Joint Buffer Management and Scheduling for Wireless Video Streaming

Transcription

1 Joint Buffer Management and Scheduling for Wireless Video Streaming Günther Liebl 1, Hrvoje Jenkac 1, Thomas Stockhammer 2, and Christian Buchner 2 1 Lehrstuhl f. Nachrichtentechnik, TUM, D Munich, Germany liebl@tum.de 2 Nomor Research GmbH, D Bergen, Germany stockhammer@nomor.de Abstract. In this paper we revisit strategies for joint radio link buffer management and scheduling for wireless video streaming. Based on previous work [1], we search for an optimal combination of scheduler and drop strategy for different end to end streaming options. We will show that a performance gain vs. the two best drop strategies in [1], ie drop the HOL packet or drop the lowest priority packet starting from HOL, is possible: Provided that basic side-information on the video stream structure is available, a more sophisticated strategy removes packets from an HOL group of packets such that the temporal dependencies usually present in video streams are not violated. This advanced buffer management scheme yields significant improvements for almost all investigated scheduling algorithms and streaming options. In addition, we will demonstrate the importance of fairness among users when selecting a suitable scheduler. 1 Introduction Optimization and adaptation of video streaming strategies to both wired and wireless clients, eg for High Speed Downlink Packet Access (HSDPA), has become a challenging task. The heterogeneous network structure results in a number of conflicting issues: On the one hand, significant performance gains for video transmission over wireless channels can be achieved by appropriate adaptation. On the other hand, optimization of the media parameters or streaming server transmission strategies exclusively to wireless links will result in suboptimal performance for a wired transmission and vice versa. Hence, cross-layer design of the following components is required: Streaming server, wireless streaming client, media coding, intermediate buffering, channel resource allocation and scheduling, receiver buffering, admission control, media playout, error concealment, etc. Since the search for an optimal joint set of all parameters is usually not feasible, suboptimal solutions have to be considered, which yield sufficient performance gains by jointly optimizing a subset of the above parameters. In this work we focus once again on strategies for joint radio link buffer management and scheduling for incoming IP based multimedia streams at the P. Lorenz and P. Dini (Eds.): ICN 05, LNCS 34, pp , 05. c Springer-Verlag Berlin Heidelberg 05

2 Joint Buffer Management and Scheduling for Wireless Video Streaming 883 radio link layer. Based on the wireless shared channel scenario in [1], we search for an optimal combination of scheduler and drop strategy for different end to end streaming options. In addition to the previously proposed drop strategies at the radio link buffers, we will investigate the gains achievable by incorporating basic side-information on the structure of the video stream. Our advanced drop strategy removes elements from an HOL group of packets such that the temporal dependencies usually present in video streams are not violated. We will assess the performance gain of this new scheme for an HSDPA scenario, and we will demonstrate the importance of fairness among users when selecting a scheduler. 2 Preliminaries for Wireless Video Streaming 2.1 End to End Streaming System As stated in [1], assume that the media server stores a packet stream, defined by a sequence of packets called data units, ie P = P 1, P 2,... Each data unit P n has a certain size r n in bits and an assigned Decoding Time Stamp (DTS) t DTS,n indicating when this data unit must be decoded relative to t DTS,1. After the server has received a request from a client it starts transmitting the first data unit P 1 at time instant t s,1 and continues with the following data units P n at time instants t s,n. Data unit P n is completely received at the far end at t r,n and the interval δ n t r,n t s,n is called the channel delay (we assume that data units are either received correctly or lost in the network due to bit errors or congestion). The received data unit P n is kept in the receiver buffer until it is forwarded to the video decoder at decoding time t d,n. Without loss of generality we assume that t DTS,1 = t d,1. Neglecting for now the session setup phase, the initial delay is defined as δ init t d,1 t s,1. Then, data units which fulfill t s,n +δ n t DTS,n can be decoded in time. Small variations in the channel delay can be compensated for by this receiver-side buffer, but long term variances result in loss of data units. Several advanced streaming techniques have been proposed in the literature to cope with this late-loss [2]. However, most streaming systems available on the market do not apply any of them yet. Hence, we will not consider them here, but we note that their use is feasible in our framework and is currently investigated. 2.2 Source Abstraction for Streaming According to [1], the video encoder Q e maps the video signal s = {s 1,...,s N } onto a packet stream P Q e (s). We assume a one to one mapping between source units s n,(ie video frames) and data units (ie packets). Encoding and decoding of s n with a specific video coder Q results in a reconstruction quality Q n q(s n, Q(s n )), where q(s, ^s) measures the rewards/costs when representing s by ^s. We restrict ourselves in the following to the Peak Signal to Noise Ratio (PSNR), as it is accepted as a good measure to estimate video performance. According to [3], the result of the encoding is a set of data units for the presentation which can be represented as a directed acyclic graph. If such a set is received by the client, only those data units whose ancestors have all

