TY - JOUR
AU - Khan,, Rehanullah
AB - Abstract The amount of online video content is exponentially increasing, which spurs its access demands. Providing optimal quality of service (QoS) for this ever-increasing video data is a challenging task due to the number of QoS constraints. The system resources, the distributed system platform and the transport protocol thus all need to collaborate to guarantee an acceptable level of QoS for the optimal video streaming process. In this paper, we present a comprehensive survey on QoS management for the video-on-demand systems. First, we focus on load management and replication algorithms in content delivery networks and peer-to-peer (P2P) networks for their shortcomings. We also address the problem of admission control and resource allocation with the objectives of congestion avoidance and frame-loss reduction. Besides, we introduce and discuss various replication schemes. For both the client–server architecture and P2P networks, we highlight the need for a specific storage management policy to preserve system reliability and content availability. We also focus on content distribution and streaming protocols scaling. We deduce that content availability is linked to the characteristics and the performance of the streaming protocols. Finally, we create a comparison table that presents the different contributions of the discussed approaches as well as their limitations. We believe that such a comprehensive survey provides useful insights and contributes to the related domains. 1. INTRODUCTION Nowadays, there is a lot of concern regarding the quality of service (QoS) of video-on-demand (VoD) systems. As the amount of video traffic accessed via the Internet is steadily growing, providing an optimal QoS is still a challenge for VoD systems. Therefore, the system resources, the distributed system platform and the transport protocol must collaborate to guarantee an acceptable level of QoS for the video streaming process. Naturally, the QoS is straightly tied to the hierarchical structure of the network. It is assumed that content delivery networks (CDNs) benefit from the ease of management and are able to provide a high QoS [1]. Thus, the use of servers and trackers to provide content storage and distribution ensures the system’s reliability and enhances data availability. On the other hand, the main issues with CDNs are the high cost and the low scalability comparing to peer-to-peer (P2P) networks [2]. More precisely, a large-scale P2P system whereby peers contribute resources to store and share content can reduce the load that would otherwise be managed in data centers. The main challenge facing VoD systems is to serve a large number of end users among all network’s issues and resource limitations. The QoS and the quality of experience (QoE) are tightly linked measurements to show how far the provided QoS parameters conform to the client’s specifications [3]. Thus, the end user should be able to receive the streams continuously without interruption, neither a quality degradation problems caused by the content delivery process. Hence, the system has especially to be aware of scalability, content availability and load balancing. Therefore, regardless of the network hierarchy, VoD systems must provide replication and caching schemes that can optimize content placement and track video popularity growth properly [70]. Data replication and content storage techniques are keys strategies in VoD systems to increase data availability and reduce response time. This can be performed, for example, by making more copies in the most popular data center or by forwarding the end user query to the nearest available copy [4]. Besides, the replication scheme is supposed to reduce the query search latency, to enhance reliability and to achieve load balancing. For data caching, it is essential to determine which content should be discarded from the memory disc to be replaced by the incoming replicas. Some studies proposed to use simple content placement schemes (e.g. second chance frequency least recently used [5], first-in-first-out [6], least frequently requested [7], etc.) to manage cache replacement. Some others have proposed to define the content replacement as an optimization problem to be solved using an approximation method [8, 9, 10]. Moreover, replication concerns more P2P networks since data storage and sharing are performed using the individual resources of the peer. Therefore, the system has to be aware of which piece needed to be replicated in such a hierarchy in order to efficiently manage the limited cache. We notice the possibility of introducing an extra upload capacity provided by content servers, which are normally used in P2P networks as index servers (i.e. they maintain an updated list of peers possessing the content) [11]. The primary objective of this work is to highlight the axis of QoS enhancement for distribution management in VoD systems [75, 76, 77]. The rest of the paper is organized as follows: in Section 2, we introduce the QoS of VoD systems, and we address the need to define what parameters should be considered to identify the system’s level QoS. Section 3 explores the issues of QoS supervision and admission control. Section 4 is devoted to reviewing existing approaches to QoS enhancement. Also, the replica placement problem is highlighted in both CDNs and P2P networks. Furthermore, a discussion of the issue of static replication and the performance of some popularity-based replication approaches is presented in this section. The impact of the hierarchical structure of the network on distribution management and content availability is addressed. Moreover, an overview of the importance of the streaming protocol as an element of QoS enhancement for VoD services is presented as well. Section 5 presents a practical comparison between different schemes that have been reviewed, highlighting their pros and cons. Finally, the paper is concluded in Section 6. 2. QOS IN VOD SYSTEMS VoD systems are designed to perform scheduled tasks in real time to respond to end-user requests. Each system is concerned depending on the degree to which it accomplishes its mission to evaluate the performance. The first step in requesting granted performance from the VoD system is to specify QoS requirements. Then, the underlying system is used to allocate the appropriate resources to guarantee the required QoS. The process of QoS guarantee in VoD systems should meet the following goals: ▪ The video system should be scalable to serve multiple unsynchronized requests of video while avoiding congestion and minimizing packet loss. ▪ End user should receive video data at a constant rate in order to preserve the same video quality during the streaming. Although, the system should adapt its process to the user’s bit rate to preserve the data consistency. ▪ End user has to be able to carry out different features such as pause, play, fast forward and rewind within the very first seconds of the supplied video data received. ▪ VoD system must perform a fast recovery without prominent degradation of the streaming process when a data source fails or becomes out of range for P2P architectures. ▪ The process of resource allocation should be appropriate to the specified QoS requirements, which are the supplied video quality, end user’s bit rate, etc. ▪ The system has to oversee the video segments during their transition toward their destination in order to be aware of efficiency’s change of the corresponding system. ▪ An appropriate scheme must perform a quick regulation of the system’s performance to ensure that video data may flow within the required QoS specification. The QoS requirements are formally modeled by a set of parameters that can be assigned numerical values according to both internal and external measurements. The QoS parameters defined for VoD systems can be categorized according to three orientations as the transmission’s performance, delivered content quality and cost. For the evaluation of transmission’s performance in VoD systems, the measurements of packet loss rate, latency, variable delay and jitter are crucial since they can affect video consistency and data synchronization. For some real-time applications such as video streaming, receiving data with an acceptable change of content integrity is tolerated compared to the data with delay. The delivered content quality is also significant for the end user. Thus, video resolution, frame rate, compression scheme, subjective image and sound quality needs to be defined due to the importance of enhancing the user’s experience. Likewise, a VoD system should consider cost issues while implementing a new architecture as well as the integration of some modifications within the existing distribution. Consequently, data transmission and storage charges have to be taken care of. However, this definition of QoS parameters is very general; thereby some studies have proposed alternative measurement factors to improve the evaluation of QoS level in their systems. 3. QOS MANAGEMENT AND PERFORMANCE EVALUATION Traditionally, video traffic has requirements in terms of QoS. Notably, it necessitates a lot of network bandwidth to distribute media, a massive disc space to save video content, real-time management of the process of data caching, an effective admission control policy, efficient congestion avoidance method, etc. Therefore, the process of QoS management and performance evaluation is critical to define the current state of the VoD system [74]. Indeed, it will be able to moderate its method of data delivery to overcome the different challenges of video data transmission and to enhance the end-user experience. 