Distributed control and optimization of peer to peer video streaming systems

Abstract

Video distribution over Internet has already become enormously popular. It is a majorapplication due to users‘ growing demand and extraordinary growth of networktechnologies. Future media Internet will need to be able to distribute high qualityvideo contents in an efficient, supple and personalized way through dynamic andheterogeneous network environments. Live video streaming applications have gainedgreat popularity among users but exert great pressure on video servers and on theInternet traffic.Peer-to-Peer (P2P) networks have drawn great attention from both the researchcommunity and the industry as they provide an attractive solution for video streamingdue to their low cost and high scalability. A major advantage in using a P2Parchitecture is that each peer which involved in a content delivery contributes with itsown resources to the streaming distribution. In a P2P live streaming system a peer notonly downloads data from the network, but also uploads the downloaded data to otherusers in the network. In this way the uploading bandwidth of end users is efficientlyutilized to reduce the bandwidth burden which is placed otherwise on the servers.In live video streaming applications all the users play a video in approximatesynchronicity with the video source and download the video at its playback rate. Themajor objective of a P2P live video streaming system is to distribute video blocksfrom the video source to all the peers in a way that all the peers receive all the packetsbefore their playback deadlines and each peer receives each packet only once.Subsequently, the major requirements for a P2P live video streaming application canbe summarized to the followings: Efficiency of the video distribution in terms of utilization of peers‘ uploadbandwidth, in order to minimize any additional bandwidth contributed by aset of media servers (cloud) and/or maximize the playback rate of the videowhich the system is able to deliver. Efficiency has a direct impact on thetrade-off between bandwidth costs and the video quality that the system isable to distribute. Stability of the system which is defined as the uninterrupted and completestream delivery in each participating peer in the presence of dynamicconditions (unrelated network traffic, bandwidth changes, peer arrivalsand/or departures), which affect the amount of the available uploadbandwidth of the participating peers. These dynamic conditions may haveserious impact on the quality of experience (QoS) perceived by the users. Scalability which is determined by the amount of resources (bandwidth,storage, processing overhead) that the media server/cloud, which managesthe system, has to contribute in order to sustain the uninterrupted deliveryof the stream, as the number of participating peers grows. The design of ascalable system requires low amount of resources even if the number ofparticipating peers is high.Based on the recent advances of the research community, there are two strategies inorder to adapt the P2P live streaming service to the dynamic upload bandwidthconditions of participating peers. The first is to dynamically adapt the playback rateaccording to the existing upload bandwidth of participating peers, while the second isto dynamically allocate upload bandwidth from auxiliary sources (e.g. clouds). Theselection of a strategy has to do with the QoE that participating users desire and thebusiness model of the service provider. In case that users and the service providerdesire a costless live steaming service the first strategy has to be selected. At the otherhand, in case that they desire a live streaming service with high QoE the secondstrategy has to be selected.In this dissertation is presented a P2P live video streaming system which is scalableand stable. The proposed system is able to guarantee the complete and on time videodistribution to every participating peer based on the two aforementioned strategies.The contribution of the dissertation is summarized to the development of thesestrategies with respect to the aforementioned P2P live streaming requirements.The system consists of a media server in a cloud and a set of peers connected viaInternet. The video, with constant or dynamic playback rate, is divided into videoblocks. The function of the system can be described through four basic components(the P2P Overlay, the Distributed Block Transmission Scheduler (DBTS), theScalable Bandwidth Monitoring (SBM) and the QoS Enabler). In order to allow peers to exchange video blocks each peer maintain network connections with a small subsetof other peers, noted as neighbors. These connections can change dynamically andform a dynamic graph called the P2P Overlay. Video block exchanges are coordinatedby the Distributed Block Transmission Scheduler (DBTS), which is comprised by aset of algorithms that executed, by every peer which dynamically communicates withits neighbors. The Scalable Bandwidth Monitoring is a centralized component in thecloud, that applies a monitoring protocol, which aggregates all the requiredmonitoring information and forms respectively all the required metrics that the QoSEnabler needs. According to the selected strategy, QoS Enabler is responsible eitherto dynamically determine the desired video playback rate in case of the first strategyor to dynamically calculate and add/remove the amount of surplus/deficit in case ofthe second strategy. Then, video blocks are generating and passing to DBTS in orderto be diffused to all participating peers.In the first control strategy that has been developed, it is proposed an architecturewhich adapts dynamically the video playback rate according to the existing uploadbandwidth of participating peers. It is a low cost technique that guarantees the smoothoperation of the system even in cases where the total available bandwidth is low orchanges very abruptly. Towards this goal it is developed a scalable idle bandwidthmonitoring mechanism and a playback rate control mechanism. More precisely, in thisdissertation, is modeled analytically the relationship between the dynamic playbackrate and the available bandwidth resources through difference equations and is appliedcontrol theory. In order to stabilize the system, are quantified the monitoringinaccuracies and dynamic bandwidth changes and is calculated analytically theamount of bandwidth resources that need to be kept