3 884 G. Liebl et al. been also received can be decoded. In case of a lost data unit, the corresponding source unit is represented by the timely nearest received and reconstructed source unit (ie a direct or indirect ancestor). If there is no preceding source unit, eg I frames, the lost source unit is concealed with a standard representation, eg a grey image. In case of consecutive data unit loss, the concealment is applied recursively. The concealment quality Q n,ν (i), ifs n is represented with s i, is defined as Q n (i) q(s n, Q(s i )). We express the importance of each data unit P n as the increase in quality at the receiver if s n is correctly decoded, ie I n 1 Q n N Q n (c (n)) + N i=n+1 n i [ Qi (n) Q ] i (c (n)), (1) with c (n) the number of the concealing source unit for s n, and n i indicating that i depends on n. Additionally, Q n (0) indicates concealment with a standard representation. The overall quality for a sequence of length N is then Q = 1 N N Q n = Q 0 + n=1 N I n, (2) with Q 0 the minimum quality, if all frames are presented as grey. Hence, quality is incrementally additive w.r.t. to the partial order in the dependency graph. n=1 2.3 Streaming Parameters and Performance Criteria The video decoder might experience the absence of certain data units P n due to loss related to bit errors or congestion in the network (δ n = ), late-loss at the client (δ n > t DTS,n t s,n ), or the server not even having attempted to transmit the unit. Whereas the former two reasons mainly depend on the channel, the latter can be viewed as temporal scalability and a simple means for offline rate control and is not used here. Another important parameter in our end to end streaming system is the initial delay at the client. On the one hand, this value should be kept as low as possible to avoid annoying startup delay to the end user. On the other hand, longer initial delay can compensate for larger variations in the channel delay and reduce late-loss. Since we have ruled out more advanced streaming strategies, the single-user performance can be determined using a sequence of channel delays δ = {δ 1,...,δ N } for each data unit and a predefined initial delay δ init as (1 {A} equals 1 if A is true and 0 otherwise) Q(δ, δ init )=Q 0 + N I n 1 {δ n δ init } n=1 n 1 m=1 m n 2.4 Streaming in a Wireless Multiuser Environment 1 {δ m δ init }. (3) We assume that M users in the serving area of a base station in a mobile system have requested to stream multimedia data from one or several streaming servers.

4 Joint Buffer Management and Scheduling for Wireless Video Streaming 885 We assume that the core network is over provisioned such that congestion is not an issue on the backbone. The streaming server forwards the packets directly into the radio link buffers, where packets are kept until they are transmitted over a shared wireless link to the media clients. A scheduler then decides which users can access the wireless system resources bandwidth and transmit power, and a resource allocation unit integrated in the scheduler assigns these resources appropriately. Obviously, for the same resources available different users can transmit a different amount of data, eg a user close to the base station can use a coding and modulation scheme which achieves a higher bit rate than one at the boundary of the serving area. In general, the performance of the streaming system should significantly depend on many parameters such as the buffer management, the scheduling algorithm, the resource allocation, the bandwidth and power share, the number of users, etc. As done in [1], we have concentrated on the first two aspects in our investigations. Hence, the performance criterion for the single user system has been extended by averaging (3) over all users, ie Q(M, δ init )= 1 M M Q(δ m,δ init ). (4) m=1 3 Scheduling and Buffer Management Strategies 3.1 Scheduling Several scheduling algorithms for wireless multiuser systems have already been proposed in the literature [5, 6, 7, 8]. We will briefly characterize them here: 1. Basic scheduling strategies: Well known wireless and fixed network scheduling algorithms include, for example, the Round Robin scheduler, which serves users cyclically without taking into account any additional information. 2. Channel State Dependent Schedulers: The simplest, but also most appealing idea for wireless shared channels in contrast to fixed network schedulers is the exploitation of the channel state of individual users. Obviously, if the flow of the user with the best receiving conditions is selected at any time instant, the overall system throughput is maximized. This scheduler is therefore referred to as Maximum Throughput (MT) scheduler and may be the most appropriate if throughput is the measure of interest. However, as users with bad receiving conditions are blocked, some fairness is often required in the system. For example, the Proportional Fair policy schedules the user with the currently highest ratio of actual to average throughput. 3. Queue Dependent Schedulers: The previously presented algorithms do not consider the buffer fullness at the entrance of the wireless system except that flows without any data to be transmitted are excluded from the scheduling process. Queue dependent schedulers take into account this information, eg the Maximum Queue (MQ) scheduler selects the flow whose Head-Of-Line packet in the queue currently has the largest waiting time.