3.1. QoS supervision QoS management is mainly applied in supervising and controlling QoS properties. Our studies of QoS provision mechanism employ video traffic descriptor (i.e. traffic specification) for performing admission control. The supervision process is based on real-time evaluation of QoS parameters’ measurements. Indeed, the system should make sure not only that video data flows within the specified QoS specification but also that the supplied QoS properties are sustained during the transmission. In this context, a proposed technique to study the behavior of real-time video streaming over data distribution system (DDS) middleware is presented in [12, 13]. It refers to the quality of video traffic by evaluating different QoS parameters. The approach defines three main QoS parameters namely `lifespan QoS’, `history QoS’ and `reliability QoS’. The `lifespan’ parameter refers to the maximum time the system is allowed to proceed with one video segment’s treatment. Thus, when a segment reaches its expiration date before being treated by the system’s resources, it should be automatically discarded since video streaming systems do not tolerate delays, which boost the congestion avoidance between different segments as well. The `history QoS’ parameter defines a threshold that indicates the maximum number of segments that can be cached within the system. Once this number is reached, the system tends to replace the oldest segments in the middleware by the newest ones to maintain the video streaming consistency. The `reliability QoS’ value is set to `reliable’ or `best-effort’ by the system depending on the QoS requirement of the content delivered. The integration of those QoS parameters within a DDS is beneficial in terms of management of video streaming consistency as well as data presentation (i.e. the different video segments should be delivered in the same order). Moreover, a QoS evaluation can be established through the analysis of the proposed features and according to the number of transmitted/discarded segments, data delay, etc. 3.2. Global scheduling and admission control In VoD systems, the data that arrives late is considered useless since it can affect the consistency of video streaming. Thus, streaming is a process that allows you to play audio or video content as you download it. On the other hand, the only difference (see Fig. 1) is that at the level of video streaming the Internet user has no right to know how to stop or to pause the video or audio content, for example television and radio, while VOD has all the right, for example YouTube. The system must be able to decide if a request can be supported on a resource without being harmful to other previously admitted requests. A step before performing resource allocation is defined as QoS negotiation [14]. This operation interrogates each resource about the level of QoS. It can provide before initiating a new task. Therefore, the system concludes about the QoS level that should be assigned to every single component within the system. In case the estimated QoS level is below the required QoS specification, the system report to the initiator task that the system resources cannot handle the request for the moment. Admission control performs a comparison between the QoS specification related to the new task and the present resources to provide QoS negotiation policy with the estimated QoS level that can be afforded. Moreover, the admission control process is closely tied to the QoS negotiation operation since the decision whether a request is initiated or not depends on the QoS level of the whole system that has been stated after the QoS discussion. FIGURE 1. Open in new tabDownload slide Live streaming vs VOD. FIGURE 1. Open in new tabDownload slide Live streaming vs VOD. On the other hand, a VoD system must adhere to the adequate policy of periodic QoS parameters’ measurements to determine whether the system’s resources are available or not. The admission control policy allows the system to schedule user requests examination according to system resource availability as well as their properties of QoS guarantee. This is done by performing a measurement algorithm. It determines the state of system resources periodically and verifies if all users are receiving an acceptable QoS. An effective feedback control scheduling for distributed multimedia systems (FCS-MS) is presented in [15, 16]. The technique proposed a policy to handle real-time QoS parameters’ variation according to the network load estimation. The feedback aspect is achieved within a centralized architecture by the fact that each client, once served, must return to the master server a feedback report containing QoS parameters’ measurements performed nearby the client itself. The master server carries out a comparison between the supplied QoS specifications and the received measurements to verify the system’s performance. Thus, the master server requests the adjustment of QoS parameters within all video servers when the system performances are degraded. Generally, the approach proposes to discard some video frames to enhance content delivery’s QoS specifications. Moreover, the proposed approach defines a distinctive internal structure for the master server to control the scheduling process according to the predefined QoS specifications. The proposed admission control approach is mainly developed for feedback control architecture for distributed multimedia systems. Nevertheless, it can be useful for all VoD systems. However, the approach does not consider the integration of the suggested admission control policy for rejections of client requests. Other than admission control and scheduling, data replication techniques are keys strategies in VoD systems to increase data availability and reduce response time. Besides, the replication scheme is supposed to reduce the query search latency, to enhance reliability and to achieve load balancing. In the next section, we discuss the importance of data replication for VoD systems. 4. QOS ENHANCEMENT IN VOD SYSTEMS 4.1. Adjustment of the system’s replication strategy For VoD systems, the process of data replication is considered a critical operation to reduce access latency and variable delay, minimize the number of rejected requests and maximize the degree of content availability. Thus, video data replication’s management represents a relevant field that yields the attention of many researchers. 4.1.1. Replication model In VoD applications and due to the variable size of the videos, it is challenging for replication strategies to provide a high QoS to the end user while consuming a minimum storage capacity. By achieving those goals, a video streaming system acquires a valid replication profile. A general system model (Fig. 2) that captures the essential characteristics of the replication-based system is presented in [17]. For simplicity, the set of N video servers and the set of M video objects are denoted, respectively, by S and V where $$\begin{equation} \boldsymbol{S}=\sum_{\boldsymbol{i}=\mathbf{0}}^{\boldsymbol{N}}\boldsymbol{S}\mathrm{i}\ \mathrm{and}\ \mathbf{V}=\sum_{\boldsymbol{i}=\mathbf{0}}^{\boldsymbol{M}}\boldsymbol{Vi}\ \end{equation}$$(1) FIGURE 2. Open in new tabDownload slide The basic model for the replication system. FIGURE 2. Open in new tabDownload slide The basic model for the replication system. such as N > 0 and M > 0. Let TSC be the total storage capacity and TBC the total available bandwidth that can be provided by the video streaming system. The total storage capacity required for requested videos cannot exceed the full storage capacity of the system. On the other hand, to preserve a valid replication profile, the total intentional bandwidth to transmit demanded videos should remain below TBC i.e. $$\begin{equation} \boldsymbol{TSC}\ge \sum_{i=0}^M\boldsymbol{SC}(Vi)\ and\ \boldsymbol{TBC}\ge \sum_{i=0}^M\boldsymbol{BC}(Vi) \end{equation}$$(2) where |$Vi$| represent the video object i. In the literature, many authors propose various approaches to enhance the replication scheme’s scalability according to video popularity, which refers to the measurement of client access frequency of a video. Those studies were based on the fact that the more the video popularity increases, the more the streaming system evaluates a possible growth of the number of viewers. Hence, a popularity Pi is associated with each video (i.e. |$|\boldsymbol{Vi}|\propto \boldsymbol{Pi}(\boldsymbol{Vi})$|). Thus, a video Vi is replicated into a set of replicas R and we assume that the number of copies RVi is proportional to the popularity of the video Vi i.e. $$\begin{equation} \left|\boldsymbol{RVi}\right|\propto \boldsymbol{Pi}\left(\boldsymbol{Vi}\right). \end{equation}$$(3) This straight proportionality between the popularity measurement and the replication profile i.e. the optimal number of copies to make affects the storage capacity of the system. We indicate |$\mathbf{ST}$| as the storage required capacity for a video Vi where $$\begin{equation} \left|\boldsymbol{RVi}\right|\to \mathbf{ST}\ \left(\mathbf{Vi}\right). \end{equation}$$(4) 4.1.2. Storage management Providing VoD streaming services to a large number of users on the Internet requires making as many replicas as it is possible to enhance video data availability. However, the intended performance of the system is supposed to be degraded with the increase of replication degree. Indeed, the number of stored replicas and the amount of memory consumed to cache replicated segments are directly proportional. Generally, a replication strategy has three roles [18], which are segment storage, segment selection and segment update. In this section, we address the issue of storage capacity in VoD systems. 