CS2P: Improving Video Bitrate Selection and Adaptation with - PowerPoint PPT Presentation

CS2P: Improving Video Bitrate Selection and Adaptation with Data-Driven Throughput Prediction Y. Sun, F. Lin, N. Wang @ X. Yin, J. Jiang, V. Sekar, B. Sinopoli @ T. Liu @

Bitrate adaptation is key for QoE • DASH = Dynamic Adaptive Streaming over HTTP • Entail new QoE metrics, e.g., low buffering, high video quality • Need intelligent bitrate control and adaptation Prior work: Accurate throughput prediction can help! 2

Accurate throughput prediction à Better initial bitrate selection — Fixed bitrate — Adaptive bitrate 3

Accurate throughput prediction à Better midstream adaptation — Replicate the analysis by Yin et al. at SIGCOMM2015 [1] /01234 567 p 𝑂𝑝𝑠𝑛𝑏𝑚𝑗𝑨𝑓𝑒 𝑅𝑝𝐹 = 89:6;:1<034 =>?@ABC [1] X. Yin, et al. “A Control-Theoretic Approach for Dynamic Adaptive Video Streaming over HTTP”. ACM SIGCOMM, 2015 . 4 [2] T. Y. Huang, et al. “A Buffer-Based Approach to Rate Adaptation: Evidence from a Large Video Streaming Service”. ACM SIGCOMM, 2014 .

Open questions on predictability! — Our understanding of throughput variability and predictability is quite limited. — What types of prediction algorithms to use? — In the context of video bitrate adaptation — Prior approaches: 30%+ of predictions with error ≥0.2 5

Our work and contributions A large-scale analysis, providing data-driven insights for predicting the throughput accurately. Design of CS2P (Cross-Session Stateful Predictor): Improving bitrate selection and adaptation via throughput modeling. A practical implementation of CS2P and the demonstration of improvements in video QoE. 6

Outline — Motivation è Data-driven Observations — CS2P Approach — Evaluation 7

Dataset description — From operational platform of iQIYI . — iQIYI is a leading online video Feature Coverage content provider in China. Client IP 3.2M Client ISP 87 — 20M+ sessions, 8 days in Sep. 2015, Client AS 161 — Each session records avg. Province 33 throughput per 6-second epoch . City 736 Server 18 8

Observation 1: Significant variability within a session. 20% of sessions with n-stddev ≥ 50% 50% of sessions with n-stddev ≥ 30% 9

Observation 2: Stateful/persistent characteristics. Throughput variation across two consecutive An example session. epochs with a particular IP/16 prefix. 10

Observation 3: Similar session à Similar throughput Throughput at different session clusters with particular IP/8 prefixes. 11

Observation 4: Complex relationship between session feature throughput Combinations of multiple features often have a The impact of the same feature on different much greater impact than the individual feature. sessions could be variable. 12

Outline — Motivation — Data-driven Observations è CS2P Approach — Evaluation 13

Observation Idea Many sessions exhibit stateful CS2P learns Hidden-Markov characteristics in the evolution of Models (HMM) to capture the the throughput. states and state transitions. CS2P groups similar sessions Sessions sharing similar critical sharing the same critical feature characteristics tend to exhibit values and uses Cross-Session similar throughput patterns. prediction methodology. The relationship between session CS2P learns a separate model for features and throughput are quite each similar session clusters complex. instead of using a global model. 14

Workflow of CS2P Throughput Measurements Step 1: Session Clustering Video Server Step 2: Model Training Prediction Engine 1.Initial Throughput Step 3: Throughput 2.Prediction Model Prediction and Bitrate Selection 15 Video Player Clients

Session clustering-finding critical features All the sessions for training Session under prediction A given subset of session features Try another session feature subset Sessions matching selected Repeat these procedures to find the critical feature set, features with which yields the most accurate throughput prediction of Predict the throughput of with these filtered sessions 16

Throughput prediction with HMM 0.972 Model Train: A Hidden-Markov Model per cluster via EM State 1 Gaussian (0.43,0.05 2 ) algorithm (offline). Mbps 0.016 0.055 0.012 0.020 0.069 State 2 State 3 Gaussian(2.41,1.49 2 ) Gaussian(1.20,0.10 2 ) Mbps Mbps 0.010 0.876 0.970 Hidden State X t- 1 X t Throughput prediction and 𝑌 𝑢 ϵ 𝝍 bitrate selection (online). Throughput W t- 1 W t 𝑋 𝑢 ϵ 𝑺 17

Outline — Motivation — Data-driven Observations — CS2P Approach è Evaluation 18

Trace-driven simulation setup Algorithms to compare: iQIYI throughput trace: 1. History-based predictor: • Non-overlapping traces Last Sample, Harmonic- • of training and testing Mean, Auto Regression Video source: 2. ML-based predictor: SVR, Gradient Boosting • • “Envivio” from dash.js test Regression trees website 3. CFA [1] • Encoded in H.264/MPEG-4 Bitrate selection method: in 5 bitrate levels • State-of-art: MPC [2] [1] J. Jiang, et al. “CFA: A Practical Prediction System for Video QoE Optimization”. In Proc. of USENIX NSDI, 2016. [2] X. Yin, et al.“A Control-Theoretic Approach for Dynamic Adaptive Video Streaming over HTTP”. In Proc. of 19 ACM SIGCOMM, 2015 .

Throughput Prediction Accuracy Takeaway l Midstream Epoch p Reduce median error by 50% l Multi-epoch Ahead p 9% prediction error for 10 epoch Reduce by 50% ahead p 50% improvement Midstream Throughput 20

Video QoE /01234 567 — 𝑂𝑝𝑠𝑛𝑏𝑚𝑗𝑨𝑓𝑒 𝑅𝑝𝐹 = 89:6;:1<034 =>?@ABC — QoE [1] is a linear combination of avg. video quality, quality variation, total rebuffer time and startup delay. 19% Midstream epoch 5% 8% 21 [1] X. Yin, et al. “A Control-Theoretic Approach for Dynamic Adaptive Video Streaming over HTTP”. In Proc. of ACM SIGCOMM, 2015 .

Pilot deployment: multi-city test Metrics vs. HM+MPC vs. BB Avg. Bitrate 10.9% 9.3% Good Ratio 2.5% 17.6% Bitrate Variability -2.3% 5.6% Startup Delay 0.4% -3.0% Overall QoE 3.2% 14.0% Takeaway: 1. CS2P improves most of the QoE metrics, except longer startup delay than BB and higher bitrate variability than HM. 2. The overall QoE improvement of CS2P is 3.2% to HM and 14% to BB. 22

Conclusions l Good prediction è Better bitrate selection & adaptation è Improved video QoE l Key insights on throughput variability p Evolution of intra-session throughput exhibits stateful characteristics. p Similar sessions have similar throughput structures. l CS2P: Cross-session HMM-based approach l Outperform prior predictors by 50% in midstream prediction error. l Achieve 3.2% improvement to HM and 14% to BB in video QoE. 23