for the caller, indirection server buffers that interest and delays generating another corresponding interest toward the respective caller. When caller conveys new hierarchical path name to the server, the indirection server continues to generate corresponding interests and its service toward the caller.

4.3.7 Interest Forwarding

In order to cope with the drawbacks in previous schemes, layered approach known as “interest forwarding” is proposed. Unlike the previous schemes, in this approach a new name is not required for seamless service. It requires the modifiable routers, as incoming interests are buffered in CCN routers toward a hand-off client. Overhead caused by addition of new entry routers on the way between the access routers to corresponding hand-offs is somewhat reduced. In addition to this, hand-off process does not require a new routable prefix, while hand-off latency is diminished as well.

4.3.8 Proxy-based Mobility Management

This scheme is based on the concept that user proxies with CCN environment are configured over overall architecture of IP network. Contents are available to users via proxy node when required. User will send query packet to proxy node instead of making connection with all other devices having required contents. CCN on receiving the query packet (interest) will try to get content, thus reducing the overhead of network configuration. Said scheme is composed of the following steps:

• Handover detection. By using the information of physical link or router advertisement, mobile node detects the change network status.
• Handover indication. While the handover detection event is in progress, hold request message will be sent to proxy node by the mobile node before detection of actual handover. On receiving the hold message, proxy node will stop the process of sending content data to MS's old location and stores content data only to its local repository for future retransmissions.
• Handover completion. In case of acquiring new IP address, mobile node notifies the new IP to proxy node by using handover message piggybacking. CCN proxy node transmits the stored data to new location and does not transmit the repeated interest packets. On receiving content packet at old location, the mobile node can ask for next content data segment using normal CCN interest packet.

4.3.9 Tunnel-based Redirection (TBR)

This scheme performs the redirection of incoming interest packets between content router of MSC's home domain and the content router in the moved domain. Tunneling is used for this redirection. To guarantee the presence of a router, it periodically broadcasts the prefix names. Point of attachment in CCN is located by the comparison of this broadcasted information. This information is also used to know whenever a network initiates a prefix update (PU) round to the home domain. Home content router announces the prefix of the name of the MSC initially to inform the other cognitive radio (CR). The PU message about its current movement is broadcasted by MSC and its active state is also announced to the local routers of the domain. Then, the CR of the new domain sends the PU message to the home domain. The redirection path can be extracted by the exchange of the prefix update and its acknowledgment messages (P-ACK). The data packets and interests are then exchanged among the MSC and home CR through this name prefix. This scheme works in the following three steps Figure 4.2:

• Step 1. Indicates movement as the first step that identifies whether MSC would change the status of networks by the network address information or through the underlying physical link information. Normally, data is provided by wireless access point (AP) to MSC containing all CRs in the access domain. When the change of domain occurs in network location, depending on new domain router's prefix, then initially MSC forms a tentative name-prefix in order to tunnel the interest packets. The tentative name prefix composed of MSC's original name prefixes and moved domain name. In order to check validity of tentative name prefix, MSC performs duplicate name prefix detection (DND). Interest packets are broadcasted by MSC to the all in-range peripherals that are on a physical link residing in CR's domain that have been moved. This broadcast checks whether its tentative name prefixes is used by someone else. If it is used by anyone else, then the MSC repeats the DND process for creating new tentative name. Thus, repeated DND is just to guarantee unique tentative name-prefix to adjacent neighbors. After a successful DND process MSC forwards the PU message indicating its new location to its home-based domain CR. Tentative name prefixes and MSC's signature in the PU is used for checking its validity.
• Step 2. This step deals with the configuration of the redirect path. The CR that has received a PU message compares name prefix domain of the PU message with that of its own. CR sends the PU message to the content router based on FIB reference. This information is based on difference of its own domain prefix from that of the PU message. Otherwise, the PU message is not forwarded, instead a PACK message is sent to the MSC. When a PU message is received, each content router pulls out the name prefix information immediately from the router to find the message's home domain router. Then, PIT entry is formed with the name-prefix of the MSC's home domain CR. On receiving PU, a home domain CR checks whether the PU message is valid or not. If the signature validity is proved, the CR sends an acknowledgment message to indicate a successful PU. At that point, the home switch's entrance for that MSC is changed to the new one which the PU message has set. While getting the PACK message, middle of the route CRs look into the direction table once more. The CRs consume the significant PIT section and afterwards forward the PACK message.
• Step 3. This is the last step of the scheme and deals with interest redirection. The consumer requests are directed toward the MSC. Home CR looks up in the routing table to send the interest packet. The newly created interest packet is encapsulated by the home router and forwarded to the tentative name prefix that was extracted from the routing table. Intermediate CRs between the home content router and the MSC deliver this message through regular CCN forward techniques. The MSC receives and decrypts the encapsulated interest packet and transmit the data response.