4.2. Structured P2P network Naturally, storage management depends on the architecture and the content placement as well as its distribution. The recent literature [19] [20] [21] focuses on the P2P design approach with replicas storage as the primal issue to deal with. In P2P networks, systems can be classified into either structured or unstructured P2P systems. Structured networks are established when the topology is monitored and all the information concerning the different connected peers is always available within the monitor (i.e. segment location, storage capacity supported by each peer, etc.). Hence, the replication system can store the new replicas efficiently in the peers having the highest storage capacity. Thus, those systems have to maintain an efficient mapping permanently between the replicated data and the available placement to storing for providing content availability and fast response to data queries. The authors in [72] presented an efficient chunk replication algorithm for QoS enhancement in heterogeneous P2P VoD systems (gathering mobile and fixed peers). The approach called enhanced chunk regulation algorithm (ECRA) combines three main algorithms namely local demand-based chunk download (LDCD), peer ranking factor (PRF) and local demand-based chunk replication (LDCR). The LDCD algorithm handles the chunk selection and pre-fetching in the sliding window to serve clients according to the availability of chunks. LDCD is basically invoked in the case of a request for standard chunk download. The selection of the appropriate peers to download video chunks is performed by the PRF algorithm based on their PRF value. Hence, only peers characterized by good signal strength, available energy and bandwidth are considered within the list of peers to download video chunks. On the other hand, LDCR algorithm chooses which chunk has to be replicated next. This process follows the policy of `most demanded first’ to pick up the suitable chunk to be replicated. So, the QoS of the VoD system can be improved. Besides, the LDCR algorithm performs peers selection for chunk replication i.e. only peers having the best available download bandwidth and the highest cache space have to be selected. The authors compared their proposed algorithms (chunk download algorithm and replication algorithm) with some eager algorithms, for example, popularity-based content replication (PBCR) [22], closest playback-point first [23], deadline-aware scheduling [24] and rarest first [25]. The comparison considered the number of source server downloads, the number of missed chunks, the startup delay and the stalling delay. The evaluation of the proposed downloading algorithm shows better results in terms of the average startup delay since it serves the temporal demands according to the availability of chunks in peers. A comparative analysis shows that the PRF and LDCR algorithms outperformed the ECN and PBCR approaches. They reduce the number of missed chunks and adjust chunks’ availability according to the dynamic changes in the number of requests. However, the issue of source server downloads and missed chunks increments with the growth of the mobile peers joining the network was not considered. 4.3. Unstructured P2P network Unlike the structured P2P systems, the topology cannot be predicted and the connections between different peers seem to be arbitrarily constructed in the unstructured P2P network systems. Thus, both segment’s storage and downloading are challenging since the replication process places new segments randomly within peers without being cautious about overload or congestion problems. As a solution, there are several techniques based on peer clustering to form temporary centralized units gathering connected peers. To reduce network overload and congestion between different nodes, the authors of [26] studied the effect of the introduction of an interconnected structure called distributed spanning tree (DST) within a P2P network. The proposed approach consists of an algorithm aiming to logically convert the peer network into a set of DSTs where nodes were differentiated into head nodes (HNs) and leaf nodes (LNs). Initially, the algorithm chooses a set of HNs among the regarded network, referring to some criteria (user approval, traffic load, etc.). Then, the clustering process is initiated. This process thus associates an LN list gathering the details of the nodes already elected to form the tree’s stem for each root node (HN). The scheme’s rationality is based on a dynamic formation of DSTs in which each HN and its LNs tend to attract further LN in the network to gather all adjacent nodes in the corresponding DST. Moreover, the paper outlined an appropriate algorithm for replication management in a DST structure based on global replica management [27]. The proposed algorithm defines the process of the read and writes operations performed by different nodes in the DSTs’ network. The main purposes of the observed study are the enhancement of replica consistency and the reduction of the peer network overload by decreasing the number of exchanged messages between different members of the network. Hence, when an LN requests a segment, the `read’ process is as follows: it verifies the state of the HN holding the segment. If the corresponding HN is busy with another operation, the LN receives `wait’ reply from the HN. Moreover, the same LN forwards the `wait’ reply to all other LNs in its DST network. For the `write process’, it updates the replicas in different root nodes to ensure that the anticipated LN nodes will receive the latest updated model. The performance evaluation of using the read and write algorithm within a DST structure shows a remarkable decrease in the number of messages passes over a DST applied peer network compared to a traditional peer network. Furthermore, the proposed approach presents an efficient solution for congestion between nodes while accessing the same content because the read operation performs locally in each DST. The approach has also been tested in terms of scalability, and the results showed that the introduction of the DST structure improves the overall efficiency of the system (i.e. enhancing the number of successful read and write operations). The analysis over a theoretical perspective proves that the number of messages exchanged for both the read and the write operations is proportional to the number of nodes added dynamically to a DST through the clustering process. Consequently, the more the P2P network is dense, the lower is the efficiency of the approach in terms of the reduction of bandwidth consumption. 4.4. Storage optimization in client/server architecture The process of segment storage is related to P2P architecture since it employs end-user equipment to perform a self-sufficient streaming service. Moreover, some research work deals with the case of content storage and other replication features within a client–server architecture. A selective segment replication policy in VoD systems is proposed in [28]. The solution is a remedy to the congestion problems affecting especially the centralized architectures. The replication process utilizes predictive dynamic segment replication (P-DSR) method detailed in [29, 30] and is based on two steps: the first one focus on the segment selection scheme where only segments having a future reading load above a predefined threshold are chosen to be replicated. Secondly, the P-DSR policy decides a suitable placement for future replicas. Video server’s selection scheme chooses the least loaded video server, which has enough disk space to store the future and the current reading load for segment duration. Thus, the policy avoids the use of an overloaded video server for replicas storage. The simulation results showed that the proposed approach achieves the best performance in terms of the waiting-frames rate and the received-frame rate. In another study [31], the authors proposed a QoS management method taking into account the high-speed satellites using Big-LEO systems and Ka-Sat satellite [32]. A priority queue has been proposed as an OBP satellite. In order to obtain optimal performance for multiservice application traffic, an intelligent route selection algorithm has been enunciated. In this study, the applications are bidirectional VoIP, video streaming, hypertext transfer protocol (HTTP) Web and file transfer via FTP. Simulations results showed an efficiency of the proposed algorithm with some variation of bit error rate and several connection rates. 4.4.1. Popularity aware replication Video-popularity replication systems give priority to popular videos when performing the replication process over unpopular videos to reduce caching load. Video data popularity refers to the weighted value of views frequency (i.e. client access frequency to a specific video). Generally, the system dedicates an algorithm to perform video popularity measurement (e.g. the popularity value increments in every access detection, estimation of the number of requests supplying the same video, etc.). Some studies [33, 34, 35] proposed to classify stored video content in popular and non-popular classes. This classification sets a replication factor for each video in the CDNs according to the history of new replication degree attributed to the video. It is very common to exploit the history of a replication degree to estimate the video popularity profile and decide if it is considered popular or non-popular. To adapt the replication process of each video on the Internet to its estimated popularity, a popularity growth aware replication scheme for CDN systems is proposed in [36]. The proposed scheme introduces the establishment of central coordination servers set to handle the admission control of the data traffic and to perform periodic measurements of QoS parameters nearby the client. As such, the different measurements are communicated from the end user to central coordination servers. Hence, this approach defines two policies (Fig. 3), HERMES [37, 38] and AREN [39]. The first method is an adaptive replication policy based on a video popularity prevision scheme to provide highly available Internet video using hybrid CDN. The second method AREN is a popularity aware replication scheme used mainly for cloud storage. It is a replication strategy based on collaborative video data caching [40, 41], which utilize an optimized bandwidth allocation technique that adjust video replication degree according to bandwidth availability. Moreover, AREN’s performance is based on some critical measurements such as video size, network availability, actual number of viewers, number of replica for each video, inter-arrival time between two requests, total number of views, average time between requests, etc. According to the different measurements, the method proposes adding a label to each video in order to define its class, which thus refers to its popularity. Therefore, AREN and HERMES perform similarly to the popular and non-popular video classification on the Internet. FIGURE 3. Open in new tabDownload slide Learning and predicting model using HERMES and AREN. FIGURE 3. Open in new tabDownload slide Learning and predicting model using HERMES and AREN. Using Hermes, the popularity is determined according to the replication degree of the video. Thus, video popularity increases if the system detects a rise of replication degree for the examined video as a new request arrival is detected, so the video is classified as `popular’. In terms of performance, AREN and HERMES outperform the non-collaborative caching techniques [42, 43]. The results show that the proposed approach reduces storage memory consumption by 2-folds during the replication process. However, the deployment of such architecture is costly since the association of the two-replication schemes requires the introduction of new features in the network stack. Besides, the method prioritizes the treatment of popular video that can affect content availability since client preference is not apparent and cannot be