| Network Working Group | X. Zhu |
| Internet-Draft | S. Mena |
| Intended status: Informational | Cisco Systems |
| Expires: January 20, 2019 | Z. Sarker |
| Ericsson AB | |
| July 19, 2018 |
Video Traffic Models for RTP Congestion Control Evaluations
draft-ietf-rmcat-video-traffic-model-05
This document describes two reference video traffic models for evaluating RTP congestion control algorithms. The first model statistically characterizes the behavior of a live video encoder in response to changing requests on target video rate. The second model is trace-driven, and emulates the output of actual encoded video frame sizes from a high-resolution test sequence. Both models are designed to strike a balance between simplicity, repeatability, and authenticity in modeling the interactions between a live video traffic source and the congestion control module. Finally, the document describes how both approaches can be combined into a hybrid model.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on January 20, 2019.
Copyright (c) 2018 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
When evaluating candidate congestion control algorithms designed for real-time interactive media, it is important to account for the characteristics of traffic patterns generated from a live video encoder. Unlike synthetic traffic sources that can conform perfectly to the rate changing requests from the congestion control module, a live video encoder can be sluggish in reacting to such changes. Output rate of a live video encoder also typically deviates from the target rate due to uncertainties in the encoder rate control process. Consequently, end-to-end delay and loss performance of a real-time media flow can be further impacted by rate variations introduced by the live encoder.
On the other hand, evaluation results of a candidate RTP congestion control algorithm should mostly reflect performance of the congestion control module, and somewhat decouple from peculiarities of any specific video codec. It is also desirable that evaluation tests are repeatable, and be easily duplicated across different candidate algorithms.
One way to strike a balance between the above considerations is to evaluate congestion control algorithms using a synthetic video traffic source model that captures key characteristics of the behavior of a live video encoder. To this end, this draft presents two reference models. The first is based on statistical modeling; the second is trace-driven. The draft also discusses the pros and cons of each approach, as well as how both approaches can be combined into a hybrid model.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.
A live video encoder employs encoder rate control to meet a target rate by varying its encoding parameters, such as quantization step size, frame rate, and picture resolution, based on its estimate of the video content (e.g., motion and scene complexity). In practice, however, several factors prevent the output video rate from perfectly conforming to the input target rate.
Due to uncertainties in the captured video scene, the output rate typically deviates from the specified target. In the presence of a significant change in target rate, it sometimes takes several frames before the encoder output rate converges to the new target. Finally, while most of the frames in a live session are encoded in predictive mode, the encoder can occasionally generate a large intra-coded frame (or a frame partially containing intra-coded blocks) in an attempt to recover from losses, to re-sync with the receiver, or during the transient period of responding to target rate or spatial resolution changes.
Hence, a synthetic video source should have the following capabilities:
While there exist many different approaches in developing a synthetic video traffic model, it is desirable that the outcome follows a few common characteristics, as outlined below.
These distinct behavior features can be characterized via simple statistical modelling, or a trace-driven approach. Section 5 and Section 6 provide an example of each approach, respectively. Section 7 discusses how both models can be combined together.
Figure 1 depicts the interactions of the synthetic video traffic source with other components at the sender, such as the application, the congestion control module, the media packet transport module, etc. Both reference models --- as described later in Section 5 and Section 6 --- follow the same set of interactions.
The synthetic video source dynamically generates a sequence of dummy video frames with varying size and interval. These dummy frames are processed by other modules in order to transmit the video stream over the network. During the lifetime of a video transmission session, the synthetic video source will typically be required to adapt its encoding bitrate, and sometimes the spatial resolution and frame rate.
In our model, the synthetic video source module has a group of incoming and outgoing interface calls that allow for interaction with other modules. The following are some of the possible incoming interface calls --- marked as (a) in Figure 1 --- that the synthetic video traffic source may accept. The list is not exhaustive and can be complemented by other interface calls if deemed necessary.
An example of outgoing interface call --- marked as (b) in Figure 1 --- is the rate range [R_min, R_max]. Here, R_min and R_max are meant to capture the dynamic rate range and actual live video encoder is capable of generating given the input video content. This typically depends on the video content complexity and/or display type (e.g., higher R_max for video contents with higher motion complexity, or for displays of higher resolution). Therefore, these values will not change with R_v, but may change over time if the content is changing.