4.4.1 Risks Due to Caching

The feature of caching in generic content-centric networks makes them different from traditional TCP/IP networks as it stores all the data in the router caches. Since data leaves its traces even after it is removed or forwarded, and can be retrieved by an attacker by probing and other such techniques. Hence caching puts the system security at risk in spite of being a defining feature of the CCNs. Attacks due to caching of data can be reduced by keeping data strictly up to date and restricting the replay of any old data.

4.4.2 DOS Attack Risk

DOS attacks, although difficult to carry out in these networks, are still possible. They are carried out through interests. Since data packets of request and response always follow the same path, they can be exploited and a destination can be flooded with irrelevant data packets. The complications arise due to the fact that the origin of the packet with respect to its location cannot be identified. Due to the architecture of the Internet and the generic CCNs, DOS attackers can easily cover their tracks. In order to cope with these attacks, a few security patches (to be added to the existing system) and trust mechanisms like firewalls and spam filtering have been developed. New security protocols that complement the existing networking protocols are also introduced to solve this problem [4].

4.4.3 Security Model

Establishing security of data in such a system, which is largely content-oriented, is mandatory before it can be implemented on a large scale. This has been a relatively unexplored area as compared to the caching or mobility of generic CCNs and needs further investigation to determine a better security model.

4.5.1 Cache Allocation Approaches

The authors in [15] presented an algorithm to compute optimal cache allocation among routers. The algorithm focuses on maximization of aggregate benefit, where benefit means reduction in hops taken by an interest packet from client to server. The interest packet is assumed to follow shortest path from client to server. Therefore, a shortest path tree rooted at the server that has the desired data for clients is formed. The shortest path tree structure leads the problem of cache allocation to be divided in two subproblems: (1) maximizing benefit of a data packet that has highest probability of being requested in a shortest path tree and (2) maximizing overall benefit of all data packets in whole network that will be sum of all maximizations in previous step. The former problem is being solved with help of k-means clustering algorithm while a greedy approach is adopted for the latter. The proposed greedy algorithm produces a binary matrix having nodes in rows and content packets in columns, depicting content allocations on nodes. The complexity of proposed optimization algorithm is max (O(sn³),O(c_total logN)) where s is the number of servers, n is total number of nodes, N is number of contents, and c_total is the total cache space of the network. Experimental evaluation showed that cache allocation should be heterogeneous and related to topology and content popularity.

In [16], the authors modeled network topology as an arbitrary graph and computed several centrality metrics of the graph (betweenness, closeness, stress, graph, eccentricity, and degree centralities). Total amount of network's cache is distributed heterogeneously based on centrality values among nodes of the graph. One homogeneous and 12 heterogeneous cache allocations were simulated using ccnSim, an open-source simulator. Simulation results depicted that cache allocation based on degree centrality is better among all centrality metrics. Moreover, heterogeneous cache allocation had very little (2.5%) performance gain over homogeneous cache allocation.

The cache allocation problem is described as a part of content-aware network planning problem in a budget constrained scenario in [17]. The content-aware network planning derives an optimal strategy to migrate from an IP network to content-aware network. The proposed algorithm for that purpose applies linear programming approach. The algorithm finds optimal placement of content on network routers based on minimization of flow and price per unit traffic. Probability of caching each content on each router is calculated. If cost of storage addition on a router for caching of a content with certain probability is less than budget, then cache is allocated to that router. Content popularity distribution has a major impact on the proposed approach for cache allocation in [17]. For highly skewed content popularity, there is about 87% less storage needed than the one deployed for uniformly distributed popularity content.