predicted. Data availability and fast response to client’s queries are challenging for all video streaming systems to maintain an agreeable QoS level. However, the preservation of data availability is an issue due to the limited storage capacity. Hence, the number of client requests rises, the replication factor increases and this initiates the replica allocation and placement problems [78, 79]. Usually, to overcome replication’s issues, the static replication (i.e. in which the number of the replica is predefined and manually moderated) is not a valid option. The authors in [44] introduce a data popularity-based replication algorithm called `Pop Store’. This method is based on popularity estimation according to client access frequency to the considered video. Pop Store is designed mainly for cloud data storage systems to overcome the content availability problems [45], which is traditionally considered using a static replication scheme [46]. Moreover, the proposed technique initially assigns one or more constant value(s) of the popularity threshold in a way that each value corresponds to a replication degree. Thus, the approach defines a set of popularity thresholds to optimize replication under video popularity variation. It then assigns to the same video a different replication degree per time interval according to one of the predefined thresholds. The system is also equipped with a manager to determine the suitable number of replica per file as well as its placement (i.e. which data center to be used for caching the incoming replicas). Pop Store ensures that every stored video is considered in the replication factor setting process and not only the most popular video. In addition to access frequency measurements, this approach defines a second condition to decide about video popularity: the replication manager verifies if the video has been recently consulted or not to confirm whether it is popular or not. It considers that the latest consulted video will be probably the most supplied video in the incoming requests. The simulation results show that Pop Store outperforms the other replication schemes used in the cloud environment, such as the static replication schemes and the latest access largest weight algorithm [47]. To enhance replica storage efficiency and to avoid remote access to download requested data, Pop Store tends to place the copies of popular videos in data centers with the highest access frequency to download videos. 4.4.2. P2P vs CDN P2P is an overlay network. One of the main characteristics of P2P is that a peer (i.e. node or computer) can/has to perform both function of server and client. In simple terms, a peer can share his/her data with other peer; in this way, it serves data to other peer (client), thus act as server. At the same time, it can download data/files from other peer, so it becomes a client. In CDN, it is more of like server–client system. The servers that are placed geographically in different positions are forming an overlay network that connects all edge servers together for data synchronization and update purpose. However, this server cannot download files form the node/computers connected it. So server has only serving capability and nodes connected to it has only client capabilities. 4.5. Infrastructure scaling and distribution management In this section, we outline the impact of network infrastructure on video data distribution. The most fundamental role of using a specific architecture is to distribute data content efficiently. In the literature, there are two classical system models [48], which are the P2P model and client/server architectures. However, we assume that each hierarchy is an embodiment of the data management process. The main objective of a distribution management hierarchy is to carry data content with the minimum delay and packet loss. However, with the increase of clients count, the network gets overloaded regardless of the utilized architecture. Therefore, the infrastructure mapping is critical for content availability and data dissemination management since it defines the placement of different units of the network as well as the routing process of data content. Various techniques are proposed to enhance data delivery and mapping in VoD systems to serve more end users and maintaining an acceptable QoS level. Performance evaluation of VoD content delivery over an integrated metro/access network architecture is presented in [49]. The VoD blocking probability decreases if M-metro servers (MSs) are placed in specific metro nodes, combined to an n-mesh-configuration (where n refers to the number of mesh links) and introduced in strategical positions of the metro/access network (Fig. 4). FIGURE 4. Open in new tabDownload slide Example of 1-MS and 1-MESH configuration in a metro/access network. FIGURE 4. Open in new tabDownload slide Example of 1-MS and 1-MESH configuration in a metro/access network. In the proposed pattern, the node where the MS has been introduced has a hybrid role. It acts as an optical line termination dealing with the received requests of connections. The same node performs as an optical network unit (ONU) to allow the content delivery from the MS to the child nodes of the hosting node. The same integrated architecture was being tested using both active and passive optical technology. The authors proposed a comparative analysis of the two solutions (active and passive technology). In terms of performance, the active method costs less and allows serving more end users by the same MS as it can reach all end users wherever it has been placed in the network. However, passive solution showed a better energy saving since intermediate nodes (except those equipped with an MS and/or mesh link) are passive optical nodes; however, in the active mode, all intermediate nodes must be powered (regardless of the presence of an MS and/or a mesh link). In order to preserve data content availability, the authors of [17] proposed a multithreading based approach to enhance content distribution in a personalized client–server hierarchy and to serve a higher number of end users as well. The system model presented in this work (Fig. 5) defines a set of clients, a collection of intermediate servers (ISG servers), the third set of distributed servers (DSG servers), a local tracker separating clients and ISG servers and finally a global tracker placed between DSG and ISG servers. FIGURE 5. Open in new tabDownload slide Proposed network hierarchy [17]. FIGURE 5. Open in new tabDownload slide Proposed network hierarchy [17]. Trackers maintain the updates of served videos (size, format, popularity, etc.) in a specific register and preserve a list referring to available servers in every time interval. The proposed approach utilizes the concept of multithreading to manage client requests. This choice is based on the flexibility and the capability of this concept to execute several threads simply by executing one process called `main thread’. Thus, the proposed approach defines a thread per client request addressed to the ISG or DSG server to serve a higher number of end users. Video demands are received first by the local tracker, which consults its register to verify video location. If the supplied content is available on one or several ISG servers, the tracker picks out the less loaded servers to treat the request. The nominated server determines the number of threads in progress, managing the same video content and verify whether the measured value exceeds a predefined threshold. The request is placed on the server queue if the ISG server is overloaded (i.e. the number of threads in progress is below the threshold). The demand is rejected when the queue is full and is placed for a specific time interval in the queue of the local tracker. Then, it is dispatched to the global tracker if all ISG servers remain overloaded within the time interval. When the global tracker receives a video request, it verifies its availability on DSG servers, and it proceeds in the same way as the local tracker. The unique difference, in this case, is that the global tracker discards the waiting request automatically if no DSG server can treat it within the fixed time interval. The architecture detailed above outperforms in terms of data content availability since it uses a distributed hierarchy controlled via trackers to reduce the server’s load (i.e. the controlled structure reduces the signalization traffic). Besides, the simulation results report a remarkable decrease within the rejected requests (the number of rejects is below 3% of the total number of received requests). 4.6. Streaming protocol based QoS enhancement Selecting relevant streaming protocol is very important for video streaming systems since it determines which options are available in terms of the quality of videos, what kind of devices will be able to support video content, etc. In this section, we address the different schemes of video content delivery (i.e. the different ways video content can be delivered and played in the end user’s device.). Next, we present an enticing QoS enhancement approach based on the HTTP. 4.6.1. Video content downloading and streaming The simplest way of video content streaming is called `Download and Play’ [50], and it is one of the primary streaming schemes. This method proposes that the client can frugally download the entire required video and then play it locally on his device. The main drawback of this method is that the client has to wait until receiving the totality of the video. This can be very unpleasant for the end user in case the video is large. However, this approach gives the client the possibility to enjoy video playing options such as fast forwarding instantly, reverse playing, playing back some points of interest in the video, etc. The second way of video streaming content is the `Live Streaming video’ [51]. This method performs a constant stream of the video to allow the client to instantly watch the content as soon as the downloading is initiated. This type of streaming video is traditionally used to broadcast live events (e.g. conferences, matches, etc.). There are no allowed playing options in live streaming. The live streaming process is strikingly opposite to the download-and-play method since it does not involve the downloading of the entire content. On the other hand, a hybrid scheme exists to define the “Progressive video Downloading” [52] that permits the progressive transfer of video content from its