Differentiated Quality Video Delivery in Overlay Multicasting Environment Ying Qiao Carleton University Project Presentation at the class: Quality of Service Management for Multimedia Applications Provided by: Professor Bochmann 12/6/2018
Outline Introduction Overlay multicast environment -- Internet multimedia delivery -- Types of Video service -- multimedia multicast Overlay multicast environment -- Video coding -- Video delivery Layered Peer-to-Peer Streaming Supporting Large-Scale Live Streaming Applications with Dynamic Application End-Points Incentive mechanism for Peer-to-Peer Media Streaming Conclusion 12/6/2018
Introduction (1) Internet media delivery Types of Video Service -- No VOD -- Pay-Per-view -- True VOD -- Near VOD (NVOD) -- Quasi-VOD (QVOD) Basic multicast functionality -- Group membership management -- Data delivery path maintenance -- Replication and forwarding 12/6/2018
Introduction (2) Internet media multicast IP multicast Overlay multicast Ref 4 12/6/2018
Overlay Multicasting Environment (1) Resources provided by peer end node -- Network bandwidth -- Storage space -- CPU power Features -- Overlay Multicast is deployed with the basic uni-cast routing infrastructure -- End hosts only maintain state for the groups they are participating in 12/6/2018
Overlay Multicasting Environment (2) Three architectures -- Dedicated-Infrastructure -- Application-Endpoint -- Waypoint [Ref 4] 12/6/2018
Overlay Multicasting Environment (3) Video Coding -- Replicated streaming -- Layered streaming -- Multiple Description Coding [Ref 1] 12/6/2018
Overlay Multicasting Environment (4) Video Delivery tree -- Single tree -- Multiple tree ZIGZAG [Ref 5] SplitStream [Ref 6] 12/6/2018
Overlay Multicasting Environment (5) Challenge for overlay multicast -- Bandwidth constraints -- Receiver scalability -- Network dynamics -- Receiver heterogeneity [Ref 4] 12/6/2018
Layered Peer-to-Peer Streaming (1) Layered video [Ref 2] -- Video is encoded into one base layer and multiple enhancement layers -- The base layer can be decoded independently -- The enhancement layers can be decoded cumulatively Network heterogeneity [Ref 3] 12/6/2018
Layered Peer-to-Peer Streaming (2) Large-scale on-demand multimedia distribution -- Asynchrony of user requests -- Heterogeneity of client resource capabilities Layered Peer-to-Peer Streaming -- Cache-and-relay -- Layer-encoded streaming 12/6/2018
Layered Peer-to-Peer Streaming (3) -- Cache-and-relay -- Layer-encoded streaming Goal -- Maximize the number of the received streams from end nodes other than the source -- Subject to (1) number of received streams for one receiver <= inbound bandwidth of the receiver (2) total number of received streaming from one sender <= outbound bandwidth of the sender 12/6/2018
Basic Algorithm Receiver k, inbound bandwidth a set of the hosts qualified as the supplying peers of and sorted the Hosts with the available layers Arranging the layers from the beginning of S 12/6/2018
Performance Evaluation (1) Request composition: -- Modem/ISDN peers, 50%, 112kbps -- Cable Modem/DSL peers, 35%, 1Mbps -- Ethernet Peers, 15%, 10Mbps Quality satisfaction -- The ratio of received quality and expected quality of a peer Result --The layered approach is able to fully utilize the marginal outbound bandwidth of supplying peer, and more adapted to the bandwidth asymmetric 12/6/2018
Performance Evaluation (2) Longer buffer enables a supplying peer to help more later-coming peers by prolonging the supplying chain Further increasing buffer size has very little help at prolonging the supplying chain Request chain (tree) in both cases Layered approach relieves the server bandwidth request with peer bandwidth 12/6/2018
Fairness Outbound/inbound < 1 40% Ethernet Peers are not fully satisfied Reason: the limiting inbound of the Modem/ISDN, and Cable Modem/DSL peers can not satisfied the Ethernet Peers 12/6/2018
Robustness Robustness -- 50% of the supplying peers depart early before the playback is finished -- Reconfiguration through buffer -- Failure ratio is the percentage of failed peers among all departure peers 12/6/2018
Conclusion for the layered Peer-To-Peer Streaming Be optimal at maximizing the streaming quality of heterogeneous peers Be scalable at saving server bandwidth Be efficient at utilizing bandwidth resource of supplying peers Evaluation -- Whether establishing fairness among peers, in terms of streaming quality satisfaction and bandwidth contribution -- Whether being robust against unexpected peer departures/failures 12/6/2018
Supporting Large-Scale Live Streaming Applications Key requirements -- Resource constraints -- Stability -- Efficient overlay structure Live Streaming Workload -- Large scale: the peak group size is 1,000 to 80,000 hosts -- A large number of short participations -- Heavy tail with some very long participations 12/6/2018
Bandwidth Resource Constraints Single Tree Protocols -- Resource Index: -- Trace study shows sufficient bandwidth resource Multiple Tree Protocol -- Increase the overall resilience -- Tightly coupled with specialized video encoding -- Resource Index: SupplyOfBW/DemandOfBW -- Increase the supply of the resources 12/6/2018
Stability (1) Metrics Simulation of single tree Simulation Results -- Mean interval between ancestor change for each participation -- Number of descendants of a departing participation Simulation of single tree -- Host join: asks the source to get m current group members, picks one host as parent -- Host leave: all of its descendants pick one host -- Parent Selection Algorithms: Oracle; Longest-First; Minimum depth; Random Simulation Results -- Oracle is the best -- Minimum depth tree can provide good performance 12/6/2018