+-------------+
| | encoded video
| Synthetic | frames
| Video | -------------->
| Source |
| |
+--------+----+
/|\ |
| |
-------------------+ +-------------------->
interface from interface to
other modules (a) other modules (b)
Figure 1: Interaction between synthetic video encoder and other modules at the sender
This section describes one simple statistical model of the live video encoder traffic source. Figure 2 summarizes the list of tunable parameters in this statistical model. A more comprehensive survey of popular methods for modeling video traffic source behavior can be found in [Tanwir2013].
+===========+====================================+================+
| Notation | Parameter Name | Example Value |
+===========+====================================+================+
| R_v | Target rate request | 1 Mbps |
+-----------+------------------------------------+----------------+
| FPS | Target frame rate | 30 Hz |
+-----------+------------------------------------+----------------+
| tau_v | Encoder reaction latency | 0.2 s |
+-----------+------------------------------------+----------------+
| K_d | Burst duration during transient | 8 frames |
+-----------+------------------------------------+----------------+
| K_B | Burst frame size during transient | 13.5 KBytes* |
+-----------+------------------------------------+----------------+
| t0 | Reference frame interval 1/FPS | 33 ms |
+-----------+------------------------------------+----------------+
| B0 | Reference frame size R_v/8/FPS | 4.17 KBytes |
+-----------+------------------------------------+----------------+
| | Scaling parameter of the zero-mean | |
| | Laplacian distribution describing | |
| SCALE_t | deviations in normalized frame | 0.15 |
| | interval (t-t0)/t0 | |
+-----------+------------------------------------+----------------+
| | Scaling parameter of the zero-mean | |
| | Laplacian distribution describing | |
| SCALE_B | deviations in normalized frame | 0.15 |
| | size (B-B0)/B0 | |
+-----------+------------------------------------+----------------+
| R_min | minimum rate supported by video | 150 Kbps |
| | encoder type or content activity | |
+-----------+------------------------------------+----------------+
| R_max | maximum rate supported by video | 1.5 Mbps |
| | encoder type or content activity | |
+===========+====================================+================+
* Example value of K_B for a video stream encoded at 720p and
30 frames per second, using H.264/AVC encoder.
Figure 2: List of tunable parameters in a statistical video traffic source model.
While the congestion control module can update its target rate request R_v at any time, the statistical model dictates that the encoder will only react to such changes tau_v seconds after a previous rate transition. In other words, when the encoder has reacted to a rate change request at time t, it will simply ignore all subsequent rate change requests until time t+tau_v.
The output rate R_o during the period [t, t+tau_v] is considered to be in transient. Based on observations from video encoder output data, the transient behavior of an encoder upon reacting to a new target rate request is modelled in the form of high variation in output frame sizes. It is assumed that the overall average output rate R_o during this period matches the target rate R_v. Consequently, the occasional burst of large frames are followed by smaller-than-average encoded frames.
This temporary burst is characterized by two parameters:
It can be noted that these burst parameters can also be used to mimic the insertion of a large on-demand I frame in the presence of severe packet losses. The values of K_d and K_B typically depend on the type of video codec, spatial and temporal resolution of the encoded stream, as well as the video content activity level.
The output rate R_o during steady state is modelled as randomly fluctuating around the target rate R_v. The output traffic can be characterized as the combination of two random processes denoting the frame interval t and output frame size B over time. These two random processes capture two sources of variations in the encoder output:
Both values of SCALE_t and SCALE_B can be obtained via parameter fitting from empirical data captured for a given video encoder. Example values are listed in Figure 2 based on empirical data presented in [IETF-Interim].
The output rate R_o is further clipped within the dynamic range [R_min, R_max], which in reality are dictated by scene and motion complexity of the captured video content. In the proposed statistical model, these parameters are specified by the application.
The second approach for modelling a video traffic source is trace-driven. This can be achieved by running an actual live video encoder on a set of chosen raw video sequences and using the encoder's output traces for constructing a synthetic video source. With this approach, the recorded video traces naturally exhibit temporal fluctuations around a given target rate request R_v from the congestion control module.
The following list summarizes the main steps of this approach:
In the following, Section 6.1 explains the first three steps (1-3), Section 6.2 elaborates on the remaining two steps (4-5). Finally, Section 6.3 briefly discusses the possibility to extend the trace-driven model for supporting time-varying frame rate and/or time-varying frame resolution.