placement in the network to the client’s device until the end of the supplied file. This method simulates adaptive streaming [83, 84], but the client cannot benefit from video playing options until receiving the entire part he is seeking to re-play. 4.6.2. Dynamic Adaptive Streaming over HTTP Recently, dynamic adaptive streaming over HTTP (DASH) presents a novel streaming video standard for several studies [18, 53, 54]. It is based on the streaming protocol HTTP that allows the end user to switch video quality during the streaming, referring to the provided Internet bit rate. The main reason for using HTTP for the transfer of video content is that it provides progressive streaming allowing the end user to play the received segments instantly without waiting for the downloading of the entire content. DASH define a particular client/server architecture: DASH server containing media segments. Media presentation description (MPD). DASH client. Caching server to reduce the DASH server load. HTTP adaptive streaming initiates a chunked download instead of file download, which reduces client awareness about network’s influence (i.e. bit rate variation, congestion, packet loss, etc.). A pre-treatment is performed on each video enclosing splitting the content into segments and the attribution of a URL referring to the segment’s address on the server for each video segment. Moreover, the DASH server provides an MPD document containing media metadata and the different representation (e.g. available encoding bit rates, video frame rates, video resolutions, etc.) of each segment used to facilitate segment’s selection among several contents. In effect, segment representation helps the client pick out the desired content to download according to network measurements and buffer fill levels. When an end user requests a video segment, i.e. a client embeds the segment’s URL via HTTP GET request, DASH client gets the demanded media and serves it for the client. DASH is usually used in client/server architecture since the treatment of DHSH content requires an ample disc space. However, the authors of [55] proposed the introduction of DASH over a hybrid network called the P2P Web. This method combines the use of client/server architecture and the P2P network in one structure. The server is used to store DASH content as well as to hold instantly a list indexing the information of each peer in the network. The server communicates an update of the index and the MPD to the tracker to set it permanently on a `stand by’ mode to be ready to perform when the central server is overloaded. The system model (Fig. 6) performs according to the server load when a new request is received (diagram in Fig. 7): If the server is not overloaded (i.e. the number of client requests in progress does not exceed a predefined threshold.), the required system resources are allocated to the client as it is done in a classic client/server network using the streaming protocol HTTP. If the server is overloaded, all incoming requests are dispatched to the tracker. In case the tracker is not available, the primary server switches its performance to the `tracker mode’. Hence, it provides for clients requesting the video a list indexing all the available peers. The list refers to the peers that have already downloaded the desired content and authorize sharing it with other peers. Therefore, the client will download the requested content from one or several peers, which reduces the server load and decreases the rejected request’s rate. P2P Web introduces an adapted version to the Web exchanges of the BitTorrent protocol, which allows the downloading of a content from one or several peers having received the requested media on their own devices. FIGURE 6. Open in new tabDownload slide P2P Web network architecture. FIGURE 6. Open in new tabDownload slide P2P Web network architecture. FIGURE 7. Open in new tabDownload slide Diagram of P2P Web performance. FIGURE 7. Open in new tabDownload slide Diagram of P2P Web performance. The main contribution of the P2P Web approach is the collaboration of two different streaming protocols (i.e. HTTP and BitTorrent) in the same network architecture to enhance system performance by reducing the amount of rejected requests [56]. Besides, the MPD contains representations of one content (e.g. various qualities of the video, different formats, etc.). This allows the end user to choose content according to his preference. 5. RESULTS AND COMPARISON When referring to content replication techniques regardless of the network structure, they provide mainly data availability and resource optimization. Integrity is also considered since data replication is a matter of discussion in assuring content integrity. Therefore, most of the mentioned approaches address resource optimization, quick query response, availability and scalability. The resource optimization is generally performed by optimizing the storage capacity. Table 1 summarizes the performance of each of the replication approaches discussed in this paper in terms of availability increasing and caching management. TABLE 1. Performances of replication approaches. Performance . ECRA . DST . SACM-FCADMS . Hermes and AREN . Pop Store . Availability enhancement Content download from peers’ high signal strength, available energy and available bandwidth. Reducing the number of exchanged messages and decreasing request rejection rate. Only the high requested segments are replicated so that the availability of the most demanded content will increase. Classifying video on popular/unpopular and only replicating the popular video. Optimizing replication for all videos (not only for the most popular) and considering dynamic popularity variation. Caching management Choosing peers with the highest cache space to store replicas. Reserving specific nodes for replicas storage and optimizing the read and write process. Choosing the least loaded server which has available disk space for replicas storage. No specific caching strategy. Avoiding distant access by placing replicas in the data centers with the highest access frequency. Performance . ECRA . DST . SACM-FCADMS . Hermes and AREN . Pop Store . Availability enhancement Content download from peers’ high signal strength, available energy and available bandwidth. Reducing the number of exchanged messages and decreasing request rejection rate. Only the high requested segments are replicated so that the availability of the most demanded content will increase. Classifying video on popular/unpopular and only replicating the popular video. Optimizing replication for all videos (not only for the most popular) and considering dynamic popularity variation. Caching management Choosing peers with the highest cache space to store replicas. Reserving specific nodes for replicas storage and optimizing the read and write process. Choosing the least loaded server which has available disk space for replicas storage. No specific caching strategy. Avoiding distant access by placing replicas in the data centers with the highest access frequency. Open in new tab TABLE 1. Performances of replication approaches. Performance . ECRA . DST . SACM-FCADMS . Hermes and AREN . Pop Store . Availability enhancement Content download from peers’ high signal strength, available energy and available bandwidth. Reducing the number of exchanged messages and decreasing request rejection rate. Only the high requested segments are replicated so that the availability of the most demanded content will increase. Classifying video on popular/unpopular and only replicating the popular video. Optimizing replication for all videos (not only for the most popular) and considering dynamic popularity variation. Caching management Choosing peers with the highest cache space to store replicas. Reserving specific nodes for replicas storage and optimizing the read and write process. Choosing the least loaded server which has available disk space for replicas storage. No specific caching strategy. Avoiding distant access by placing replicas in the data centers with the highest access frequency. Performance . ECRA . DST . SACM-FCADMS . Hermes and AREN . Pop Store . Availability enhancement Content download from peers’ high signal strength, available energy and available bandwidth. Reducing the number of exchanged messages and decreasing request rejection rate. Only the high requested segments are replicated so that the availability of the most demanded content will increase. Classifying video on popular/unpopular and only replicating the popular video. Optimizing replication for all videos (not only for the most popular) and considering dynamic popularity variation. Caching management Choosing peers with the highest cache space to store replicas. Reserving specific nodes for replicas storage and optimizing the read and write process. Choosing the least loaded server which has available disk space for replicas storage. No specific caching strategy. Avoiding distant access by placing replicas in the data centers with the highest access frequency. Open in new tab To present the different challenges covered in the studied approaches, we select some parameters to demonstrate the capability of each technique in terms of QoS enhancement. The comparison among different techniques that have been outlined in this paper is shown in Table 2. TABLE 2. Comparison of the different approaches. Feature . FCS-MS [30] . Multi threading concept [50] . Pop Store [44] . SACM-FCADMS [43] . P2P Web [57] . MS integration [49] . ECRA [72] . HERMES and AREN [36] . DST [39] . Streaming over DDS [12] . Consider bandwidth consumption Yes (fair bandwidth sharing) Yes Yes Yes Yes Yes Yes Yes Yes Yes Architecture Centralized client/server (use of master server) Client/ server HDFS [73] Client/ server Hybrid (P2P and C/S) Metro/ access network P2P CDN P2P Client/ server Availability improvement Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Frame-loss reduction Yes Yes No No No No No No No Yes Response-time reduction Yes (but not effective in case of heavy workload) No Yes Yes No No Yes Yes Yes Yes Scalability Yes Yes No Yes Yes Yes Yes No Yes Yes Reliability No Yes No Yes Yes No No No No Yes Data consistency enhancement No No Yes No No No No No Yes Yes Load balancing Yes No Yes Yes Yes Yes Yes Yes Yes Yes Congestion reduction Yes No Yes Yes No Yes No No Yes Yes Storage capacity No Yes Yes Yes No No Yes Yes No No Cost reduction No No Yes No No Yes Yes No Yes Yes Energy-efficient No No