1 A Buffer-Aware HTTP Live Streaming Approach for SDN-Enabled 5G Wireless Networks Speaker: Bo-Yu Huang Advisor: Dr. Ho-Ting Wu Date: 2015/04/15 2 Outline Introduction Multimedia Streaming for SDN-enabled 5G Wireless Networks Buffer-Aware HTTP Live Streaming Mechanism Analysis of implementation results Conclusion Reference 3 Introduction 5G communication technology will be able to achieve some challenging requirements. Thus it makes the performance of traditional network architecture increasingly unable to keep up with 5G communication. Therefore, software-defined networks (SDNs) were regarded as a revolutionary technology to subvert the traditional networking industry. 4 Introduction(Cont.) Currently, people tend to watch movies, videos, and TV programs via streaming servers, resulting in the discussion of various subjects. Hence the adaptive streaming technologies were proposed in the past that can dynamically select video content based on current network conditions or mathematical capabilities of hardware in order to provide higher-quality streaming service to users. However, in heterogeneous networks some studies have attempted to improve the bandwidth prediction for effective transmission rates 5 Introduction(Cont.) Different from other studies on 5G networks, this study aims at a dynamic adjustment video streaming mechanism with HTTP live streaming protocol according to the utilization and stability of the routers and switches of SDN and the network condition of 5G, enhancing service quality. 6 Multimedia Streaming for SDN-enabled 5G Wireless Networks SDN presents a software layer to make the network device adjustments through SDN defined. Therefore, there are two planes in SDN network devices: the control plane and the data plane. Because of this separation, network administrators are no longer executing all the control rules on the physical network devices individually. 7 Multimedia Streaming for SDN-enabled 5G Wireless Networks(Cont.) However, multimedia streaming, which is one of the most bandwidthconsuming services, is an urgent challenge, and researchers have been designing new architectures and mechanisms for providing a multimedia streaming mechanism on SDN-enabled 5G wireless networks. That is the motivation for the buffer-aware HTTP live streaming approach proposed in this study for SDN-enabled 5G wireless networks, as shown in Fig. 1. 8 Buffer-Aware HTTP Live Streaming Mechanism This study proposes a dynamic adaptive streaming mechanism, based on quality of service (QoS) by researching the buffer status of UEs buffer, and aims to make a dynamic adjustment according to the 5G wireless communication network. The streaming mechanism, preloads media segments under quality permission, where better quality media segments are downloaded on the local side for buffering. This mode is called the buffering mode. If the network condition does not allow better buffering, the media segment quality is lower than the threshold and buffering is abandoned, and segments most suitable for the bandwidth of the time are adjusted and selected for real-time streaming according to the network prediction. This mode is called the real-time streaming mode. 9 Buffer-Aware HTTP Live Streaming Mechanism(Cont.) The initialization is executed after the UE starts, and the index file is analyzed in this stage, where the analyzed resolution and quantification parameters are entered to obtain the score value to determine the lowest-level media information that can be preloaded in the play list. This level is called the basic level. Afterward, the system enters the preloading stage, the basic level segment is preloaded for t seconds, and the upper limit of buffering for preloading is 6t second playing time. The size is obtained from 10 Buffering Mode This mode aims to preload media segments of better quality into the buffer of the local side so the user will not receive content of worse quality. There are two dynamic adaptation strategies in buffering mode: 1. scale up mode 2. scale down mode In order to effectively evaluate the heterogeneous network conditions and adjust the quality of the media segment to be buffered in this mode, this study proposes a combination of a buffer-based bandwidth forecasting method and level equation for the dynamic adaptive adjustment strategy in buffering mode. 11 Buffer Bandwidth Forecasting Method This method is designed by observing the change in the stream playing length stored in the buffer. This approach mainly detects the changes when there is large growth in the network condition. In terms of method design and implementation, the number of downloaded segments is recorded once per second, five consecutive data packets are multiplied by the segment playing length, and calculated by the least square method to determine slope m. 12 Scale Up Mode In scale up, when the buffer bandwidth forecasting method detects the network condition becoming increasingly better, the media segment of better quality is downloaded. The precondition of this strategy is that the available playing time exceeds 2t seconds. 