In [18], a centralized global cooperative caching placement strategy named generalized dominating set-based caching (GDSC) has been proposed for content-centric ad hoc networks. GDSC minimizes the network power consumption by selection of proper cache nodes and also achieves a better trade-off between round trip nodes and cost of caching. The selection of nodes depends upon nodes degree and number of hops, and the node with high degree will cache the content. Changing the number of hops selection results in different nodes being selected for caching. The number of hops that results in reduced power consumption for the whole network is selected.

The problem of trade-off between in-network storage provisioning cost and network performance has been addressed in [19]. The authors initially proposed a holistic model for intra-domain networks to portray network performance of routing contents to clients and network cost sustained by globally coordinating in-network storage capability. Furthermore, an optimal strategy for provisioning of storage capability has been derived that results in optimization of overall network performance and cost. The authors also demonstrated that the optimal strategy can result in significant improvement on both the load reduction at origin servers and routing performance. Moreover, provided an optimal coordination level, a routing-aware content placement algorithm has been designed in [19] that runs on a centralized server. The algorithm computes and assigns contents to every router for caching, which leads to minimization of overall routing cost in terms of transmission delay or hop counts.

4.5.2 Data Allocation Approaches

Cache management and content request routing policies have been proposed in [20]. Cache management policy decides whether a piece of data that passes through a router must be cached or not. If so, then the policy determines which portion of data residing in the cache must be removed to make room for data to be cached. In-caching and removal process of data, link congestion along the path of data retrieval, and data popularity are considered. A utility function is used in the decision of caching a piece of content on a node. The utility function takes minimum bandwidth of content retrieval path and content popularity as parameters. If a piece of content has higher utility value than lowest utility value of any content at a node, the content is cached in place of lowest utility value content. The main idea of cache management policy is that content forwarded over congested links is retained by caches while evicting the content forwarded over uncongested links. The content request routing policy utilizes a scoped-flooding protocol. The protocol locates requested cached content with minimum download delay as there are multiple caches having a piece of content. The scope is boundary for the protocol to stop searching for cached content, and is calculated in terms of number of hops that a packet traverses. An issue with such caching policy is that there might be no interested node for a content after some time that is residing with a high utility value in a cache. Such a content will never be removed from a cache due to higher utility value.

Caching approach presented in [21] focuses on fair sharing of the available cache capacity on a path among various content flows. Several clients connected to a server with different hop counts may share a network path. A probability factor called ProbCache is proposed in to resolve which content must be cached among all the contents of several users passing through a router. ProbCache is a product of two factors called TimesIn and CacheWeight. TimesIn determines the number of times a path can afford to cache a packet. TimesIn takes into consideration the remaining hop count to client and cache size of the path. CacheWeight resolves contention for caching among contents of several users on a router. CacheWeight decides where to cache the number of copies indicated by TimesIn. A content with highest CacheWeight on a router gets cached on that router. Only that client's content will have higher CacheWeight for which the client is nearest to the cache. CacheWeight is calculated by the ratio of number of hops from cache to server to the number of hops from client to server. The ultimate objective of ProbCache is to utilize caches efficiently to reduce caching redundancy and, in turn, network traffic redundancy. The caching approach achieved 20% fewer server hits and 8% hop count reduction in simulation results. However, only binary tree topology was being considered in evaluation.

In [22], the authors proposed a latency-aware cache management mechanism for CCN. The mechanism is based upon the principal that whenever a content is retrieved, it is stored into cache with a probability proportional to its recently observed retrieval latency. There is a trade-off between cache size and delivery time as reducing delivery time of every content requires a cache of size appropriate to store every content at every node [22]. The proposed mechanism minimizes this trade-off by assigning caching priority to contents that are more distant. It is a fully distributed mechanism since the network caches do not exchange caching information. Two main benefits of the proposed mechanisms are: (1) faster delivery time as compared to latency insensitive approaches, and (2) fast convergence to optimal caching situation. However, dynamic network conditions have not been considered while calculating optimal data caching.