The first step is a careful choice of a set of video sequences that are representative of the target use cases for the video traffic model. For the example use case of interactive video conferencing, it is recommended to choose a low-motion sequence that resembles a "talking head", e.g. from a news broadcast or recording of an actual video conferencing call.
The length of the chosen video sequence is a tradeoff. If it is too long, it will be difficult to manage the data structures containing the traces. If it is too short, there will be an obvious periodic pattern in the output frame sizes, leading to biased results when evaluating congestion control performance. In our experience, a sequence with a length between 2 and 4 minutes is a fair tradeoff.
Given the chosen raw video sequence, denoted S, one can use a live encoder, e.g. some implementation of [H264] or [HEVC], to produce a set of encoded sequences. As discussed in Section 3, the output bitrate of the live encoder can be achieved by tuning three input parameters: quantization step size, frame rate, and picture resolution. In order to simplify the choice of these parameters for a given target rate, one can typically assume a fixed frame rate (e.g. 30 fps) and a fixed resolution (e.g., 720p) when configuring the live encoder. See Section 6.3 for a discussion on how to relax these assumptions.
Following these simplifications, the chosen encoder can be configured to start at a constant target bitrate, then vary the quantization step size (internally via the video encoder rate controller) to meet various externally specified target rates. It can be further assumed the first frame is encoded as an I-frame and the rest are P-frames. For live encoding, the encoder rate control algorithm typically does not use knowledge of frames in the future when encoding a given frame.
Given the minimum and maximum bitrates at which the synthetic codec is to operate (denoted as R_min and R_max, see Section 4), the entire range of target bitrates can be divided into n_s + 1 bitrate steps of length l = (R_max - R_min) / n_s. The following simple algorithm is used to encode the raw video sequence.
r = R_min
while r <= R_max do
Traces[r] = encode_sequence(S, r, e)
r = r + l
The function encode_sequence takes as input parameters, respectively, a raw video sequence (S), a constant target rate (r), and an encoder rate control algorithm (e); it returns a vector with the sizes of frames in the order they were encoded. The output vector is stored in a map structure called Traces, whose keys are bitrates and whose values are vectors of frame sizes.
The choice of a value for n_s is important, as it determines the number of vectors of frame sizes stored in the map Traces. The minimum value one can choose for n_s is 1, and the maximum value depends on the amount of memory available for holding the map Traces. A reasonable value for n_s is one that results in steps of length l = 200 kbps. The next section will discuss further the choice of the step length l.
Finally, note that, as mentioned in previous sections, R_min and R_max may be modified after the initial sequences are encoded. Hence, the algorithm described in the next section also covers the cases when the current target bitrate is less than R_min, or greater than R_max.
The main idea behind the trace-driven synthetic codec is that it mimics the rate adaptation behavior of a real live codec upon dynamic updates of the target rate R_v by the congestion control module. It does so by switching to a different frame size vector stored in the map Traces when needed.
The main algorithm for rate adaptation in the synthetic codec maintains two variables: r_current and t_current.
r_current = r
such that
(r in keys(Traces) and
r <= R_v and
(not(exists) r' in keys(Traces) such that r <r'<= R_v))
if t_current < SkipFrames then
t_current = t_current + 1
else
t_current = ( (t_current + 1 - SkipFrames)
% (size_traces-SkipFrames))
+ SkipFrames
where operator % denotes modulo, and SkipFrames is a predefined constant that denotes the number of frames to be skipped at the beginning of frame size vectors after t_current has wrapped around. The point of constant SkipFrames is avoiding the effect of periodically sending a large I-frame followed by several smaller-than-average P-frames. A typical value of SkipFrames is 20, although it could be set to 0 if one is interested in studying the effect of sending I-frames periodically.
The initial value of r_current is set to R_min, and the initial value of t_current set to 0.
When a new frame is due, its size can be calculated following one of the three cases below:
size_lo = Traces[r_current][t_current]
size_hi = Traces[r_current + l][t_current]
distance_lo = (R_v - r_current) / l
framesize = size_hi*distance_lo + size_lo*(1-distance_lo)
factor = R_v / R_min
framesize = max(1, factor * Traces[R_min][t_current])
factor = R_v / R_max
framesize = factor * Traces[R_max][t_current]
In case b), we set the minimum output size to 1 byte, since the value of factor can be arbitrarily close to 0.
