A novel adaptive logic for dynamic adaptive streaming over HTTP

Highlights

• Throughput estimation method based on the throughput observed over past segments.

• Rate adaptive algorithm based on the buffer occupancy and estimated throughput.

• Selects high-quality video segments, while minimizing video quality changes.

• Minimizes the risk of playback interruption.

• Competing HTTP clients achieve equitable video rates.

Abstract

In this paper, we propose an estimation method that estimates the throughput of upcoming video segments based on variations in the network throughput observed during the download of previous video segments. Then, we propose a rate-adaptive algorithm for Hypertext Transfer Protocol (HTTP) streaming. The proposed algorithm selects the quality of the video based on the estimated throughput and playback buffer occupancy. The proposed method selects high-quality video segments, while minimizing video quality changes and the risk of playback interruption, improving user’s experience. We evaluate the algorithm for single- and multi-user environments and demonstrate that it performs remarkably well under varying network conditions. Furthermore, we determine that it efficiently utilizes network resources to achieve a high video rate; competing HTTP clients achieve equitable video rates. We also confirm that variations in the playback buffer size and segment duration do not affect the performance of the proposed algorithm.

Introduction

According to [1], 75 percent of global mobile data traffic will be mobile video traffic by 2019. Hypertext transfer protocol (HTTP) streaming video services have become a cost-effective means for transmitting multimedia content. This trend is expected to continue as the Internet infrastructure has evolved to support HTTP, and HTTP is firewall friendly. Additionally, an HTTP server does not have to maintain a session state; therefore, a large number of streaming clients can be supported [2].

HTTP-based video streaming solutions provide multiple representations (e.g., different bitrate/quality) of the same content, and divide those representations into small segments. The content is stored at the server, and the rate-adaptive algorithm at the client decides which segment to download next.

The estimation of throughput is important in the selection of the next segment. The segment throughput is calculated as the ratio of the segment size divided by the time it takes to download that segment. Put simply, measured segment throughput can be used as the throughput estimate for the next segment [3]. However, due to short term-fluctuations, a throughput estimate calculated in this manner will result in a high frequency of fluctuations. Ran et al. [4] use the median of the throughput of the last several segments to estimate the throughput of the next segment. Rahman et al. [5] show that the McGinely dynamic indicator offers a stable response to throughput fluctuations while maintaining a stable playback buffer. The moving average technique [5] is accurate for slow throughput variations, but reacts slowly to sudden variations in the throughput.

HTTP clients make an estimate of future throughput from past observations to select the video rate for the next segment [3], [6], [7]. An accurate estimate of the throughput becomes an important challenge for the client. An inaccurate estimation may lead to video bit rates that degrade the user’s experience. Many rate adaptive algorithms add playback buffer occupancy as an adjustment parameter in addition to throughput estimation [5], [8], [9], [10], [11], [12].

The algorithms attempt to maximize the quality of the video by meeting conflicting objectives in a manner that improves the user’s viewing experience. Some of the potential objectives include selecting the highest feasible set of video bit rates, avoiding needless video bitrate changes, and preserving the buffer level to avoid an interruption in the playback [13], [14], [15], [16], [17]. The rate adaptation algorithms work fairly well when a client operates alone. When the client buffer is full, the client enters a periodic On-Off phase [18]. During the On-Off phase, the competing clients result in bandwidth underutilization and unfair bandwidth sharing [18], [19], [20]. In [18], the authors observed unfair bandwidth sharing among three Microsoft Smooth Streaming clients. This behavior is observed in the presence of the competing TCP as well as other competing HTTP clients. In [21], the authors suggest that in the presence of competing HTTP clients, the rate-adaptive algorithm selects variable and low-quality video, which is undesirable to the users. In this respect, an adaptive video algorithm must strive to choose the highest feasible video rates to maximize the user’s experience [20]. In addition, the competing clients should be able to achieve comparable video rates while striving to select the highest feasible video rates.

In this paper, we first propose an estimation algorithm that accurately estimates the throughput based on variations in the network throughput. It offers a stable response to small and short term fluctuations and is sensitive to persistent and large fluctuations. We then propose a rate adaptation algorithm that dynamically selects the video bit rates based on the estimated throughput and the playback buffer occupancy. The objective of the proposed algorithm is to improve the user’s viewing experience. We adopt three rate-adaptive algorithms as benchmarks. In addition to the rate-adaptive algorithm proposed in our previous work [5], we adopt the algorithms proposed in [8], [9], as benchmarks to demonstrate the efficiency of our proposal. In the results, we refer to the algorithms proposed in [5], [8], [9] as BBA, AAA, and SARA, respectively. We analyze the algorithms under varying network conditions, buffer sizes, and segment durations. We evaluate the proposed algorithm for single and multi-client scenarios. Our key results can be summarized as follows:

• The proposed algorithm streams high-quality video and avoids unnecessary video rate changes while avoiding playback interruption.

• In the single client scenario, the proposed rate adaptive algorithm achieves average video rate better than BBA, AA and SARA up to 58%, 37.5% and 7% respectively. The proposed algorithm achieves comparable video rate changes compared to BBA and AAA and achieves up to 6.5 times less number of video rate changes compared to SARA.

• Furthermore, in a multi-client environment, we show that the proposed scheme efficiently utilizes network resources to stream high-quality video and that the HTTP clients achieve equitable video rates.

• We perform experiments to show that irrespective of the buffer size and segment duration, the proposed algorithm improves the user’s experience.

It is important to note that this work does not attempt to reduce the extent of multi-client problems. Its objective is to show that the proposed algorithm performs better than other state-of-the-art algorithms in a multi-client environment.

The remainder of this paper is organized as follows: Section 2 offers an overview of HTTP adaptive streaming, and reviews existing video streaming algorithms. Section 3 presents the proposed throughput estimation algorithm. Section 4 explains our throughput- and buffer-based rate adaptive algorithm. Section 5 provides the simulation results. Finally, Section 6 concludes the paper.