336 IEEE TRANSACTIONS ON BROADCASTING, VOL. 52, NO. 3, SEPTEMBER 2006 CAR: A Low Latency Video-on-Demand Broadcasting Scheme for Heterogeneous Receivers Chin-Tsai Lin and Jen-Wen Ding Abstract—As different types of wireless networks are converging wireless environments for the following two reasons. First, wire- into an all-IP network, i.e., the Internet, it can be expected that less networks have very limited bandwidth compared to conven- in the near future video-on-demand (VoD) will be widely applied tional wired networks. Second, users in wireless environments to many interesting services, and users can access these services using heterogeneous terminals via heterogeneous wired/wireless usually employ heterogeneous terminals, which are quite dif- access networks. Many periodic broadcasting protocols have ferent in terms of their processing capability, storage space, and been proposed to reduce the implementation cost of VoD systems. power. However, most of the protocols assumed homogeneity for user In the past few years, much research has been devoted to terminals, while in practice, user terminals are usually quite the design of efficient VoD servers and VoD transmission different in their processing power, buffer space, and power. To address this problem, a few periodic broadcasting protocols protocols (or multicast protocols) [1]–[9], [11]–[15], [17]–[19], providing the same video quality for all heterogeneous clients [21]–[24]. Many studies showed that a well-designed VoD have been proposed recently. In this paper, we proposed a novel broadcasting protocol is a key technology that can effectively heterogeneous VoD broadcasting technique called Catch and Rest lower the very high server I/O bandwidth and network band- (CAR) to accommodate bandwidth heterogeneity without sacri- width requirements of a large-scale VoD system. Hu and Hua ficing user video quality. Then, we provide mathematic analysis to calculate the client bandwidth and buffer space requirements of et al. provide a comprehensive survey of VoD broadcasting CAR. Finally, we present our performance evaluation results for protocols in [13], [14]. These protocols allow a large number of CAR. Our results show that under the same system resources (i.e., clients to demand different videos at any time with low service server and network bandwidth), CAR provides more uniform and latency ranging from a few seconds to a few minutes depending acceptable service latency for all heterogeneous clients compared on the design of the system. The service latency, in this paper, to previous works. is defined to be the time between the demand for a video and Index Terms—Broadcast channels, catch and rest, multimedia the reception of the beginning of the video. communication, multimedia systems, periodic broadcasting, VoD broadcasting protocols can be classified into two quality-of-service (QoS), video-on-demand (VoD). categories: reactive and proactive [13], [14]. In the reactive approach, clients make requests to a video server and the server I. INTRODUCTION employs broadcast (or multicast) to service a batch of requests that demand the same video and arrives closely in time. Ex- ITH the rapid development of multimedia and commu- amples of this type of approach include first-come-first-serve Wnication technologies over the past decade, it is now (FCFS) batching, maximum-queue-length-first (MQLF) feasible to provide video-on-demand (VoD) services to a large batching [6] and maximum-factored-queued-length-first number of users using broadband wired networks. It is antici- (MFQLF) [1], and various types of patching [3], [4], [9]. In pated that in the near future, as different types of wireless net- the proactive approach, videos are divided into a series of works are converging into all IP networks, VoD will be widely segments and broadcast periodically on dedicated channels. applied to many services (e.g., home entertainment, distance The proactive approach is delicately designed such that all learning, digital libraries, staff training, on-line games) [16], clients arriving at different times can receive and watch videos [20], and users can access these services using different termi- continuously with low service latency. For a large-scale system, nals (e.g., PDA, smart-phone, set-top box, notebook, PC) via the reactive approach is far less efficient compared to the different access technologies (e.g., 2.5G/3G, WLAN/WMAN, proactive approach. The VoD service provided by broadcasting xDSL) [10]. Fig. 1 shows the general architecture for providing protocols is usually referred to as near VoD service since a VoD services to heterogeneous users in hybrid wired/wireless true VoD service does not require users to wait and supports networks. However, to date, we still face many problems to video-cassette-recorder-like (VCR-like) interactivities [13], achieve the above goal. One of the main difficulties is that con- [14]. ventional VoD delivery technology cannot be readily applied to Although many proactive video broadcasting protocols have been proposed, much research has assumed that all clients are homogeneous, i.e., all clients have the same reception capability Manuscript received September 14, 2005; revised March 16, 2006. C.