[Rafael Silva Pereira] QoS Parameters for Video Streaming

When we talk about professional media streaming, there is a lot concern with the quality of service, especially when we have large volumes of requests, or when we delivery contents in high definition. In this scenario, is essential to have a fairly clear idea about what is quality of service, and how we can identify, through objective parameters, if we’re or not at an acceptable level, which ensures a high quality experience for end users.

In the case of video streaming, quality of service means that the user should watch a certain content, without interruption (rebuffering), and without degradation of video quality caused by the delivery process (loss of frames, problems in the reconstruction of b-frames and p-frames due packet loss).

Thus, the best known parameter for evaluating the quality of a streaming service is the buffer length, which is the ability to maintain the user’s buffer always at an acceptable level, being greater enough to ensure a continuous experience independent of small variations in bandwidth available, and in the bitrate of the encoded video. Associated with this parameter, we have the average rebuffering rate, which consists of a ratio between the total time used filling the buffer, and the total playing time. This parameter is very interesting to assess the quality of service for live streaming, where users can stay connected for a long time, and, of course, they would like to play the content without interruptions.

A high quality streaming service should always keep the rate of rebuffering below 0.5%, discarding the initial buffer, at normal bandwidth availability. Moreover, this parameter can be very useful to identify network bottlenecks at content distribution process.

Another key parameters are the average frame rate and the rate of frame loss. These two parameters give us an overview of the capacity, of the server to stream the video content, and, of the users to receive and decode the stream. In case of server overloading , one of the first alternatives used to maintain a continuous experience is to reduce the frame rate, or basically, drop frames. These parameters usually indicate that either the server is not able to serve the content, or, the clients do not have the minimum conditions to receive and decode the video. These problems are usually associated with intensive CPU usage, due an excess of clients in the server side, or by lack of resources for decoding the video on the client.

For Flash Media Server specifically, some parameters related with memory usage for input and output buffer, size of buckets, and use of aggregated messages, can influence directly in the video frame rate, but in this case, we can detect the problem even with a small amount of users.

Finally, we have some parameters, not least importants, to complement the set of basic checkpoints to ensure the quality of streaming video service, including, the time to establish the connection, the rate of reconnection, the sync between audio and video, and psnr between the video decoded by the user and the video that comes out of encoder (since the loss of packets may influence the reconstruction of the frames, especially for temporal video compression). Measuring and evaluating all these parameters we can identify with much more efficiency any problem that might compromise the quality of service, which is essential for professional media streaming business.