13 Scale Down Mode The scale down mode is used when the network condition is poor and the balance between download and consumption has failed. In order to continuously provide the user with picture quality above the basic level, the level is corrected down in order to maintain buffering mode as possible When the average buffer time is shorter than the threshold, the downloaded media level is scaled down. The precondition of scale down is that the slope is less than zero. 14 Buffer-Aware HTTP Live Streaming Mechanism(Cont.) When scale up or scale down is executed, it will enter into the idle mode within five seconds, and without any adjustment. As the recording continues, the number of occupied buffers, or the downloaded quantity, is recorded once per second. 15 Level Equation The level equation aims to calculate the level difference between other representations recorded in index and the current streaming download. The difference refers to different bit rates. The influential factors discussed in this article are resolution and quantification parameters. According to : The two equations are combined to calculate the gap of the current bitrate, expressed as 16 Level Equation(Cont.) The Gt calculated by Eqs. 4 and 5 refers to the bit rate gap between two representations, and according to the bandwidth increasing trend, as calculated from the level difference and buffer bandwidth forecasted by scale up and scale down, the level equation of Eq. 6 is defined, The level equation determines integers between –2 and 4, where the level differences calculated by the aforesaid two equations may be equal to each other, meaning there may be multiple representations equal to the level difference of the media segment under streaming download. 17 Level Equation(Cont.) If the scale up or scale down mechanism determines the level, then the choice best meeting economic benefit is calculated, namely, the sum of the scores of the representation, as derived from the score graph, is divided by Gt, where a higher ratio means better quality can be obtained by lower bandwidth, expressed as 18 Level Equation(Cont.) The flow chart of buffering mode streaming mechanism in the system 19 Real-time Streaming Mode This streaming mode is designed for poor network conditions. In buffering mode, when the UE has exhausted the media segments temporarily stored in the local side, the system enters into streaming mode and abandons buffering. The downloaded quantity per second, as recorded in the scale down state, is used for polynomial regression. Five pieces of historical information are used for two-dimensional polynomial regression to calculate the estimated bandwidth at the next second. The polynomial regression is defined as: 20 Real-time Streaming Mode(Cont.) A two-dimensional polynomial regression curve is deduced from the collected data cluster, which is used for forecasting at the next time point, as shown in Fig. 4. 21 Buffer-Aware HTTP Live Streaming Mechanism(Cont.) After initializing the streaming service, the SDN controller directly presets the UE into buffering mode, and changes into real-time streaming mode when the number of buffered media segments is reduced to 0. When buffering mode changes into real-time streaming mode, the downloading task is directly abandoned, and the media segment closest to the bandwidth, as calculated by polynomial regression, is downloaded until the ratio of total playing time of downloaded basic level media segments to the download time is higher than 1. 22 Analysis of implementation results 23 Analysis of implementation results (Cont.) 24 Analysis of implementation results (Cont.) 25 Analysis of implementation results (Cont.) 26 Conclusion In this study, a buffer-aware HTTP live streaming approach for SDN-enable 5G wireless networks was proposed, and the two modes were adopted for controlling the streaming quality based on the router management of the SDN controller and the bandwidth prediction of 5G. It makes the most suitable quality for the network environment to be determined. Finally, the results prove that the mechanism could maintain a certain level of streaming quality for SDN-enabled 5G wireless networks and ensure smooth and complete streaming services. 27 Reference Chin-Feng Lai, Ren-Hung Hwang, Han-Chieh Chao, Mohammad Mehedi Hassan, Atif Alamri, “A buffer-aware HTTP live streaming approach for SDNenabled 5G wireless networks,” Network, IEEE, Vol. 29, no. 1, Jan.-Feb. 2015, pp. 49–55.