Intent-based Application and Network Cooperative Control Technology for Video-streaming Services

2.1 Rate control technology for streaming video in applications

Adaptive bitrate (ABR) is one of the application controls at the user device that controls the rate of video streaming. In ABR, multiple-bitrate videos of different quality are pre-populated on the video-streaming server, and the application controls the rate request by switching the bitrate of the video to the streaming server in accordance with network quality such as throughput. This enables the application to increase, for example, user satisfaction by reducing playback stops. As shown in Fig. 1, we also developed a technology for predicting future quality of experience (QoE) using our QoE-estimation technology [1] on the basis of throughput measured at the user device and application information (video/audio bitrate time-series information, etc.), and controlling rates to satisfy the QoE set by the user (requested-QoE) in the application [2]. This technology ensures higher user satisfaction than conventional technologies. However, when network quality deteriorates due to congestion, application control alone cannot satisfy the requested-QoE because it can only lower the bitrate and reduce video quality. Thus, application control alone cannot maintain user satisfaction when network quality deteriorates.

Fig. 1. Intent-based rate control technology.

2.2 Network control technology

Current network control technology estimates quality of service (QoS) by using traffic forecasts for each traffic flow on the basis of communication logs observed with network devices and controls routes so that the resource-utilization efficiency is high while maintaining actual QoS. Traffic flow is communication traffic with the same source Internet protocol (IP) address, destination IP address, port number, and protocol number. We developed a technology that groups flows with similar characteristics into macro flows by using non-negative tensor factorization then accurately predicts future traffic for each macro flow by using seasonal autoregressive integrated moving average (SARIMA) [3, 4]. Network control using accurate traffic prediction improves the efficiency of network-resource utilization. Therefore, the network control technology controls routes on the basis of the actual QoS estimated from communication logs, but it is difficult to correctly estimate the QoS required to satisfy a user’s desired QoE from communication logs alone. The QoS required to satisfy the user’s desired QoE is called the requested-QoS. Therefore, if there is a large discrepancy between the requested-QoS and the actual QoS, the QoE may decrease due to insufficient quality, or network-resource-utilization efficiency may decrease due to excessive quality. In addition, if the application-rate control technology described above and network control technology operate independently in video delivery, the QoE may not be satisfied. When network quality degrades due to congestion, the application-rate control technology delivers low-quality video to resolve the congestion. In this case, the network control technology determines that congestion is naturally resolved and does not change the route. As a result, low-quality video continues to be delivered and the QoE decreases. Thus, the network control technology alone cannot satisfy the requested-QoE because requirements of the requested-QoS cannot be accurately estimated and the effect of the application-rate control technology cannot be taken into account.

3.1 Overall structure

The overall structure of the proposed technology is shown in Fig. 2. The technology consists of a rate-control function and a route-control function using our technology described above, and an application-network-cooperation function that enables cooperation between these functions. In the rate-control function, the user sets the requested-QoE for each user device, and the user device notifies the application-network-cooperation function of the requested-QoE and the representation information (a list of video and audio bitrates, video resolution, and frame rate that can be selected by the rate control). The application-network-cooperation function predicts the required bandwidth as future QoS requirements for each macro flow that bundles flows of users with the same requested-QoE on the basis of the requested-QoE for each user notified by all user devices. Network control satisfies the requested-QoE by executing optimal routing for each macro flow to satisfy the predicted required bandwidth. Therefore, application control and network control enable intent-based cooperative control by using the requested-QoE as a common index.

Fig. 2. Intent-based application and network cooperation function.

3.2 Application-network-cooperation function

As shown in Fig. 3, the application-network-cooperation function consists of a user-QoS-extraction function, active-user-prediction function, and macro-flow-QoS-extraction function. The user-QoS-extraction function calculates the network QoS requirements on the basis of the throughput required from the requested-QoE and application information reported through user devices. The relationship between the requested-QoE and throughput required to achieve it depends on the representation information designed for each video-streaming service. Therefore, the relationship between the requested-QoE and required throughput is modeled in consideration of the representation information, and the model is used to estimate the required throughput from the requested-QoE and representation information. The active-user-prediction function predicts the future number of active users for each macro flow. It derives this number using SARIMA, a time-series prediction method, from the time-series data of the number of active users for each macro flow notified from user devices to the application-network-cooperation function. The macro-flow-QoS-extraction function calculates the required bandwidth for each macro flow on the basis of the estimated required throughput per user and the predicted number of active users per macro flow. These functions enable network routing to be controlled on the basis of network requirements considering the requested-QoE.

Fig. 3. Application and network cooperation function.

4.1 Simulation overview

The simulation scenario assumes routing control when one path is congested in a network structure where network bottlenecks are duplicated, such as a point of interface between a carrier network and the Internet. Simulations were conducted using ns-3, an open source network simulator. Figure 4 shows the simulation environment. The network structure consists of three routers fully mesh-connected in a triangular structure. There is one video server connected to one router and 90 user terminals connected to each of the remaining two routers. There are also two routes between the video server and user devices. User devices can set their requested-QoE taking values from 1.0 to 5.0 to 4.5, 4.0, or 3.5, and 30 user devices with each requested-QoE were connected to each router. Congestion was generated by flowing background traffic to the video server from a user device. The effectiveness of the proposed and conventional technologies was evaluated by comparing the results of routing control during congestion in terms of achieving the requested-QoE and network-resource-utilization efficiency.

Fig. 4. Simulation environment.

4.2 Effectiveness verification through simulation

Figure 5 shows the percentage of users who achieved the requested-QoE after routing with the proposed and conventional technologies during congestion. About 22% of all users did not achieve the requested-QoE with the conventional technology. When limited to users with a requested-QoE of 4.5, about two-thirds did not meet the requested-QoE. This is because the network control was unable to correctly estimate the required bandwidth because it derived this bandwidth using only the information in the network and could not take into account the requested-QoE or application control. The proposed technology satisfied the requested-QoE for all users. This is because this technology estimates the required bandwidth to satisfy the QoE with higher accuracy than the conventional technology by using the requested-QoE obtained from the user-device application. Thus, the proposed technology can achieve higher user satisfaction than the conventional technology. In addition, if the percentage of users who achieved the requested-QoE with the conventional technology by increasing the link bandwidth is equal to that with the proposed technology, a 16% increase in link bandwidth is required. This indicates that the proposed technology is superior to the conventional technology in terms of network-resource-utilization efficiency.

Fig. 5. Results of effectiveness evaluation through simulation.