No No No Yes Yes No No No Feature . FCS-MS [30] . Multi threading concept [50] . Pop Store [44] . SACM-FCADMS [43] . P2P Web [57] . MS integration [49] . ECRA [72] . HERMES and AREN [36] . DST [39] . Streaming over DDS [12] . Consider bandwidth consumption Yes (fair bandwidth sharing) Yes Yes Yes Yes Yes Yes Yes Yes Yes Architecture Centralized client/server (use of master server) Client/ server HDFS [73] Client/ server Hybrid (P2P and C/S) Metro/ access network P2P CDN P2P Client/ server Availability improvement Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Frame-loss reduction Yes Yes No No No No No No No Yes Response-time reduction Yes (but not effective in case of heavy workload) No Yes Yes No No Yes Yes Yes Yes Scalability Yes Yes No Yes Yes Yes Yes No Yes Yes Reliability No Yes No Yes Yes No No No No Yes Data consistency enhancement No No Yes No No No No No Yes Yes Load balancing Yes No Yes Yes Yes Yes Yes Yes Yes Yes Congestion reduction Yes No Yes Yes No Yes No No Yes Yes Storage capacity No Yes Yes Yes No No Yes Yes No No Cost reduction No No Yes No No Yes Yes No Yes Yes Energy-efficient No No No No No Yes Yes No No No Open in new tab TABLE 2. Comparison of the different approaches. Feature . FCS-MS [30] . Multi threading concept [50] . Pop Store [44] . SACM-FCADMS [43] . P2P Web [57] . MS integration [49] . ECRA [72] . HERMES and AREN [36] . DST [39] . Streaming over DDS [12] . Consider bandwidth consumption Yes (fair bandwidth sharing) Yes Yes Yes Yes Yes Yes Yes Yes Yes Architecture Centralized client/server (use of master server) Client/ server HDFS [73] Client/ server Hybrid (P2P and C/S) Metro/ access network P2P CDN P2P Client/ server Availability improvement Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Frame-loss reduction Yes Yes No No No No No No No Yes Response-time reduction Yes (but not effective in case of heavy workload) No Yes Yes No No Yes Yes Yes Yes Scalability Yes Yes No Yes Yes Yes Yes No Yes Yes Reliability No Yes No Yes Yes No No No No Yes Data consistency enhancement No No Yes No No No No No Yes Yes Load balancing Yes No Yes Yes Yes Yes Yes Yes Yes Yes Congestion reduction Yes No Yes Yes No Yes No No Yes Yes Storage capacity No Yes Yes Yes No No Yes Yes No No Cost reduction No No Yes No No Yes Yes No Yes Yes Energy-efficient No No No No No Yes Yes No No No Feature . FCS-MS [30] . Multi threading concept [50] . Pop Store [44] . SACM-FCADMS [43] . P2P Web [57] . MS integration [49] . ECRA [72] . HERMES and AREN [36] . DST [39] . Streaming over DDS [12] . Consider bandwidth consumption Yes (fair bandwidth sharing) Yes Yes Yes Yes Yes Yes Yes Yes Yes Architecture Centralized client/server (use of master server) Client/ server HDFS [73] Client/ server Hybrid (P2P and C/S) Metro/ access network P2P CDN P2P Client/ server Availability improvement Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Frame-loss reduction Yes Yes No No No No No No No Yes Response-time reduction Yes (but not effective in case of heavy workload) No Yes Yes No No Yes Yes Yes Yes Scalability Yes Yes No Yes Yes Yes Yes No Yes Yes Reliability No Yes No Yes Yes No No No No Yes Data consistency enhancement No No Yes No No No No No Yes Yes Load balancing Yes No Yes Yes Yes Yes Yes Yes Yes Yes Congestion reduction Yes No Yes Yes No Yes No No Yes Yes Storage capacity No Yes Yes Yes No No Yes Yes No No Cost reduction No No Yes No No Yes Yes No Yes Yes Energy-efficient No No No No No Yes Yes No No No Open in new tab In Table 2, regardless of their architecture, almost all approaches considered the bandwidth consumption, availability improvement, scalability and load balancing. They focused mostly on the content replication management and the optimization of the hierarchical structure of the network to absorb data centers overloading. Although the problem of high cost is only scaled in the CDNs domain, we notice that most of the works based on the client–server architecture did not consider cost reduction in their contributions. Moreover, few of them addressed the issues of frame loss, data consistency and especially energy efficiency. 6. OPEN ISSUES AND FUTURE DIRECTIONS Although research on QoS enhancement of VoD systems has already achieved much, there are still some problems that remain to be considered [80, 81]. To help define a better grasp of research directions in this field, more insight into some future trends and research challenges are presented as follows: 6.1. Measurement and modeling for QoS of VOD system QoS measurement and QoS parameter negotiation are critical for video streaming systems to make their service more efficient. Therefore, this issue remains challenging and it will engage researches in the field of QoS management for a long time. It is assumed that the QoE determines the QoS [82, 83], and at the same time, the quality of the delivered content affects the degree of satisfaction of the end user [57]. For instance, video streaming systems focus on the perspective of user experience to estimate the customer satisfaction of the VoD system [58]. Furthermore, the process of QoS parameter measurement is tightly linked to the QoE representation. Hence, a good representation of the QoE of end users over data analysis can help provide QoS metrics to be better assumed by the video streaming system. However, the process of experience sampling is still an open issue for VoD systems. The streaming system had to use effective policies to get users to collaborate and share their QoS metrics. The authors of [71] believe that the hardest part of QoS parameters management is the retrieving of the metrics defining the QoE of users [84], whereas the number of served customers is essential. Therefore, adopting optimal data analytics that retrieves QoE parameters from end users and then performs a comparison between the provided quality and the desired one is vital. 6.2. Video popularity variation VoD systems need to determine video popularity to help optimize replication system operations [59]. For instance, various studies [60] [61] assume that video popularity affects the performance of the caching strategy directly. Data popularity distribution is quite simple to calculate in a given time interval. Popularity estimation is generally based on analyzing video access frequency over time. Furthermore, dynamic data popularity variation makes the process of content popularity prevision more complex [62]. The main issue is that the number of users and view time is not regular per day. For the same user, the probability of replying to the same video per day is not apparent. Moreover, the continuous arrival of new videos can change the popularity distribution between two-time intervals. It remains a research challenge to handle dynamic popularity variation, and a better popularity estimation mechanism has to be addressed in the future. 6.3. Synchronized multimedia data It is challenging to capture the different factors affecting video streaming quality, given the change of the network statue on a time scale of several minutes. For instance, there is an extensive literature on characterizing the effect of network performance parameters on client buffer requirements in real-time video streaming systems [63, 64]. To optimize the content dispatching strategy, low end-to-end delay remains a design goal for distributed VoD systems. The end-to-end delay depends on the scheduling algorithm running on the active network switches and the source traffic representation [65]. Ideally, the end user must receive the demanded content continuously without breaks. Thus, the system should be aware of the delay bounds for various traffic specifications. The main issue regarding delay bounds supervision is that the sources in an integrated network supporting video services have various characteristics [66]. Besides, the complexity of delay determination rises in a heterogeneous networking environment (i.e. with multiple specifications, different scheduling algorithms for each switch, etc.). Moreover, some studies consider placing the demanded content on the most accessed data centers. Others propose to geo-locate the end user and then correlate its request with the nearest node [67]. This can help to reduce the end-to-end delay. However, network traffic is naturally unpredictable, and its characteristics should be permanently measured. Future researches should address more the concept of the delay guarantee to a video segment based on its expected arrival time. 6.4. Mobility and data errors Mobile nodes use various transmission ranges (Low, medium and high) to keep connected to the Internet and the other nodes in the network. During mobility, frequent disconnections from the streaming server or the node sharing the content make it hard to avoid transmission errors [68]. Moreover, it is quite difficult to recover from mistakes that occurred while the node is moving. For instance, the placement of data streaming servers at different levels of the network may be the fluent strategy to deal with streaming discontinuity during mobility. However, this solution seems to be very costly considering extensive video CDNs. In this case, a better solution may consist of using a heterogeneous environment (P2P and client/server exchange) [69]. Furthermore, the mobile node collaborates with the network to estimate the transmission range and the upcoming disconnection from the streaming server. The network is then alerted, and it may dispatch the streaming specifications to the available server or peer at the new range of the mobile node. This field requires a lot of attention in future work since the astonishing increase in the use of mobile streaming networks (typically in vehicular ad hoc networks.). 7. CONCLUSION In this paper, we have presented a survey of issues of QoS specification and QoS enhancement in VoD systems to provide a better experience to the end user. We address the feedback technique as a practical and useful scheme to perform QoS measurements within the end user so that the system can manage the self-adjustment of provided QoS specifications. We present as well the impact of admission control and resource allocation scheduling on the frame-loss rate and the congestion issues. We employ the following topics to distinguish between different QoS enhancement approaches replication strategy, replica placement, distribution management among different hierarchical structures of a network and the streaming protocol used to carry data flows. We discuss the contribution and the limitation of the surveyed literature. In conclusion, most of the approaches focused on reducing bandwidth consumption and improving video data availability. In some cases, they dealt with energy consumption and cost reduction by blending CDN and P2P networks in the same hierarchical structure or by reducing signalization traffic load, etc. More work has to be done to reduce frame loss and to increase video data consistency so that the system can be able to provide a continuous streaming service. REFERENCES [1] Pathan , M. , Rajkumar , B. and Vakali , A. ( 2008 ) Content Delivery Networks: State of the Art, Insights, and Imperatives. In Content Delivery Networks , pp. 3 – 32 . Springer , Berlin, Heidelberg . