Data-Driven QoE Measurement
Recently, the strict definition of data-driven QoE measurement has not been given. The International Telecommunication Union (ITU-T) has defined the QoE concept as the entire thing of availability of services subjectively perceived by end users. The definition of QoE by the European Qualinet is the degree of satisfaction or annoyance of the end users of services because the utility and/or the expectations regarding services are based on end-user attitudes and current service situations (He et al., 2018). Furthermore, data-driven QoE measurement can be defined from the perspective of objective-based or subjective-based metrics. Generally speaking, data-driven QoE measurement is the end-user-centric metric that is used to measure end-user acceptance or satisfaction of data services or applications. In addition, data-driven QoE measurement is the extended version of data-driven QoS measurement used for judging delay-sensitive data from the end-user perspective (Wang et al., 2018). In summary, the common understanding of data-driven QoE measurement is as follows: data-driven QoE measurement is a new measurement for data services which is based on the interactions between end users and services.
The data-driven QoE concept is a well-known measurement mechanism for determining the overall perception of the data-driven QoS, i.e., the evaluation of data-driven QoS as experienced by end users. Recently, many data-driven QoS-based methods have been developed to optimize the efficiency and performance of the environment, as proposed by the authors in Fang et al. (2016). Even though the parameters of data-driven QoS offer good objective measurement criteria, they cannot directly determine the quality of end-user perceptions. Therefore, both academic and industry researchers have shifted their attention from data-driven QoS parameters like jitter, throughput, packet loss, and delay to the concept of data-driven QoE. Data-driven QoE, in contrast, can refer to both the performance and efficiency of services as measured by data-driven QoS, as well as the subjective opinions of end users. Therefore, data-driven QoE is more suitable with respect to end users in a form of measurement than is data-driven QoS.
In fact, data-driven QoE analysis is paid close attention to the data-driven QoS metrics, i.e., bitrate, rebuffering, and delay. Data-driven QoE analysis models have appeared of late due to availability of large-scale data accompanied by the promising popularity of traffic and/or streaming around the Internet. In particular, video streaming/traffic has the availability of large-scale data in order to analyze QoE around the Internet (Wang et al., 2019). Later on, some extensions to the data-driven QoE analysis model have been studied. QoE analysis model with linear regression and correlation has been proposed in Du et al. (2018). The Kendall correlation coefficient can be selected to calculate the correlations between each metrics of data-driven QoS and data-driven QoE. Then the collection of information can help to get a deeper understanding of the relationship between data-driven QoS and QoE via refining how the intellectual of uncertainty of the data-driven QoE metrics increases a certain data-driven QoS metric. Finally, linear regression-based curve fitting is the application to the data-driven pairs of QoS-QoE. By observing the data-driven QoS-QoE curves, the relationship is not linear in the whole environments. Additionally, a decision tree-based QoE prediction model is developed in Wang et al. (2016) to eliminate defects of correlation and linear regression analysis.
Also, data-driven QoE management has been studied. For example, a data-driven management architecture for personalized QoE is proposed to assess QoE from user perspective involving the subjective characteristics of end users (Wang et al., 2016). Then, a reinforcement learning-based method has been used for QoE-based spectrum handoff management (Xu et al., 2018). Data-aware QoE-QoS management method based on principle component analysis proposed to balance QoE-QoS management (Usman et al., 2018).
Lastly, several data-driven QoE manners, e.g., evaluation, and visualization have been proposed to enable data-driven QoE optimization in Jiang et al. (2017). Content providers can get amazing profits from data-driven platform via transferring strategies of load balance based on QoE server statement visualization and monitoring (Wang et al., 2016).
The common understanding of data-driven QoE is a novel measurement technique for applications or services and is confirmed based on the quality of whole environments and the experience of end users. Data-driven QoE measurement has been applied to a variety of scenarios.
Design of media player buffer: The design of media player buffer is crucial adjective because the event of rebuffering has a vital impact on QoE of end users. The size of buffer will affect the delay and time of rebuffering. The delay will be longer since more and more data can be downloaded before the player plays if the buffer size is large. During the playing state, vice versa, however, fewer events of rebuffering can happen. Additionally, He et al. (2018) has been shown that most of the downloaded data in the buffer is not useful due to many end users quitting before the completions of video.