-T. Lin is with the Department of Information Management, Kun Shan Uni- and storage space. As discussed earlier, in the future clients at versity, Tainan 710, Taiwan (e-mail: [email protected]). different geographical locations will access VoD services using J.-W. Ding is with the Department of Information Management, National heterogeneous terminals via various wireless/wired access Kaohsiung University of Applied Sciences, Kaohsiung 807, Taiwan (e-mail: [email protected]). network technologies. This implies that a VoD broadcasting Digital Object Identifier 10.1109/TBC.2006.879856 protocol should take this heterogeneity into account and support 0018-9316/$20.00 © 2006 IEEE LIN AND DING: LOW LATENCY VIDEO-ON-DEMAND BROADCASTING SCHEME FOR HETEROGENEOUS RECEIVERS 337 Fig. 1. General architecture for providing heterogeneous VoD services. receivers with different reception capabilities (ranging from required bandwidth and buffer space, respectively, for clients. very limited wireless bandwidth in a cafe to very high wired Section V describes the parameter settings of our simulation bandwidth at work) and different storage space (ranging from experiments and the performance evaluation results of CAR. a few mega-bytes to a few giga-bytes). However, little research Finally, Section VI concludes this paper. has been done on providing VoD services for heterogeneous receivers in hybrid wireless/wired networks. Two innovative II. RELATED WORK schemes, HeRO [2] and BroadCatch [19], have been proposed to address this problem recently. With both schemes, users can A. Overview of Periodic Video Broadcasting Schemes choose among a range of bandwidths to use to watch the video One way to broadcast popular video is to let multiple clients at the cost of their service latency, not the video quality. Specif- share the same set of channels. To reduce client access latency, ically, clients with high reception bandwidth watch videos with Staggered Broadcast [5] is the simplest broadcasting protocol low service latency, while clients with low reception bandwidth proposed in the early days. It periodically broadcasts the whole watch videos with long service latency. Although both schemes video in channels differentiated by time of the video. can provide relatively low service latency for high-end clients, Some more efficient broadcasting protocols have proposed. In- the service latency for low-end clients is quite high. In this stead of broadcasting the whole video, they broadcast a distinct paper, we propose a novel heterogeneous VoD broadcasting portion of the video periodically in each channel. In each por- scheme, termed CAR (catch-and-rest), to solve this problem. tion, the video may be further partitioned into segments that Compared to HeRO and BroadCatch, CAR greatly reduces are of fixed size or variant size depending on the broadcasting the service latency for low-end clients at the cost of slightly schedule. increasing the service latency for high-end clients. The reduced The efficiency of the proposed broadcasting schedule is service latency is one or two order of magnitude larger than always evaluated by the bandwidths that are required by the the increased service latency. This design tradeoff is significant servers and the clients. Early periodic broadcast approaches since it implies that under a limited broadcast bandwidth, it such as Pyramid Broadcast (PB) [23], Fast Broadcast [17], Har- is more likely that CAR can provide acceptable service la- monic Broadcast [15], New Pagoda Broadcast [18], Recursive tency guarantee for both high-end and low-end clients, while Frequency-Splitting [22], etc.\ aimed at reducing the server HeRO and BroadCatch can provide acceptable service latency bandwidth. However, these protocols demand that the clients guarantee for only high-end clients. In addition to service have the same bandwidth with the server and therefore result latency, CAR greatly reduces the maximum client buffer space in the requirement of high-cost clients. To minimize the client requirement as compared to BroadCatch, and has similar client bandwidth requirement, Skyscraper Broadcast [11], Client buffer space requirement as compared to HeRO. The feasibility Centric Approach (CCA) [12] and Generalized Fibonacci of CAR is proved theoretically. The performance of CAR is