Department of Information Technology – Broadband Communication Networks (IBCN) Dynamic QoE Optimisation for Streaming Content in Large- Scale Future Networks Jeroen Famaey, Bart De Vleeschauwer, Tim Wauters, Filip De Turck, Bart Dhoedt, Piet Demeester INTEC Broadband Communication Networks (IBCN) Department of Information Technology (INTEC) Ghent University
Department of Information Technology – Broadband Communication Networks (IBCN) 2 Overview Introduction Server and Bit-Rate Selection Evaluation Conclusions & Future Work
Department of Information Technology – Broadband Communication Networks (IBCN) 3 Introduction – Current Situation Video-based services are highly popular Video sharing Internet-TV … Video quality Usually low to very low No dynamic quality adaptation Centralised delivery architecture Scales poorly Single point-of-failure
Department of Information Technology – Broadband Communication Networks (IBCN) 4 Introduction – Our Approach Content hosting Content discovery Quality adaptation Bwd monitoring Overlay routing Content discovery Content selection Bwd monitoring
Department of Information Technology – Broadband Communication Networks (IBCN) 5 Overview Introduction Server and Bit-Rate Selection Protocol Description Static Selection Algorithm Dynamic Selection Algorithm Evaluation Conclusions & Future Work
Department of Information Technology – Broadband Communication Networks (IBCN) 6 Server and Bit-Rate Selection Competing goals Improve end-user Quality of Experience (QoE) Admit as many end-user requests as possible What is Quality of Experience? Jitter Delay Packet loss Bit-rate … Transcoding plugins Bit-rate QoE Computational resources QoE End-user perception
Department of Information Technology – Broadband Communication Networks (IBCN) 7 Server and Bit-Rate Selection Discovery Selection Accept Reject
Department of Information Technology – Broadband Communication Networks (IBCN) 8 Static Selection Algorithm (1) Sorts servers and plugins based on metric Naïve metrics MinRes: minimise computational-resource usage Maximises bit-rate and QoE Best under computational-resources bottleneck MinBwd: minimise bandwidth usage Minimises bit-rate and QoE Best under bandwidth bottleneck Weighted metric Servers: Plugins:
Department of Information Technology – Broadband Communication Networks (IBCN) 9 Static Selection Algorithm (2) α in [0,1] = normalised resource usage β in [0,1] = normalised bandwidth usage αβ action low use high bit-rate plugin lowhighuse low bit-rate plugin highlowuse high bit-rate plugin high use medium bit-rate plugin Requirements: w 1 = 1 + α - βw 2 = β
Department of Information Technology – Broadband Communication Networks (IBCN) 10 Dynamic Selection Algorithm Only used for plugin selection Adapt bit-rate while video is playing Admit new stream: For all plugins P i (sorted by decreasing bit- rate) Downscale all videos that use P i-1 to P i If: enough resources and bandwidth, use P i Remove finished stream: For all plugins P i (sorted by increasing bit- rate) Upscale videos that use P i to P i+1
Department of Information Technology – Broadband Communication Networks (IBCN) 11 Overview Introduction Server and Bit-Rate Selection Evaluation Simulation Setup Static Scenarios Dynamic Scenarios Conclusions & Future Work
Department of Information Technology – Broadband Communication Networks (IBCN) 12 Evaluation – Simulation Setup Implemented in PlanetSim P2P simulator Network 20 content servers 50 client proxies (actual clients not simulated) Transcoding plugins P 1 : 350 Kbps, 1000 Mhz, 80% QoE P 2 : 700 Kbps, 500 Mhz, 90% QoE P 3 : 1400 Kbps, 100 Mhz, 100% QoE Videos 100 different videos Duration in [1,11] minutes interval (YouTube-like use-case) Popularity directly proportional to availability (linear) Popularity directly proportional to request count (zipf) Evaluation metrics Satisfied request count Globally obtained QoE
Department of Information Technology – Broadband Communication Networks (IBCN) 13 Evaluation – Static Scenarios Request rate remains constant during simulation Scenarios Computational-resource bottleneck Bandwidth bottleneck Computational-resource and bandwidth bottleneck (shown)
Department of Information Technology – Broadband Communication Networks (IBCN) 14 Evaluation – Dynamic Scenarios Scenarios Static (computational-resource and bandwidth bottleneck) Dynamic 1 (increase from 30 to 125 requests per minute) Dynamic 2 (video durations multiplied by 10) (shown)
Department of Information Technology – Broadband Communication Networks (IBCN) 15 Overview Introduction Server and Bit-Rate Selection Evaluation Conclusions & Future Work
Department of Information Technology – Broadband Communication Networks (IBCN) 16 Conclusions and Future Work Contributions Scalable QoE-aware video delivery architecture Protocol and algorithms to adapt video quality Maximise obtained QoE Admit as many client requests as possible Evaluation Intelligent weighted static metric performs up to 30% better than naïve metrics Dynamic algorithm mostly useful in context of longer video durations (e.g. Video-on- Demand) Future work Reduce assumptions made Scalable overlay routing protocols to route around bandwidth bottlenecks in the core network
Department of Information Technology – Broadband Communication Networks (IBCN) Questions ? Jeroen Famaey, Bart De Vleeschauwer, Tim Wauters, Filip De Turck, Bart Dhoedt, Piet Demeester INTEC Broadband Communication Networks (IBCN) Department of Information Technology (INTEC) Ghent University