5 886 G. Liebl et al. 4. Hybrid Scheduling Policies: It might be beneficial to take into account both criteria, the channel state information and the queue information, in the scheduling algorithm. In [8] hybrid algorithms have been proposed under the acronyms Modified Largest Weighted Delay First (MLWDF) and Exponential Rule, which yield the most promising results among the standard solutions. 3.2 Radio Link Buffer Management For better insight into the problem, we assume that a single radio link buffer can store N data units, independent of their size. If the radio link buffers are not emptied fast enough because the channel is too bad and/or too many streams are competing for the common resources, the wireless system approaches or even exceeds its capacity. When the buffer fill level of individual streams approaches the buffer size N, data units in the queue have to be dropped. We will discuss several possible buffer management strategies in the following: 1. Infinite Buffer Size (IBS): Each radio link buffer has infinite buffer size N =, which guarantees that the entire stream can be stored in this buffer. No packets are dropped resulting in only delayed data units at the client. 2. Drop New Arrivals (DNA): Only N packets are stored in the radio link buffer. In case of a full queue, new packets are dropped, which is the standard procedure applied in a variety of elements in a wired network, eg routers. 3. Drop Random Packet (DRP): Similar to DNA, but instead of dropping the newly arrived packet we randomly pick a packet in the queue to be dropped. This strategy is somewhat uncommon, but we have included it here since all other possibilities are only specific deterministic variants of it. 4. Drop HOL Packet (DHP): Same as DNA, but here we drop the Head Of Line (HOL) packet, ie the packet which resides longest in the buffer. This is motivated by the fact that streaming media packets usually have a deadline associated with them. Hence, to avoid inefficient use of channel resources for packets that are subject to late-loss at the client anyway, we drop the packet with highest probability of deadline violation. 5. Drop Priority Based (DPB): Assuming that each data unit has assigned a priority information, we drop the one with the lowest priority which resides longest in the buffer. Our motivation here is the fact that sophisticated media codecs, like H.264/AVC [4], provide options to indicate the importance of elements in a media stream on a very coarse scale. Hence, those packets are removed first which do not affect the end-to-end quality significantly. 6. Drop Dependency Based (DDB): We propose the following new strategy to avoid the major drawback of both DHP and DPB: Starting from the HOL of the buffer when dropping medium to high priority packets is suboptimal due to the temporal dependency within the media stream. For example, if I- or P-frames in a Group of Pictures (GOP) are removed, all the remaining frames in the GOP cannot be reconstructed at the media decoder. Thus, leaving them in the buffer and transmitting them leads to inefficient utilization of the scarce wireless resources. Provided that basic side-information on the

6 Joint Buffer Management and Scheduling for Wireless Video Streaming 887 structure of the media stream is available to the buffer management (eg the GOP structure and its relation to packet priorities), an optimized strategy operates on the HOL group of packets with interdependencies: While all low priority packets can be deleted starting from the beginning of the HOL group, any medium or high priority packets should be first removed from the end of the HOL group to avoid broken dependencies. Since the structure of the media stream is usually fixed during one session, the buffer management only has to determine this information once during the setup procedure, which we believe to be feasible at least in future releases of wireless systems. 3.3 Streaming Server Rate Control As in [1], we only consider two basic streaming server rate control models: 1. Timestamp Based Streaming: In case of TBS the data units P n are transmitted exactly at sending time t s,n. If the radio link buffer is emptied faster than new data units arrive, then it possibly underruns. In this case, this flow is temporarily excluded from the scheduling. 2. Ahead of Time Streaming: In case of ATS, the streaming server is notified that the radio link buffer can still accept packets. Hence, the streaming server forwards data units to the radio link buffer even before their nominal sending time t s,n such that the radio link buffer never underruns and all flows are always considered by the scheduler. However, the streaming server eventually has to forward a single data unit no later than at t s,n regardless of the fill level notification. Thus, a drop strategy at the radio link buffer is still required. Note that this mode requires pre-recorded streams at the server and a sufficiently large decoder buffer at the media client. 4 Experimental Results 4.1 Definition of Test Scenario We have used the same HSDPA scenario as in [1] for our performance evaluations, which consists of one serving base station and 8 tiers of interfering base stations. We omit the detailed system parameter settings here and refer the interested reader to the above publication. A total of M = 10 randomly distributed users are attached to the serving base station and each of them has requested a streaming service for the same H.264/AVC coded QCIF sequence of length N = 2698 frames (looped six times) as in [1], with QP = 28, fps, and no rate control. The GOP structure is IBBPBBP..., with an I frame distance of 1 s. The PSNR results in Q(N) =36.98 db, and the average bit rate is kbit/s. We have evaluated the performance in terms of average PSNR Q(M = 10, δ init ) vs. initial delay δ init for selected parameter settings. All presented results with limited buffer size assume a restriction of N = packets (ie 1 second of video).