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [2] Ha , I. , Wildman , S.S. and Bauer , J.M. ( 2010 ) P2P, CDNs, and hybrid networks: the economics of Internet video distribution . Int. Telecommun. Policy Rev. , 4 , 1 – 22 . Google Scholar OpenURL Placeholder Text WorldCat [3] Cerqueira , E. , Zeadally , S., Leszczuk , M., Curado , M. and Mauthe , A. ( 2010 ) Recent advances in multimedia networking . Multimed. Tools Appl. , 54 , 635 – 647 . doi: 10.1007/s11042-010-0578-z . Google Scholar Crossref Search ADS WorldCat [4] Vijendran , A.S. and Thavamani , S. ( 2012 ) Analysis study on caching and replica placement algorithm for content distribution in distributed computing networks . Int. J. Peer Peer Netw. , 3 , 6 . Google Scholar OpenURL Placeholder Text WorldCat [5] Alghazo , J. , Akaaboune , A. and Botros , N. ( 2004 ) SF-LRU Cache Replacement Algorithm . In IEEE Int. Workshop on Memory Technology, Design and Testing (MTDT’04) . [6] Haque , M.S. , Peddersen , J. and Parameswaran , S. ( 2011 ) CIPARSim: Cache Intersection Property Assisted Rapid Single-Pass FIFO Cache Simulation Technique . In Proc. Int. Conf. Computer-Aided Design (ICCAD '11) , pp. 126 – 133 . [7] Zhou , Y. , Fu , T.Z.J. and Chiu , D.M. ( 2013 ) On replication algorithm in P2P VoD . IEEE/ACM Trans. Networking , 21 . Google Scholar OpenURL Placeholder Text WorldCat [8] Zhou , Y. , Fu , T.Z.J. and Chiu , D.M. ( 2012 ) Server-assisted adaptive video replication for P2P VoD . Signal Process. Image Commun. , 27 , 484 – 495 . Google Scholar Crossref Search ADS WorldCat [9] Floratou , A. , Megiddo , N., Potti , N., Ozcan , F., Kale , U. and Schmitz-Hermes , J. ( 2015 ) Adaptive Caching Algorithms for Big Data Systems . Computer Science, RJ10531 (ALM1509-001) . [10] Gomes , C. and Hempstead , M. ( 2015 ) Combative Cache Efficacy Techniques: Cache Replacement in the Context of Independent Prefetching in Last Level Cache. In 33rd IEEE Int. Conf. Computer Design (ICCD) , New York , pp. 423 – 426 . [11] Ma , Z. , Xu , K. and Zhong Y. ( 2012 ) Exploring the policy selection of P2P VoD System — A simulation based research . 2012 IEEE 20th International Workshop on Quality of Service Coimbra , pp. 1–4, doi: 10.1109/IWQoS.2012.6245991 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [12] Al-Madani , B. , Al-Roubaiey , A. and Baig , Z.A. ( 2014 ) Real-time QoS-aware video streaming: a comparative and experimental study . Adv. Multimedia , 2014 , 11 . Google Scholar Crossref Search ADS WorldCat [13] Al-hammouri , M. , Madani , B., Aloqaily , M., Ridhawi and Jararweh , Y. ( 2018 ) Scalable Video Streaming for Real-Time Multimedia Applications over DDS Middleware for Future Internet Architecture . In 2018 IEEE/ACS 15th Int. Conf. Computer Systems and Applications (AICCSA) , Aqaba, pp. 1 – 6 . [14] Torres-Cruz , N. , Rivero-Angeles , M.E., Rubino , G., Menchaca-Mendez , R. and Menchaca-Mendez , R. ( 2018 ) An efficient resource allocation scheme for VoD services over window-based P2P networks . Multimed. Tools Appl. , 77 , 31427 – 31445 . Google Scholar Crossref Search ADS WorldCat [15] Alaya , B. , Duvallet , C. and Sadeg , B. ( 2009 ) A new approach to manage QoS in distributed multimedia systems . Int. J. Comput. Sci. Inf. Secur. , 2 . Google Scholar OpenURL Placeholder Text WorldCat [16] Alaya , B. , Duvallet , C. and Sadeg , B. ( 2010 ) Feedback Architecture for Multimedia Systems . In ACS/IEEE Int. Conf. Computer Systems and Applications—AICCSA . [17] Vinay , A. , Prakash , A., Kiran Kumar , D.S., Nagabhushanet , K. and Anitha , T.N. A Novel and Optimal Video Replication Technique for Video-on-Demand Systems . In Int. Conf. and Workshop on Emerging Trends in Technology (ICWET 2011)—TCET , Mumbai, Maharashtra, India, February 25–26, 2011. [18] Kim , S. , Oh , H. and Kim , C. ( 2015 ) Energy cognitive dynamic adaptive streaming over HTTP . KSII Trans. Internet Inf. Syst. , 9 . Google Scholar OpenURL Placeholder Text WorldCat [19] Lo , C.W. and Su , Y.Y. ( 2014 ) P2P Video Streaming Replication Scheme for P2P VoD Services. In Int. Conf. Information and Communication Technology Convergence (ICTC) . Busan . [20] Jagadeesh , M. , Dyaberi , K.K. and Pai , V.S. ( 2010 ) Storage Optimization for a Peer-to-Peer Video-on-Demand Network. In Proc. First Annual ACM SIGMM Conf. Multimedia Systems , MMSys 2010, Phoenix, Arizona, USA, February 22–23, 2010. [21] Hareesh , K. and Manjaiah , D.H. ( 2011 ) Peer-to-peer live streaming and video on demand design issues and its challenges . Int. J. Peer Peer Netw. , 2 . Google Scholar OpenURL Placeholder Text WorldCat [22] Kawasaki , Y. , Matsumoto , N. and Yoshida , N. ( 2006 ) Popularity-based content replication in peer-to-peer networks . Alexandrov, V.N. et al. (eds.) . ICCS 2006, Part IV. LNCS , 3994 , 436 – 443 . Google Scholar OpenURL Placeholder Text WorldCat [23] Wen , Z. , Liu , N., Yeung , K.L. and Lei , Z. ( 2011 ) Closest Playback-Point First: A New Peer Selection Algorithm for P2P VoD Systems . In Global Telecommunications Conf. (GLOBECOM 2011) , pp. 1 – 5 . IEEE . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC [24] Yang , Y. et al. ( 2010 ) Improving QoS in BitTorrent-like VoD Systems . In INFOCOM, 2010 Proceedings . IEEE . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [25] D’Acunto , L. et al. ( 2013 ) BitTorrent-like P2P approaches for VoD: a comparative study . Comput. Networks , 57 , 1253 – 1276 . Google Scholar Crossref Search ADS WorldCat [26] Silvestre , G. , Monnet , S., Krishnaswamy , R. and Sens , P. ( 2012 ) Aren: A Popularity Aware Replication Scheme for Cloud Storage . In IEEE 18th Int. Conf. Parallel and Distributed Systems (ICPADS) . [27] Hara , T. and Madria , S.K. ( 2009 ) Consistency management strategies for data replication in mobile ad hoc networks . IEEE Trans. Mob. Comput. , 8 , 950 – 967 . Google Scholar Crossref Search ADS WorldCat [28] Alaya , B. , Zidi , S., Laouamer , L. and Moulahi , T. ( 2015 ) Effect of selective replication strategy and dynamic admission control to QoS management in video on demand systems . J. Innovation Digital Ecosyst . Google Scholar OpenURL Placeholder Text WorldCat [29] Beji , M. , Alaya , B. and Duvallet , C. ( 2016 ) Payoff-Based Dynamic Segment Replication Policy in Distributed VOD System . In 2016 Eleventh Int. Conf. Digital Information Management (ICDIM) , Porto, pp. 151 – 156 . [30] Xiaobo , Z. and Zhong , X.C. ( 2007 ) Efficient algorithms of video replication and placement on a cluster of streaming servers . J. Netw. Comput. Appl. , 30 , 1084 – 8045 . Google Scholar OpenURL Placeholder Text WorldCat [31] Lukman , A. , Sun , Z. and Haitham , C. ( 2017 ) QoS based admission control using multipath scheduler for IP over satellite networks . Int. J. Electr. Comput. Eng. , 7 , 2958 – 2969 . Google Scholar OpenURL Placeholder Text WorldCat [32] Guidotti , A. , Vanelli-Coralli , A., Foggi , T., Colavolpe , G., Caus , M., Bas , J., Cioni , S. and Modenini , A. ( 2018 ) LTE-based satellite communications in LEO mega-constellations . Int. J. Satell. Commun. Networking , 1 – 16 . Google Scholar OpenURL Placeholder Text WorldCat [33] Pires , K. , Monnet , S. and Sens , P. ( 2014 ) POPS: a popularity-aware live streaming service . 20th IEEE Int. Conf. Parallel and Distributed Systems (ICPADS) , Hsinchu . Google Scholar OpenURL Placeholder Text WorldCat [34] Wang , Q. , Daudjee , K. and Özsu , M.T. ( 2008 ) Popularity-Aware Prefetch in P2P Range Caching . Eighth Int. Conf. Peer-to-Peer Computing Aachen, September . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [35] Carlier , A. , Charvillat , V. and Ooi , W.T. ( 2015 ) A Video Timeline with Bookmarks and Prefetch State for Faster Video Browsing . In Proc. 23rd ACM Int. Conf. Multimedia , Brisbane, Australia. Google Scholar OpenURL Placeholder Text WorldCat [36] Silvestre , G. , Monnet , S., Buffoni , D. and Sens , P. Predicting Popularity and Adapting Replication of Internet Videos for High-Quality Delivery . Seoul , 2013, pp. 412–419, doi: 10.1109/ICPADS.2013.64 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [37] Gui-min , H.U.A.N.G. , Qian , B.A.I., Ping-shan , L.I.U. and Ya , Z.H.O.U. ( 2017 ) A Dynamic Popularity-based Cache Replication Strategy for P2P VoD Systems . In The Int. Conf. Computer Science and Technology , pp. 134 – 140 . [38] Borsje , J. , Levering , L. and Frasincar , F. ( 2008 ) Hermes: A Semantic Web-Based News Decision Support System . Proc. ACM Symposium on Applied Computing , 2415 – 2420 . Google Scholar OpenURL Placeholder Text WorldCat [39] Victer Paul , P. , Saravanan , N., Jayakumar , S.K.V., Dhavachelvan , P. and Baskaran , R. ( 2012 ) QoS enhancements for global replication management in peer to peer networks . Future Gener. Comput. Syst. , 28 , 573 – 582 . Google Scholar Crossref Search ADS WorldCat [40] Gharaibeh , A. , Khreishah , A., Ji , B. and Ayyash , M. A provably efficient online collaborative caching algorithm for multicell-coordinated systems . IEEE Trans. Mob. Comput. doi: 10.1109/TMC.2474364.2015 . OpenURL Placeholder Text WorldCat Crossref [41] Sun , F. , Liu , B., Hou , F., Zhou , H., Chen , J., Rui , Y. and Lin , G. ( 2015 ) A QoE centric distributed caching approach for vehicular video streaming in cellular networks . Wireless Commun. Mobile Comput. . doi: 10.1002/wcm.2636 . Google Scholar OpenURL Placeholder Text WorldCat Crossref [42] Li , Z. and Simon , G. ( 2014 ) Cooperative caching in a content centric network for video stream delivery . J. Netw. Syst. Manag. . doi: 10.1007/s10922-014-9300-1 . Google Scholar OpenURL Placeholder Text WorldCat Crossref [43] Vo , P.L. , Le , T.-A., Hong , C.S., Moon , S.I., Lee , S. and Tu , N.L. ( 2015 ) Cooperative Caching for Adaptive Bit Rate Streaming in Content Delivery Networks . In Int. Conf. Ubiquitous Information Management and Communication (ICUIMC) . [44] Myint , J. and Hunger , A. ( 2014 ) Comparative Analysis of Adaptive File Replication Algorithms for Cloud Data Storage. In Int. Conf. Future Internet of Things and Cloud . [45] Sun , D.