Video streaming represents a significant part of Internet traffic. During the playback, a video player monitors network throughput and dynamically selects the best video quality in given network conditions. Therefore, the video quality depends heavily on the player’s estimation of network throughput, which is challenging in the volatile environment of mobile networks. In this work, we improved the throughput estimation using prediction produced by LSTM artificial neural networks (ANNs). Hence, we acquired data traces from 4G and 5G mobile networks and supplied them to two deep LSTM ANNs, obtaining a throughput prediction for the next four seconds. Our analysis showed that the ANNs achieved better prediction accuracy compared to a naive predictor based on a moving average. Next, we replaced the video player’s default throughput estimation based on the naive predictor with the LSTM output. The experiment revealed that the traffic prediction improved video quality between 5% and 25% compared to the default estimation.

The increase of HTTP-based video popularity causes that broadband and Internet service providers’ links transmit mainly multimedia content. Network planning, traffic engineering or congestion control requires an understanding of the statistical properties of network traffic; therefore, it is desirable to investigate the characteristic of traffic traces generated by systems which employ adaptive bit-rate streaming. Our first contribution is an investigation of traffic originating from 120 client-server pairs, situated in an emulated content distribution network, and multiplexed onto a single network link. We show that the structure of the traffic is distinct from the structure generated by the first and second generation of HTTP video systems, and furthermore, not similar to the structure of general Internet traffic. The obtained traffic exhibits negative and positive correlations, anti-persistence, and its distribution function is skewed to the right. Our second contribution is an approximation of the traffic by ARIMA/FARIMA processes blue and artificial neural networks. As we show, the obtained traffic models are able to enhance the performance of an adaptive streaming algorithm.

Classification of video traffic is crucial for network management, enforcing quality of service, and optimising bandwidth. Feature selection plays a vital role in traffic identification by reducing data volume, enhancing accuracy, and reducing computational cost. This paper presents a comparative study of three feature selection approaches applied to video traffic identification: filter, wrapper, and embedded. Real-world traffic traces are collected from three popular video streaming platforms: YouTube, Netflix, and Amazon Prime Video, representing diverse content delivery characteristics. The main contributions of this work are (1) the identification of traffic generated by these streaming services, (2) a comparative evaluation of three feature selection methods, and (3) the application of previously untested algorithms for this task. We evaluate the examined methods using F1-score and computational efficiency. The results demonstrate distinct trade-offs among the approaches: the filter method offers low computational overhead with moderate accuracy, while the wrapper method achieves higher accuracy at the cost of longer processing times. The embedded method provides a balanced compromise by integrating feature selection within model training. This comparative analysis offers insights for designing video traffic identification systems in modern heterogeneous networks.

Video sharing services like YouTube have become very popular which consequently results in a drastic shift of the Internet traffic statistic. When transmitting video content over packet based networks, stringent quality of service (QoS) constraints must be met in order to provide the comparable level of quality to a traditional broadcast television. However, the packet transmission is influenced by delays and losses of data packets which can have devastating influence on the perceived quality of the video. Therefore, we conducted an experimental evaluation of HTTP based video transmission focusing on how they react to packet delay and loss. Through this analysis we investigated how long video playback is stalled and how often re-buffering events take place. Our analysis revealed threshold levels for the packet delay, packet losses and network throughput which should not be exceeded in order to preserve smooth video transmission.

Predicting throughput is essential to reduce latency in time-critical services like video streaming, which constitutes a significant portion of mobile network traffic. The video player continuously monitors network throughput during playback and adjusts the video quality according to the network conditions. This means that the quality of the video depends on the player’s ability to predict network throughput accurately, which can be challenging in the unpredictable environment of mobile networks. To improve the prediction accuracy, we grouped the throughput trace into clusters taking into account the similarity of their mean and variance. Once we distinguished the similar trace fragments, we built a separate LSTM predictive model for each cluster. For the experiment, we used traffic captured from 5G networks generated by individual user equipment (UE) in fixed and mobile scenarios. Our results show that the prior grouping of the network traces improved the prediction compared to the global model operating on the whole trace.

Due to the popularity of Dynamic Adaptive Streaming Over HTTP (DASH), broadband and Internet service providers’ links transmit mainly multimedia content. As the most popular providers encrypt their video services, the attempts to identify their traffic through Deep Packet Inspection (DPI) encounter difficulties. Therefore, encrypted DASH traffic requires new classification methods. In this work, we propose to identify DASH traffic taking into account statistical dependencies among video flows. For this purpose, we employ cluster analysis which can identify groups of traffic flows that show similarity using only the application level information. In our work, we applied three unsupervised clustering algorithms, namely MinMax K-Means, OPTICS and AutoClass, to classify video traces obtained from an emulated environment. The experimental results show that the employed algorithms are able to effectively distinguish video flows generated by different play-out strategies. The classification performance depends on the network conditions and parameters of the learning process.