Congestion control: Conventional TCP protocol applied to video traffic can result in long delay owing to the following reasons: (1) considering the TCP protocol, a lost packet will be retransferred when it is received successfully, leading to a long delay and get a poor QoE; (2) the additive increase multiplicative decrease (AIMD) algorithm results in fluctuated throughput around the time, which will increase the delay and result in end-user dissatisfaction; and (3) the congestion control is QoS-based, whereas the video is QoE-based. Sterca et al. (2016) proposed the Media-TCP to design a mechanism of congestion control video-friendly for the TCP protocol, optimizing the congestion window size, which maximize the long-term QoE. The distortion delay and influence deadline of each packet are considered to offer different services or applications for different packet classes. Media-TCP is shown to improve the PSNR around methods of conventional TCP congestion control. Unfortunately, Wang et al. (2017b) proposed Media-TCP, which is still QoS-based, a MOS-based congestion control for multimedia transmission. The value of MOS is predicted in real-time manner via the Lync system of Microsoft on the basis of quantitative measurements (i.e., packet loss, bit errors, packet delay, and jitter). Then QoE-based adaption of congestion window is formulated as a partially observable Markov decision process (POMDP) and is solved by the algorithm of online learning. Another method is the use of protocol such as video-friendly application, Dynamic Adaptive Streaming over HTTP (DASH), mitigating the delay problem in video transmission and not changing the TCP protocol.
Recently, applications of VoIP (i.e., Skype and Hangouts) are useful for managed relays, optimizing performances between network clients. In contrast to any cast-based relay selection, recent work has been shown that a data-driven algorithm of relay selection can reduce the number of poor-quality calls (i.e., 42 percent fewer calls with over 1.2 percent packet loss) (Canbaz et al., 2017). Thereby, the satisfaction of QoE can be given by the proposed data-driven algorithm in the Internet telephony.
The flexibility of file-sharing applications (e.g., Dropbox) allows each client to catch files from a chosen data center or server. We can enhance the QoE for these services or applications potentially by the use of data-driven methods to predict the throughput between a server and client (Chen et al., 2015).
With the applications of social network, we will attempt to optimize server selection. The goal is to get information of the co-location. Recent work has shown that optimal caching and server selection can reduce query time of Facebook to 50 percent (Du et al., 2018). Online services (e.g., search) can also benefit from data-driven techniques. For example, compared to anycast, edge sever selection of data can reduce search latency by 60 percent serving 2× more queries. Therefore, using data-driven technology in Web services is beneficial to the reduction of latency and, furthermore, brings better QoE performance.
- Canbaz M, Thom J, Gunes MH (2017) Comparative analysis of internet topology data sets. In: Proceedings of 2017 IEEE conference on computer communications workshops (INFOCOM WKSHPS)Google Scholar
- Du M, Wang K, Xia Z, Zhang Y (2018) Differential privacy preserving of training model in wireless big data with edge computing. IEEE Trans Big Data. https://doi.org/10.1109/TBDATA.2018.2829886
- He X, Wang K, Huang H, Miyazaki T, Wang Y, Guo S (2018) Green resource allocation based on deep reinforcement learning in content-centric IoT. IEEE Trans Emerg Top Comput. https://doi.org/10.1109/TETC.2018.2805718
- He X, Wang K, Huang H, Miyazaki T, Wang Y, Sun Y (2018) QoE-driven joint resource allocation for content delivery in fog computing environment. In Proceedings of ICC.Google Scholar
- Jiang J, Sekar V, Milner H, Shepherd D, Stoica I, Zhang H (2016a) CFA: a practical prediction system for video QoE optimization. In: Proceedings of 13th USENIX symposium on networked systems design and implementation (NSDI’16)Google Scholar
- Jiang J, Sun S, Sekar V, Zhang H (2017) Pytheas: enabling data-driven quality of experience optimization using group-based exploration-exploitation. In: Proceedings of 14th USENIX symposium on networked systems design and implementation (NSDI’17)Google Scholar
- Purohit L, Kumar S (2018) A classification based web service selection approach. IEEE Trans Serv Comput. https://doi.org/10.1109/TSC.2018.2805352
- Usman M, Yang N, Jan M, He X, Xu M, Lam k (2018) A joint framework for QoS and QoE for video transmission over wireless multimedia sensor networks. IEEE Trans Mob Comput 17(4):746–759Google Scholar
- Wang K, Mi J, Xu C, Zhu Q, Shu L, Deng DJ (2016) Real-time load reduction in multimedia big data for mobile Internet. ACM Trans Mult Comput Commun App 12(5):76Google Scholar
- Wang Y, Wang* K, Huang H, Miyazaki T, Guo S (2019) Traffic and computation co-offloading with reinforcement learning in fog computing for industrial applications. IEEE Transactions on Industrial Informatics. IEEE Trans Ind Informatics 15(2):976–986Google Scholar
- Xu C, Wang K, Li L, Xia R, Guo S, Guo M (2018) Renewable energy-aware big data analytics in geo-distributed data centers with reinforcement learning. IEEE Trans Netw Sci Eng. https://doi.org/10.1109/TNSE.2018.2813333
- Zhang J, Wang Z, Wang K, Guo S, Guo M (2017) Improving power efficiency for online video streaming service: a self-adaptive approach. IEEE Trans Sust Comput. https://doi.org/10.1109/TSUSC.2017.2739798