7 888 G. Liebl et al. Maximum Throughput (MT) Scheduling, N=, TBS Maximum Throughput (MT) Scheduling, N=, ATS a) b) IBS DNA DRP DHP DPB DDB IBS DNA DRP DHP DPB DDB MLWDF Scheduling, N=, TBS MLWDF Scheduling, N=, ATS c) IBS DNA DRP DHP DPB DDB d) IBS DNA DRP DHP DPB DDB Fig. 1. Average PSNR versus initial delay for different buffer management strategies under MT (a,b) and MLWDF (c,d) scheduling 4.2 Buffer Management Performance for MT and MLWDF Policy In Fig. 1a,b we compare all the different buffer management strategies under MT scheduling for both TBS and ATS. Regardless of the drop and streaming strategy, the system performance increases with larger initial delay as the probability of late-loss decreases. As the system is overloaded (about %), in case of IBS the fullness of the radio link buffers increases over the length of the streams. Since no dropping is performed, users with bad channel conditions experience significant HOL blocking and excessive initial delay at the media client is required for sufficient performance for both TBS and ATS. Nevertheless, due to the maximum throughput scheduler at least some users, namely those close to the base station, are served with good quality, but the worse users experience too high channel delays for this setup. This fact is especially evident in Fig. 1b, where due to the persistent occupation of the channel by the good users (who always have data to be sent in case of ATS) the quality for very short initial playout delay is already quite high, but then only increases very slowly. Hence, for improving the overall system performance it is beneficial to drop data units already at the radio link buffers (irrespective of the strategy) to reduce the excess load at the air interface and convert late-loss into controlled packet removals, ie achieve an in time delivery of a temporally scaled version of the video stream. The simplest strategy DNA shows some gain, but is worse than dropping packets randomly. The by far best performance for both TBS and ATS is obtained by applying our newly proposed DDB algorithm: While DHP and DPB intersect at relatively

8 Joint Buffer Management and Scheduling for Wireless Video Streaming 889 Different Schedulers with Optimal Dropping, N=, TBS Different Schedulers with Optimal Dropping, N=, ATS a) b) Round Robin, DDB MT, DDB PropFair, DDB MQ, DHP MLWDF, DDB ExpRule, DDB Round Robin, DDB MT, DDB PropFair, DDB MQ, DDB MLWDF, DDB ExpRule, DDB Different Schedulers with Optimal Dropping, N=, TBS Different Schedulers with Optimal Dropping, N=, ATS c) MT, DDB, User 5 (best) MLWDF, DDB, User 5 (best) MT, DDB, User 1 (average) MLWDF, DDB, User 1 (average) MT, DDB, User 6 (worst) MLWDF, DDB, User 6 (worst) d) MT, DDB, User 5 (best) MLWDF, DDB, User 5 (best) MT, DDB, User 1 (average) MLWDF, DDB, User 1 (average) MT, DDB, User 6 (worst) MLWDF, DDB, User 6 (worst) Fig. 2. a),b) Average PSNR versus initial delay for different schedulers and optimal buffer management strategy. c),d) Results for the best, worst, and average user large initial delay, the curve for DDB shows good performance over the whole range of delay values, especially in case of ATS. Furthermore, an interesting observation can be made for all drop strategies and TBS: If the radio link buffer size is larger than the initial delay, an almost linear gain can be achieved by increasing the latter. However, soon after the initial delay matches the radio link buffer size, the PSNR curve runs into saturation, since the majority of packets is now dropped due to the finite buffer size and not due to deadline expiration. If we evaluate the performance of our drop strategies for a hybrid scheduler, like MLWDF, which has been designed to account for some fair trade-off between channel and queue state, we observe that IBS should never be used: Both Fig. 1c,d show disastrous consequences for the average PSNR. Hence, considering the HOL delay in the scheduling metric leads to large performance degradations for all users, if the amount of HOL blocking is not limited. On the other hand, if the radio link buffer size is limited, applying our new DDB algorithm either yields close to optimum (TBS) or strictly optimum (ATS) performance. 4.3 Scheduler Comparison and Fairness Figures 2a,b contain the average PSNR for six different combinations of scheduler and optimal drop strategy under TBS and ATS. Albeit for the MQ scheduler with TBS, optimal buffer management is achieved by our new DDB algorithm. The question which scheduler to use, however, is not as simple, but largely depends on