idle, by lowering the playbackrate, in order to ensure the complete distribution of the stream. The system isevaluated with OPNET Modeler packet level simulator and is certified that: Provides a scalable monitoring, in terms of bandwidth and processingoverhead that costs to the live streaming service provider, by exploiting thebalancing properties of the P2P overlay and the distributed block transmissionscheduler for P2P live streaming. Ensures the uninterrupted stream distribution even in a highly dynamicenvironment as it is based on an analytical model that is designed and factorizes the disturbances which a dynamic environment introduces to a P2Plive streaming system and exploits modern control theory to mitigate theireffects. Utilizes efficiently the upload bandwidth of participating peers by calculatinganalytically the maximum playback rate that the system is able to deliver as afunction of the system's resources, the accuracy of the measurements and themaximum possible disturbance on system's resources.In the second control strategy that is developed, it is proposed an architecture thatdynamically calculates and allocates/releases from/to auxiliary sources (e.g. clouds)the amount of deficit/surplus of bandwidth which is required for the video distributionwith respect to the video playback rate and system's resources. Towards this goal, isdeveloped a scalable idle bandwidth monitoring gossip protocol and a bandwidthallocation control mechanism. More precisely is modeled analytically the relationshipbetween the total bandwidth surplus/deficit and peer's bandwidth utilization throughdifference equations and is applied control theory. In order to stabilize the system, arequantified the monitoring inaccuracies and dynamic bandwidth changes and iscalculated analytically the amount of bandwidth resources that should be additionallyallocated or should be kept idle (in case of surplus) in order to guarantee theuninterrupted distribution of the stream. The system is evaluated with OPNETModeler packet level simulator and is certified that: Provides a scalable, in terms of bandwidth and processing overhead that coststo the cloud, dynamic and fault tolerant monitoring, by exploiting thebalancing properties of: i) the P2P overlay, ii) the distributed blocktransmission scheduler for P2P live streaming and, iii) an innovative gossipprotocol. Ensures the uninterrupted stream distribution even in a highly dynamicenvironment as it is based on an analytical model that is designed, whichfactorizes the disturbances that a dynamic environment (underlying networkand peer behavior) introduces to a P2P live video streaming system andexploits modern control theory to mitigate their effects. Utilizes efficiently the upload bandwidth of participating peers by calculatinganalytically the amount of bandwidth (hence the equivalent in surplus/deficit) that guarantee the successful distribution of the stream as a function of thesystem's resources, the accuracy of the measurements and the maximumpossible disturbances in the system's resources.Despite that the two aforementioned strategies guarantee a stable P2P live streamingservice, require the existence of a centralized management component that willaggregate the required monitoring information and will apply the appropriate controlstrategy. Motivated by this fact is attempted a P2P live streaming architecture thatwith a not only scalable but also totally distributed way, will be able to determinedynamically the required bandwidth (hence the equivalent in surplus/deficit) for thevideo distribution. At the same time if necessary subsidiary help, it will no longercome from a central source (e.g. cloud) but from other (additional and/ or unused)peers. More precisely is designed and implemented an innovative monitoringbandwidth and control mechanism that: It is scalable as it follows a distributed (bottom up) monitoring architecturethrough an innovative gossip protocol and a distributed control strategy. It is stable as it is based on an analytical model which is stabilized by aproposed control strategy. It efficiently utilizes the upload bandwidth of participating peers.2. Structure of the DissertationThis dissertation is structured in six chapters. The first chapter consists the state of theart about advances in P2P live video streaming. Firstly, is presented the videostreaming classification and the current architectural approaches for live videostreaming. Subsequently, there is an analysis for P2P networks as they consist awidespread approach for live video streaming implementation. Afterwards, isanalyzed the live video streaming service over P2P networks. Then, is presented anevaluation framework with which it can be certified the performance of a P2P livevideo streaming system. It follows a short reference on mobile P2P live videostreaming. Finally, there is an analysis referred to the stability on P2P live videostreaming which acted as the motivation for this dissertation.The second chapter presents the original (proposed) P2P live video streaming system'sarchitecture. There is a brief analysis for the whole system and a further detailedanalysis for the P2P Overlay and for the Distributed Block Transmission Scheduler, asare both two basic parts of the system regardless of the control strategy thatimplemented.In the third chapter is proposed an original architecture that enhance the P2P livevideo streaming service by adapting dynamically the video playback rate with respectto system's total available bandwidth resources. In this way the system guarantees itssmooth and uninterrupted operation with low cost even in cases where changes to thetotal available bandwidth are very abrupt.In the fourth chapter is proposed an original cloud assisted P2P live video streamingarchitecture. In this way the system guarantees QoS in the video distribution with theminimum amount of bandwidth overprovision. Thus is achieved QoS at the lowestpossible cost.In the fifth chapter is presented an original P2P live video streaming architecturewhich is not only scalable but also totally distributed. Additionally, if necessarysubsidiary help, it no longer comes from a central source (e.g. cloud) but from other(additional and/ or unused) peers.The sixth chapter highlights all the conclusions of the dissertation and some aspectsfor possible future work.Appendix presents the bibliography and the publications of this dissertation atinternational journal and conferences.