Publications

Detailed Information

Fairness Improved Cross-Layer Adaptive Streaming over HTTP : 공정성을 개선한 HTTP 기반의 교차 계층 적응형 스트리밍 프레임워크

Cited 0 time in Web of Science Cited 0 time in Scopus
Authors

방용배

Advisor
권태경
Major
공과대학 컴퓨터공학부
Issue Date
2018-02
Publisher
서울대학교 대학원
Keywords
Adaptive Video StreamingDASHTCPCross-LayerQoEfairness
Description
학위논문 (석사)-- 서울대학교 대학원 : 공과대학 컴퓨터공학부, 2018. 2. 권태경.
Abstract
Due to the improvement and popularization of computers, and the development of Internet and networks, we have been producing and consuming video information which consists of audio and visual information from the Internet, where we only got text and picture-oriented information in the past. In this trend, video traffic quickly grew to take the majority of the Internet traffic. In this flow, researchers are actively studying on how to handle video traffic more efficiently.

Meanwhile, in addition to QoS, which is an objective indicator, QoE, which is a subjective indicator, is emerging as a performance index of video streaming. QoE is a quality of experience that a user uses and feels in the service. In video streaming, QoE can be adversely affected if the video is interrupted during playback and re-buffered or video quality is low. Therefore, researchers are studying in various directions to find out the metrics that can affect QoE and minimize them.

CLASH is an HTTP adaptive streaming (HAS) that adopt the cross-layer technique to mitigate the conflict point between the HAS layer and the TCP layer. In TCP, if there is only one packet loss, it is determined that network congestion has occurred, and the congestion window size is reduced. This results in a reduction in bandwidth, which can have a severe impact on the QoE of HAS. For HAS, the loss of one packet may not be such a big deal, and the cause of that it becomes fatal problem is that the TCP layer and the HAS layer operate independently of each other. To solve this problem, CLASH proposed a scheme to hide packet loss and confirmed that QoE is improved by using it.

On the other hand, CLASH has an important problem in fairness. TCP is promised to solve congestion by reducing congestion windows size when congestion occurs in the network. However, CLASH does not reduce the congestion window by itself, which causes the bottleneck point be overwhelmed. It is a severe problem which means that the scheme cannot be used widely.

In this paper, we introduced the concept of debt and divide the video streaming state into an emergency state and a non-emergence state in order to solve this problem. If a packet loss is detected in the emergency state, it increases one debt and hides the packet loss. Under the non-emergency state, packet loss may be detected and not be hidden. If it has debt instead, it will deliberately pay for its debt by causing packet loss. The emergency state and non-emergency state are based on the length of the playback buffer. A deterministic model with fixed boundaries and a probabilistic model to reduce sudden changes are proposed. In the former case, it is a model that manages the debt by dividing the playback buffer length into the emergency state when it is smaller than the fixed value, and a non-emergency state when it is bigger than the fixed value. In the latter case, it is a model that manages debt by making continuous probabilities that the probability of the bearing debt action is high when the playback buffer length is small and the probability of the paying debt action is high when the playback buffer length is long.

We experimented with the performance of all packet loss hiding techniques. Next, We looked at how certain values can best improve QoE and fairness by modifying the two major coefficients of the probabilistic model. The average bit-rate, total play time, average bandwidth, QoE, and fairness index were examined for several combinations of two coefficients of the probabilistic model. The value s showed a positive correlation with average bit-rate and fairness. Because all probabilistic models showed similar fairness indicators, only the QoE index ranking is used to find the best combinations of the coefficients. The best ones were chosen as the best combination and it is a = 0.7, s = 0.5.

The performance of our proposed schemes is compared with that of normal TCP and CLASH which hides all packet losses. As expected, hiding all packet losses showed overwhelming increases bandwidth and QoE, but the worst in terms of fairness. In the cases of the debt models proposed in this paper, they show similar fairness compared with the normal TCP. The deterministic model shows the effect of reducing buffering. On the other hand, the degree of buffering reduction of the probabilistic model was weak, but it is confirmed that this model improves QoE by reducing the image quality variation. When evaluated as a whole, the probabilistic model was found to be good both in terms of QoE and fairness. As a result of adopting the concept of debt, we can confirm that QoE is better than general TCP while ensuring fairness in CLASH. However, the overwhelming QoE, which was the strength of CLASH, proved to be very soft. This is thought to be due to the excessive repayment which is one packet loss for one packet loss hiding. Therefore, there is a need for more research on the method of returning the exact gain that can be obtained by concealing packet loss.
Language
English
URI
https://hdl.handle.net/10371/141553
Files in This Item:
Appears in Collections:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share