9 890 G. Liebl et al. the initial playout delay: For very short values, the MT scheduler performs best, while all other schedulers with queue-based metrics are not very efficient. For reasonable initial playout delays larger than one second, the MLWDF scheduler would be the better choice. However, the type of scheduler has to be chosen upon system startup without knowing individual initial playout delays. Therefore, the fairness among the users in the system has to be considered by looking at the PSNR of the best, worst, and average user depicted in Fig. 2c,d for both MT and MLWDF. While MT favors both the best and the medium user by suppressing the bad user significantly, MLWDF tries to achieve a trade-off between maximum throughput of the system and fairness. Hence, the best and medium user quality is slightly reduced, while the system tries to supply the bad user with sufficient quality as well. This is especially evident for ATS, where the gain of the bad user is large compared to the decrease of the other two. Obviously, for reasonable initial playout delays, this support of the bad users also results in an increase in average quality over all users, since more of them contribute to it. 5 Conclusion In this paper we have revisited strategies for joint radio link buffer management and scheduling for video streaming over wireless shared channels. As a straightforward extension to previous work we have proposed a more sophisticated drop strategy at the radio link buffer that incorporates side-information on the temporal dependency structure in typical video streams. Albeit for one combination of scheduler and streaming mode, our newly proposed algorithm provides optimal performance over the whole range of (unknown) initial playout delays. Since the side-information only has to be determined once during the setup phase, we consider it to be feasible within future releases of wireless systems like HSDPA. Furthermore, our investigations have gained us some valuable insight into the applicability of certain types of schedulers for wireless video streaming: In particular, we showed that the combination of a queue-state dependent scheduler with infinite radio link buffer size leads to disastrous results for all users. However, DDB combined with a hybrid scheduler seems to yield a good trade-off between average quality of all users and fairness among individual users. Since including side-information on the video stream in the buffer management has proven to be successful, making part of it also available to the scheduler is the subject of ongoing research at our institute. First results already show that priority and/or deadline based scheduling policies yield significant improvements. References 1. G. Liebl, H. Jenkac, T. Stockhammer, and C. Buchner, Radio Link Buffer Management and Scheduling for Video Streaming over Wireless Shared Channels, in Proc. Packet Video Workshop 04, Irvine, CA, USA, Dec. 04.

10 Joint Buffer Management and Scheduling for Wireless Video Streaming B. Girod, M. Kalman, Y. J. Liang, and R. Zhang, Advances in video channel adaptive streaming, in Proc. IEEE Int. Conf. on Image Processing, Rochester(NY), USA, Sept P. A. Chou and Z. Miao, Rate-distortion optimized streaming of packetized media, 01, submitted. pachou. 4. Advanced Video Coding for Generic Audiovisual Services, ITU T and ISO/IEC JTC 1, M. Andrews, K. Kumaran, K. Ramanan, A. Stolyar, and P. Whiting, Providing Quality of Service over a Shared Wireless Link, IEEE Communications Magazine, vol. 39, pp. 0 4, February H. Fattah and C. Leung, An Overview of Scheduling Algorithms in Wireless Multimedia Networks, IEEE Trans. on Wireless Communications, October S. H. Kang and A. Zakhor, Packet Scheduling Algorithm for Wireless Video Streaming, in Proc. Packet Video Workshop 02, Pittsburgh, USA, April S. Shakkottai and A. L. Stolyar, Scheduling Algorithms for a Mixture of Real- Time and Non-Real-Time Data in HDR, in Proceedings of the 17th International Teletraffic Congress (ITC-17), Salvador, Brazil, September R. S. Tupelly, J. Zhang, and E. K. Chong, Opportunistic scheduling for streaming video in wireless networks, in Proc. 37th Annual Conference on Information Science and Systems, Baltimore, MD, USA, Mar. 03.