-W. , Chang , G.-R., Gao , S., Jin , L.-Z. and Wang , X.-W. ( 2012 ) Modeling a dynamic data replication strategy to increase system availability in cloud computing environments . J. Comput. Sci. Technol. , 27 , 256 – 272 . Google Scholar Crossref Search ADS WorldCat [46] Dogra , N. and Singh , S. ( 2015 ) A survey of dynamic replication strategies in distributed systems . Int. J. Comput. Appl. , 110 . Google Scholar OpenURL Placeholder Text WorldCat [47] Chang , R.-S. , Chang , H.-P. and Wang , Y.-T. ( 2008 ) A Dynamic Weighted Data Replication Strategy in Data Grids . In 2008 IEEE/ACS Int. Conf. Computer Systems and Applications , pp. 414 – 421 . [48] Majd , G. , Radwan , E.R., Abdelaziz , C. and Mohamad , R. ( 2015 ) Client/Server and Peer-to-Peer Hybrid Architecture for Adaptive Video Streaming . In Communications, Signal Processing, and their Applications (ICCSPA), Int. Conf. , Sharjah . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC [49] Fratini , R. , Savi , M., Verticale , G. and Tornatore , M. Using Replicated Video Servers for VoD Traffic Offloading in Integrated Metro/Access Networks . In European Community Seventh Framework Programme FP7/2013–2015 . [50] Liu , Y. , Yang , G. and Liang , C. ( 2008 ) A survey on peer-to-peer video streaming systems . Peer Peer Netw. Appl. , 1 , 18 – 28 . doi: 10.1007/s12083-007-0006-y . Google Scholar Crossref Search ADS WorldCat [51] Rohini , G. and Srinivasan , A. ( 2016 ) Multi server based cloud-assisted real-time transrating for HTTP live streaming . Indian J. Sci. Technol. , 9 . doi: 10.17485/ijst/2016/v9i3/78864 . Google Scholar OpenURL Placeholder Text WorldCat Crossref [52] Seufert , M. , Egger , S., Slanina , M., Zinner , T., Hoßfeld , T. and Tran-Gia , P. ( 2015 ) A survey on quality of experience of HTTP adaptive streaming . IEEE Commun. Surv. Tutorials , 17 , 469 – 492 . Google Scholar Crossref Search ADS WorldCat [53] Chang , R.-I. , Liu , Y.-C., Ho , J.-M., Chu , Y.-H., Chung , W.-C. and Wu , C.-J. ( 2015 ) Optimal Scheduling of QoE-Aware HTTP Adaptive Streaming . Research Center for Information Technology Innovation , Academia Sinica, Taiwan . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [54] Qadir , Q.M. , Kist , A.A. and Zhang , Z. ( 2015 ) Mechanisms for QoE optimisation of video traffic: a review paper . Int. J. Inform. Commun. Technol. Appl. , 1 . Google Scholar OpenURL Placeholder Text WorldCat [55] Bouzakaria , N. , Ghareeb , M. and Parrein , B. Une architecture hybride Client/serveur et Pair-a-Pair pour le streaming vidéo sur l'Internet . LUNAM Université, Université de Nantes, IRCCyN UMR CNRS 6597 . Poly Tech Nantes , France . [56] Wei , X. , Ding , P., Zho , F., Lou , J. and Gao , Y. ( 2019 ) A Load Balancing Strategy Based on Request Queue for P2P-VoD System . In 2019 15th Int. Wireless Communications & Mobile Computing Conference (IWCMC) , Tangier, Morocco , pp. 668 – 673. [57] Kryftis , Y. , Mavromoustakis , C.X., Mastorakis , G., Pallis , E., Batalla , J.M., Rodrigues , J.J.P.C., Dobre , C. and Kormentzas , G. ( 2015 ) Resource Usage Prediction Algorithms for Optimal Selection of Multimedia Content Delivery Methods . In IEEE Int. Conf. Communications (ICC) , pp. 5903 – 5909 . [58] Nourikhah , H. and Akbari , M.K. ( 2016 ) Impact of service quality on user satisfaction: Modeling and estimating distribution of quality of experience using Bayesian data analysis . Electron. Commer. Res. Appl. , 17 , 112 – 122 . Google Scholar Crossref Search ADS WorldCat [59] Li , Z. , Wu , Q., Salamatian , K. and Xie , G. ( 2015 ) Video delivery performance of a large-scale VoD system and the implications on content delivery . IEEE Trans. Multimed. , 17 , 880 – 892 . Google Scholar Crossref Search ADS WorldCat [60] Ling , Q. , Xu , L., Yan , J. and Zhang , Y. ( 2015 ) An adaptive caching algorithm suitable for time-varying user accesses in VOD systems . Multimed. Tools Appl. , 74 , 11117 – 11137 . Google Scholar Crossref Search ADS WorldCat [61] Pires , K. and Simon , G. ( March 2015 ) YouTube Live and Twitch: A Tour of User-Generated Live Streaming Systems . 6th ACM Multimedia Systems Conf. , Portland, United States , pp. 225 – 230 . [62] Ali-Eldin , A. , Kihl , M., Tordsson , J. and Elmroth , E. ( 2015 ) Analysis and Characterization of a Video-on-Demand Service Workload . Proc. 6th ACM Multimedia Systems Conf. , 189 – 200 . Google Scholar OpenURL Placeholder Text WorldCat [63] Huang , G.M. , Liu , P.S. and Gong , X. ( 2015 ) A Novel Peer Selection Strategy in P2P VoD System Using Biased Gossip . IEEE Int. Commun. Software and Networks (ICCSN) , 372 – 377 . Google Scholar OpenURL Placeholder Text WorldCat [64] Guomin , Z. , Chao , H., Wang , N., Wei , X. and Xing , C. ( 2014 ) A Novel Scheme for Improving Quality of Service of Live Streaming. In Int. Conf. P2P, Parallel, Grid, Cloud and Internet Computing (3PGCIC) , pp. 457 – 462 . [65] Kim , J. , Caire , G. and Molisch , A.F. ( 2015 ) Quality-aware streaming and scheduling for device-to-device video delivery . IEEE/ACM Trans. Networking , 24 , 2319 – 2331 . Google Scholar Crossref Search ADS WorldCat [66] Li , Y. , Dai , S. and Chang , X. ( 2015 ) Delay guaranteed VoD services over group-based integrated fiber-wireless (FiWi) access networks with energy efficiency . Opt. Fiber Technol. , 24 , 100 – 105 . Google Scholar Crossref Search ADS WorldCat [67] Hu , H. , Wen , Y., Chua , T.-S., Huang , J., Zhu , W. and Li , X. ( 2015 ) Joint content replication and request routing for social video distribution over cloud CDN: a community clustering method . IEEE Trans. Circuits Syst. Video Technol. . Google Scholar OpenURL Placeholder Text WorldCat [68] Kolios , P. , Papadaki , K. and Friderikos , V. ( 2016 ) Energy efficient mobile video streaming using mobility . Comput. Networks , 94 , 189 – 204 . Google Scholar Crossref Search ADS WorldCat [69] Ojanperä , T. , Luoto , M., Majanen , M., Mannersalo , P. and Savolainen , P.T. ( 2015 ) Cognitive network management framework and approach for video streaming optimization in heterogeneous networks . Wireless Pers. Commun. , 84 , 1739 – 1769 . Google Scholar Crossref Search ADS WorldCat [70] Arpan , M. , Nidhi , H. and Marc , L. ( 2018 ) Optimal Content Replication and Request Matching in Large Caching Systems . In IEEE Conf. Computer Communications , pp. 288 – 296 . [71] Seufert , M. , Egger , S., Slanina , M., Zinner , T., Hoßfeld , T. and Tran-Gia , P. A survey on quality of experience of HTTP adaptive streaming . IEEE Commun. Surv. Tutorials , doi: http://dx.doi.org/10.1109/COMST.2014.2360940 . OpenURL Placeholder Text WorldCat Crossref [72] Alshayeji , H. and Dias , D.N. ( 2015 ) Enhanced chunk regulation algorithm for superior QoS in heterogeneous P2P video on demand . J. Netw. , 10 . Google Scholar OpenURL Placeholder Text WorldCat [73] Kala Karun , A. and Chitharanjan , K. ( 2013 ) A Review on Hadoop—HDFS Infrastructure Extensions . Proc. 2013 IEEE Conf. Information and Communication Technologies , doi: 10.1109/CICT.2013.6558077 . Google Scholar OpenURL Placeholder Text WorldCat Crossref [74] Raimund , S. et al. ( 2018 ) QoE Management for Future Networks . In Ganchev , I., van der Mei , R., van den Berg , H. (eds) Autonomous Control for a Reliable Internet of Services . Lecture Notes in Computer Science , Vol. 10768 . Springer , Cham . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC [75] Schatz , R. et al. ( 2018 ) QoE Management for Future Networks . In Ganchev , I., van der Mei , R., van den Berg , H. (eds) Autonomous Control for a Reliable Internet of Services . Lecture Notes in Computer Science, Vol. 10768 , Springer , Cham . OpenURL Placeholder Text WorldCat [76] Wei , X. , Ding , P., Zho , F., Lou , J. and Gao , Y. ( 2019 ) A Load Balancing Strategy based on Request Queue for P2P-VoD System . In 15th Int. Wireless Communications and Mobile Computing Conf. (IWCMC) , pp. 668 – 673 . [77] Wei , X. , Ding , P., Zhou , L. and Qian , Y. ( 2019 , 2019 ) QoE oriented chunk scheduling in P2P-VoD streaming system . IEEE Trans. Veh. Technol. , 35 , 8012 – 8025 , doi 10.1109/TVT.2019.2922273 . Google Scholar Crossref Search ADS WorldCat [78] Abdelkader , B. , Korichi , A., Bourouis , A. and Alreshoodi , M. ( 2019 ) Survey on QoE\QoS correlation models for video streaming over vehicular ad-hoc networks . J. Comput. Inform. Technol. , 26 , 267 – 278 . Google Scholar Crossref Search ADS WorldCat [79] Dantas , J. , Matos , R., Melo , C., Araujo , J., Ferreira , J. and Maciel , P. ( 2019 ) Evaluation of Encoding and Network Aspects on Video Streaming Performance: A Modeling and Experimental Approach . In IEEE Int. Conf. Systems, Man, and Cybernetics (SMC) , Miyazaki, Japan , pp. 3883 – 3888 . [80] Bouraqia , K. , Sabir , E., Sadik , M. and Ladid , L. ( 2020 ) Quality of experience for streaming services: measurements, challenges and insights . IEEE Access , 8 , 13341 – 13361 . Google Scholar Crossref Search ADS WorldCat [81] Afzal , S. , Testoni , V., Rothenberg , C.E., Kolan , P. and Bouazizi , I. ( 2019 ) A holistic survey of wireless multipath video streaming . IEEE Commun. Surv. Tutorials , 1 – 42 . Google Scholar OpenURL Placeholder Text WorldCat [82] Barakabitze , A.A. et al. ( 2020 ) QoE management of multimedia streaming services in future networks: a tutorial and survey . IEEE Commun. Surv. Tutorials , 22 , 526 – 565 . Google Scholar Crossref Search ADS WorldCat [83] Seufert , M. , Egger , S., Slanina , M., Zinner , T., Hoßfeld , T. and TranGia , P. ( 2015 ) A survey on quality of experience of HTTP adaptive streaming . IEEE Commun. Surv. Tutorials , 17 , 469 – 492 . Google Scholar Crossref Search ADS WorldCat [84] Martini , M.G. , Chen , C.W., Chen , Z., Dagiuklas , T., Sun , L. and Zhu , X. ( 2012 ) Guest editorial QoE-aware wireless multimedia systems . IEEE J. Sel. Areas Commun. , 30 , 1153 – 1156 . Google Scholar Crossref Search ADS WorldCat © The British Computer Society 2020. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - QoS Enhancement In VoD Systems: Load Management And Replication Policy Optimization Perspectives
JF - The Computer Journal
DO - 10.1093/comjnl/bxaa060
DA - 2020-10-19
UR - https://www.deepdyve.com/lp/oxford-university-press/qos-enhancement-in-vod-systems-load-management-and-replication-policy-99CcXBkEht
SP - 1547
EP - 1563
VL - 63
IS - 10
DP - DeepDyve
ER -