A betweenness centrality-based caching strategy has been presented in [23]. The authors argued that a ubiquitous caching strategy results in data redundancy. Furthermore, a random caching strategy is useful where a high number of overlapping paths exists in network. The proposed caching strategy assumes that each node's betweenness centrality has already been calculated. A content request message stores every node's centrality value on its path to server. The content message generated by a server against a request message gets cached on a node with highest centrality value on the path to client. There is no collaboration between network caches, therefore it is a distributive caching strategy. The proposed caching strategy also considers dynamic network environments where topology tends to change with time. An ego network betweenness measure has been calculated instead of betweenness centrality in dynamic networks. An ego network consists of only immediate neighbors of a node.

In [24], the authors propose an age-based cache replacement policy that specifies which content objects to be cached in routers. An age is associated with each content object. The age decides lifetime of a content copy in a router. Age of a duplicate content object is calculated when it is added to a cache. The age value depends on two factors: (1) distance in terms of node count of the duplicate content object from server, and (2) popularity of the duplicate content object. The age is directly proportional to distance of duplicate content object from server and popularity of duplicate content object. This scheme allows for the content chunks to be pushed to network edges and release redundant content object storage at intermediate nodes. Moreover, this scheme assigns popular contents greater priority for getting cached by routers. The duplicate content object is removed when its age expires. Cooperation exists between routers to modify the age, hence it is a collaborative approach. The approach makes assumptions that content's popularity and network topology are already known.

It has been revealed in [25] that by jointly considering forwarding and meta-caching decisions, the performance gains are achieved in terms of content average distance traveled. The potential gains offered by smart forwarding policies such as ideal nearest replica routing (iNRR) are enabled by meta-caching policies such as leave a copy down (LCD). Cache pollution dynamics completely counterpoise these potential gains otherwise. The authors in [26] argued that iNRR with LCD acquires significant performance improvement over iNRR with leave a copy everywhere (LCE), provided best-case CCN topology. LCD has been designed for hierarchical topologies, so in general case (alternative meta-caching policy), least recently used (LRU) may be desirable.

In [26], the authors argued against the necessity of an indiscriminate in-path caching strategy in ICN and investigated the possibility to achieve higher performance gain by caching less. It was proved that a simple random caching strategy can overtake the current pervasive caching paradigm under certain network topology conditions. The authors proposed a caching policy on the basis of betweenness centrality that caches the content at the nodes having the highest probability of getting a cache hit along the content delivery path. The main goal of designing this policy is to ensure that content always spreads toward users and hence it reduces the content access latency. An approximation of the caching policy was also proposed by the authors for scalable and distributed realization in dynamic network environments where full topology is not known a priori. This approximation is based on betweenness centrality of ego networks. The research in [26] is an extension to work in [23], discussed previously. The authors realized that effectiveness of betweenness is dependent on the betweenness distribution. It was concluded that topologies that exhibit power law distribution such as WWW networks ensure effectiveness of betweenness policy.

The authors have proposed a collaborative caching policy unlike any caching policy mentioned before in [27]. The nodes collaborate by serving each other's request instead of independently determining if a specific item should be cached or which item must be eliminated to make room for new item. This way the policy allows nodes to eliminate redundancy of their cached items. The policy is based on a simple rule that an originally cached item should either be kept in place or be found at one of the neighbor nodes. This way, same performance of distributed caching policy can be preserved while increasing the available free caching slots.

In [28], a low complexity content placement scheme combined with dynamic request routing in CCN infrastructure has been proposed. The objective of the scheme is to reduce caching redundancy and make more efficient use of available cache resources. That way the scheme increases overall cache utilization and potentially increase user-perceived quality. The authors in [28] designed a self-assembly caching (SAC) scheme of collaborative content placement along a path of caches that is based on an analysis of the feasible cache size distribution. The hot content is being pulled by the SAC scheme toward the edge of the network while the cold content is pushed back to the core of the network based on popularity. Therefore, a low caching redundancy and low download delay is achieved because SAC could automatically distribute the content to the proper cache location according to its popularity. A trail having content cache locations is created at some special routers along the content delivery path. The trail is updated during the content chunk caching or eviction, leading to enabling of dynamic and absolute direction of following request search. The trail information makes the cached chunk detectable locally and so improves the in-network caching utilization.