Stability (2) Simulation Results -- Oracle is the best -- Minimum depth tree can provide good performance 12/6/2018
Stability (3) Impact of Multiple-Tree Protocols Simulation result -- Independent trees -- Load balancing -- Preemption Simulation result -- More frequent ancestor changes -- Improved performance comes at a cost of more frequents disconnects, more protocol overhead, and more complex protocols 12/6/2018
Efficient overlay structure (1) Overlay structure closely reflects the underlying IP network -- Need to discover other nearby hosts as parents -- Partition hosts into clusters -- One member of each cluster is designated as the clustered head -- Hosts in the same cluster maintain knowledge about one another Clustering Quality Metric -- Average and maximum intra-cluster distance in milliseconds 12/6/2018
Efficient overlay structure (2) Sensitivity to Number of Clusters -- More clusters smaller intra-cluster distance -- Maximum intra-cluster distance more sensitive to the change of number of clusters 12/6/2018
Efficient overlay structure (3) Sensitivity to Cluster Size and Resource Maintenance -- Bounding the cluster size doesn’t significantly affect the intra-cluster distances 12/6/2018
Conclusion for large-scale live streaming applications with dynamic application end-points Minimizing depth in single-tree protocols provides good stability performance Multiple-tree protocols can significantly improve the quality of streams Simple clustering techniques improve the efficiency of the overlay structure Opening issue: encourage application end-points to contribute their resources is an important direction 12/6/2018
Incentive Mechanism for Peer-to-Peer Media Streaming (1) System quality is: T is the total number of the packets in a streaming session, is 1 if the packet i arrives at the receiver before its scheduled play-out time, and 0 otherwise Cooperation brings quality Simultaneous uploading hurts quality 12/6/2018
Incentive Mechanism for Peer-to-Peer Media Streaming (2) Random peer selection provides random quality 12/6/2018
Score-based incentive mechanism Peer selection scheme allows a user to select peers with equal or lower rank to serve as suppliers A user wishes to receive better-than-best-effort streaming, it must earn a positive score by contributing to the system The stream quality for a receiver can be expressed as a function of contribution, score, or rank 12/6/2018
Functions Scoring function: could be: Contribution cost: Rank Computation: Quality function: 12/6/2018
Experiment system 12/6/2018
Performance evaluation Expected rate: the total bytes coming from all senders The gain increases for the incentive when the K increases When k>20, the difference of the rates decreases because the bottleneck is shifted from the hosts to the network Packets the miss their play-out deadlines are considered as lost 12/6/2018
Quality of Streaming 12/6/2018
Conclusion for incentive mechanism for Peer-to-Peer Media Streaming Motivation -- The stream quality is poor if the level of cooperation is low -- Cooperation from a few altruistic users cannot provide high quality streaming to its users in a large system Conclusion -- A rank-based incentive mechanism achieves cooperation through service differentiation -- The contribution of a user is converted into a score, then the score is mapped into a rank, and the rank provides flexibility in peer selection that determines the quality of a streaming session -- Cooperative users earn higher rank by contributing their resources to others, and eventually receive high quality streaming 12/6/2018
Conclusion Application layer multicasting Consuming the other end node’s resource while sharing own resource out The differentiated quality is realized with replicated streaming, layered streaming, and MDC Replicated streaming is used at the single tree delivery In the single tree, the minimize depth algorithm shows good performance Layered Streaming and MDC with multiple tree delivery increases resource, and improve the stability as well Cluster can improve the efficiency of the overlay structure Fairness is still an open issue Incentive mechanism is a solution to encouraging resource sharing 12/6/2018
Reference [1] Layered Peer-to-Peer Streaming [2] A Comparison of Layering and Stream Replication Video Multicast Schemes [3] Receiver-Driver layered Multicast [4] Internet Multicast Video Delivery [5] ZIGZAG: An Efficient Peer-to-Peer Scheme for Media Streaming [6] SplitStream: High-bandwidth content distribution in cooperative environment [7] The Feasibility of Supporting Large-Scale Live Streaming Applications with Dynamic Application End-Points [8] Incentive Mechanism for Peer-to-Peer Media Streaming 12/6/2018
Appendix: Receiver-Driven Layered Multicast Rate-adaptation protocol Each receiver runs the control loop: -- On congestion, drop a layer -- On spare capacity, add a layer Join-experiment -- adding layers at “well-chosen” times -- causing congestion, then the receiver drops the adding layers -- successful, the receiver start adding another join-experiment Exponential Join timer for RLM adaptation at the join experiment “Sharing learning” in multiple receivers for scaling of the receiver 12/6/2018