Quality of Service (QoS) and Call Admission Control (CAC) are not enough to manage network congestion. As video and encrypted application traffic consume over 90% of network bandwidth, enterprise network managers must apply dynamic session management if they want their networks to intelligently manage network traffic and minimize the impact of network congestion.
QoS breaks down when the majority of network traffic has the same QoS/DSCP markings such as AF41 for all video traffic and AF31 for all TLS or IPsec traffic. As more applications utilize video such as collaboration, augmented reality, real-time streaming, monitoring, and entertainment, to a network router each of these flows looks the same. As 4K video and adaptive codecs become more common, the burstiness of the video streams is overwhelming networks. CAC, while it works well for voice, limiting the maximum number of concurrent calls, does not work well for video, since the amount of bandwidth a video session varies widely, from 128Kbps up to 12Mbps and the amount of bandwidth required varies widely per second.
Because WAN bandwidth is deemed expensive, highly compressed voice and video codecs are typically used, which lowers the end user experience and puts a cap on the number of sessions that are carried. Even though the majority of the time, bandwidth is available for a high-quality end user experience, the pre-defined, static configuration of the network inhibits the use of the available bandwidth.
Once a voice/video session is established, the media is then peer to peer and the session is set. The codec type and all the other communication negotiations are done prior to session establishment. If the network runs out of bandwidth mid-session, then the communication will suffer either quality degradation or totally drop the session. Adaptive codecs were introduced to address this problem and the network communicates to the session by dropping packets. The flaw in this model is the premise that all established sessions of the same type/application are equal and all must adapt in the same way. The business reality is that not all sessions are equal and that under network congestion; some sessions should get higher priority than others.
Figure 1 illustrates a multi-vendor communications network architecture. This model works the majority of the time, but under high load, it falls apart. A few examples:
- Large Conference Call