ITU-T G.1022 PDF

This Recommendation defines models to estimate the buffer occupancy and operational state used in client media players employing TCP transport and other forms of reliable transport.

The measurement point(s) provide an observation point on TCP connection(s) primarily within the end-to-end path, where input data are collected.

The scope of work can be described in terms of inputs and outputs of the model(s):

Exact procedures for processing network layers and subsequent container processing are beyond the scope of this Recommendation, as it is highly variable and obviously must be matched to the specific circumstances. However, general guidance for readers is provided.

Inputs – Frames or partial frames (audio/visual (A/V) frames) pertaining to the reliable transport connection(s) in use and the time stamp when each packet appears at the measurement point. Normal operation considers both directions of transmission. Alternatively, meta-data from the container or streaming media application layer and network layers may be used whenever present.

Outputs – for each model state change, it is required to report the new state name, the timestamp of the change in units of measure equal to 1 millisecond resolution. Optionally, the current buffer fill level in units of milliseconds is reported for each change. There is optional periodic reporting of the buffer fill level with timestamp and current state name. Also, the frame type and size information that is played-out may be reported and for some models, the complete media stream information. Furthermore, if there are multiple video streams (e.g., in the adaptive case) a possible stream change could be reported. When available for each frame, there shall be a descriptive tuple containing a synthesized or decoded status, frame type, frame size in bytes, duration in time (ms) represented by the frame, and the payload bytes. The frame type should be registered (somewhere) and referenced in the type value.

Operation – there are scenarios where it may not be possible to access the media frame boundaries within the A/V stream, such as when the media is encrypted or an un-decodeable media format is encountered. In such cases, the set of default frame tuples within a container may be synthesised, and the distinction between decoded frame boundaries and synthesised frames shall be indicated. When a sufficiently long pause is detected in the TCP stream (or media flow) and a waiting time for resumption of the TCP stream expires, the monitoring system declares an End of stream if the buffer is coincidentally depleted, and normal operation should cease.

All stored state, including buffer state and queuing diagnostic information may be made available for on demand retrieval. Some network analysis is allowed to assist the mid-path monitoring to determine frame arrival time and other signals that are inputs to the model (the details of complex network interactions are out of the scope of the Recommendation, but some general guidance is provided).

Advertisements interjected in user sessions are extremely important to the business of video delivery, and evaluation of their delivery may need to be separate to meet the needs of performance management.

NOTE – Non-reliable and UDP streams are outside of the scope of this Recommendation; they are covered in [b-ITU-T G.1021].