Note that main algorithm as described above can be further extended to mimic some additional typical behaviors of a live video encoder. Two examples are given below:
The trace-driven synthetic codec model explained in this section is relatively simple due to fixed frame rate and frame resolution. The model can extended further to accommodate variable frame rate and/or variable spatial resolution.
When the encoded picture quality at a given bitrate is low, one can potentially decrease either the frame rate (if the video sequence is currently in low motion) or the spatial resolution in order to improve quality-of-experince (QoE) in the overall encoded video. On the other hand, if target bitrate increases to a point where there is no longer a perceptible improvement in the picture quality of individual frames, then one might afford to increase the spatial resolution or the frame rate (useful if the video is currently in high motion).
Many techniques have been proposed to choose over time the best combination of encoder quatization step size, frame rate, and spatial resolution in order to maximize the quality of live video codecs [Ozer2011][Hu2010]. Future work may consider extending the trace-driven codec to accommodate variable frame rate and/or resolution.
From the perspective of congestion control, varying the spatial resolution typically requires a new intra-coded frame to be generated, thereby incurring a temporary burst in the output traffic pattern. The impact of frame rate change tends to be more subtle: reducing frame rate from high to low leads to sparsely spaced larger encoded packets instead of many densely spaced smaller packets. Such difference in traffic profiles may still affect the performance of congestion control, especially when outgoing packets are not paced by the media transport module. Investigation of varying frame rate and resolution are left for future work.
It is worthwhile noting that the statistical and trace-driven models each has its own advantages and drawbacks. Both models are fairly simple to implement. It takes significantly greater effort to fit the parameters of a statistical model to actual encoder output data whereas it is straightforward for a trace-driven model to obtain encoded frame size data. On the other hand, once validated, the statistical model is more flexible in mimicking a wide range of encoder/content behaviors by simply varying the correponding parameters in the model. In this regard, a trace-driven model relies -- by definition -- on additional data collection efforts for accommodating new codecs or video contents.
In general, the trace-driven model is more realistic for mimicking ongoing, steady-state behavior of a video traffic source whereas the statistical model is more versatile for simulating transient events (e.g., when target rate changes from A to B with temporary bursts during the transition). It is also possible to combine both models into a hybrid approach, using traces during steady-state and statistical model during transients.
+---------------+
transient | Generate next |
+------>| K_d transient |
+-------------+ / | frames |
R_v | Compare | / +---------------+
------->| against |/
| previous |
| target rate |\
+-------------+ \ +---------------+
\ | Generate next |
+------>| frame from |
steady-state | trace |
+---------------+
Figure 3: Hybrid approach for modeling video traffic
As shown in Figure 3, the video traffic model operates in transient state if the requested target rate R_v is substantially higher than the previous target, or else it operates in steady state. During transient state, a total of K_d frames are generated by the statistical model, resulting in one (1) big burst frame with size K_B followed by K_d-1 smaller frames. When operating at steady-state, the video traffic model simply generates a frame according to the trace-driven model given the target rate, while modulating the frame interval according to the distribution specified by the statistical model. One example criterion for determining whether the traffic model should operate in transient state is whether the rate increase exceeds 10% of previous target rate. Finally, as this model follows transient state behavior dictated by the statistical model, upon a substantial rate change, the model will follow the time-damping mechanism defined in Section 5.1, which is governed by parameter tau_v.
The statistical model has been implemented as a traffic generator module within the [ns-2] network simulation platform.
More recently, the statistical, trace-driven, and hybrid models have been implemented as a stand-alone, platform-independent traffic source module. This can be easily integrated into network simulation platforms such as [ns-2] and [ns-3], as well as testbeds using a real network. The stand-alone traffic source module is available as an open source implementation at [Syncodecs].
There are no IANA impacts in this memo.
It is important to evaluate RTP-based congestion control schemes using realistic traffic patterns, so as to ensure stable operations of the network. Therefore, it is RECOMMENDED that candidate RTP-based congestion control algorithms be tested using the video traffic models presented in this draft before wide deployment over the Internet.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [RFC8174] | Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017. |