In [29], a cross-layer cooperative caching strategy for CCN is proposed. Three parameters that are user preference, betweenness centrality, and cache replacement rate are introduced as application layer, network layer, and physical layer metrics. A caching probability function is derived based on Grey relational analysis (GRA) among all nodes along the content delivery path. Multiple requests for the same content are aggregated in PIT, which keeps track of interest packets forwarded upstream so that returned data packets can follow the path back to end-point of content requester. Some nodes aggregate requests and termed as aggregated nodes. The aggregate node computes the caching probabilities and adds them in the PIT. Caching probability fields of data packet is updated by the caching probability in PIT when the data packet arrives at the aggregated node.

A novel caching method to deliver content over the CCN has been proposed in [30]. The authors analyze caching content distribution and related interest distribution by considering the mechanism of content store and pending interest table of CCNs. Based on the analysis, a caching algorithm has been proposed to efficiently use the capacity of content store and pending interest table. The caching is done based on a cost model derived in the paper. The cost model is based on total delay to retrieve a piece of content. The content store will calculate the cost to remove a cached content for caching a newly requested piece of content. If the cost is below a threshold, the new content is cached. The proposed scheme does not account for the dynamic network conditions and is a distributed caching mechanism.

In [31], the authors studied the user-behavior-driven CCN caching and jointly investigated video popularity and video drop ratio. It has been said that this is the first paper taking the video drop ratio into consideration in CCN caching policy design. The video drop ratio means that video consumer does not always watch the video till the end. An intra-domain caching in CCN system has been considered that is composed of servers/repositories, routers, and users. Content store (CS) of each router acts as a buffer memory. The routers are divided in levels hierarchy and videos are ranked according to popularity. The main idea of proposed scheme is that ith chunk is cached in jth router if, and only if, the higher ranking chunks than ith chunk can be stored in lower level routers than jth router, and if the lower level routers cannot cache all the chunks having higher ranking than ith chunk. The proposed algorithm achieves optimal values in terms of average transmission hops needed and server hit rate, given a cascade network topology.

A caching policy similar to the one mentioned in [31] has been proposed in [32], where the similarity is used for classification of routers into levels. The proposed strategy maximizes cache utilization and improves content diversity in networks by content popularity and node level matching-based caching probability. Three parameters including hop count to the requester, betweenness centrality, and cache space replacement rate are considered for evaluation of nodes' caching property along the delivery path. The proposed strategy performs classification of nodes into levels on the basis of their caching property. High caching property nodes are defined as first-level nodes; while less capable nodes are termed as second-level nodes. Moreover, in this approach, the popularity of content is matched to the affiliated node using probabilistic methods. Content is cached in nodes based on their popularity in an increasing order, starting by placing the most popular content in the first level, the lesser popular in the second level, and so on. This decreases the replication of less popular content and access delay. The proposed policy improved up to 23% cache hit ratio, reduced up to 13% content access delay, and accommodated up to 14% more contents, compared with leave copy everywhere (LCE) policy.

The research work conducted in [33] emphasizes the need of an algorithm for selection of a subset of nodes among those available to serve new incoming requests for longer time. The authors proposed a cost-effective caching (CEC) algorithm that selects eligible node based on node's remaining capacity to cache more content. In CEC, a node with highest space left in content store will be designated to cache the content. The term cost in CEC is the cost of computation carried on each node in network for the process of sending interest packet and receiving data packet. CEC tries to minimize this cost. The motivation behind CEC is that if a new content replaces already cached content then the cost increases. The cost is minimized by caching a content on some other node that has sufficient free space to store that content without replacement. That way CEC maximizes cache hit and minimizes both cache miss and content replacement.

Due to dynamic change in bandwidth in VANETs, dynamic adaptive streaming (DAS) technology is used for delivery of video content with different bit rates according to available bandwidth. DAS requires several versions of a same video content, resulting in reduced cache utilization. Hence a cache management scheme for adaptive scalable video streaming in Vehicular CCNs has been proposed. The scheme aims to provide high quality of experience (QoE) video-streaming services through caching of appropriate bit-rate content near consumers. A video content is divided into layers where each layer is the video content with specific bit rate. Chunk of layers are pushed to neighbors according to their available bandwidth through broadcast hop-by-hop.

Table 4.1 compares the CCN caching schemes mentioned so far on various critical parameters. Table 4.2 depicts the other objectives (if any) of aforementioned caching schemes coupled with caching.