Giving life to today s media distribution services FIA - Future Internet Assembly Athens, 17 March 2014 Presenter: Nikolaos Efthymiopoulos Network architecture & Management Group Copyright University of Patras, Greece FIA 2014 University of Patras - Greece 1
The future that we want! No borders FREE!!! Stability Be the center Personalized content discovery FIA 2014 University of Patras - Greece 2
The present that we have! Video on demand services (i.eyoutube) Users upload and consume for free Lexicographic search and recommendation by exploiting user s history No live streaming No personalized search and recommendation No social interaction between users Social networks (i.e. Facebook) Free-global scale entrance Personalized recommendation by exploiting posts of friends No live streaming No search and recommendation tools for media objects Closed video on demand platforms (i.e. ISP platforms) High video quality No social interaction Small market penetration because of high cost FIA 2014 University of Patras - Greece 3
The research that connects them! User information exploitation Users built dynamic social relationships Users built dynamic media object consumption patterns Interaction with other users triggered by specific media objects events The objectives are: To exploit all this information for personalized and advanced searching and recommendation. To develop a personalized advertisement system that could act as the funding source for such kind of systems. User information exploitation User resources exploitation Better & Innovative services FIA 2014 University of Patras - Greece 4
The research that connects them! User resources exploitation (a global pool of resources) Users are able to store and distribute the media objects that consume Users can act as social helpers and facilitate the distribution of media objects that their social relationships consume. The objectives are: Scalable optimization and full exploitation of the dynamic and heterogeneous user resources Scalable monitoring and control of these resources and dynamic playback rate adaptation according to them. User information exploitation User resources exploitation Better & Innovative services FIA 2014 University of Patras - Greece 5
The research that connects them! Better and innovative services Users want to broadcast Users want to consume in parallel multiple audiovisual streams Users want real time audiovisual interaction between them Users want to advertise media objects or other information Users want to generate new media objects by process the exiting set All these at global scale, stable and free. User information exploitation User resources exploitation Better & Innovative services FIA 2014 University of Patras - Greece 6
Social aware live streaming architecture Objective is to organize user (social) resources towards a scalable, efficient and stable live streaming. A set of users that want to consume a live stream. A server that acts as a tracker (service entrance point). A set of servers or social helpers acting as auxiliary bandwidth providers. FIA 2014 University of Patras - Greece 7
Functionalities of Social aware Live Streaming I Media Distribution Graph Neighbor List Distributed Block Transmission Scheduler Auxiliary Network Resources Distr. Bandwidth Control Algorithm Video Player P2P Congestion Control Service Performance Measurements Network Measurements Scalable Bandwidth Monitoring Media blocks are exchanged among participating peers and each peer runs a Distributed Media Block Scheduler. It collaboratively ensures the timely delivery of every media block to every participating node. FIA 2014 University of Patras - Greece 8
Functionalities of Social aware Live Streaming II Media Distribution Graph Neighbor List Distributed Block Transmission Scheduler Auxiliary Network Resources Distr. Bandwidth Control Algorithm Video Player Video Stream P2P Congestion Control Service Performance Measurements Network Measurements Scalable Bandwidth Monitoring The MDG is the graph in which each node is a user or a media server (cloud) and each edge bidirectional transmission of media blocks. The objective of the MDG is to determine the set of nodes with which each node exchanges media blocks in order to dynamically exploit available bandwidth. FIA 2014 University of Patras - Greece 9
Functionalities of Social aware Live Streaming III Media Distribution Graph Neighbor List Distributed Block Transmission Scheduler Auxiliary Network Resources Distr. Bandwidth Control Algorithm Video Player Video Stream P2P Congestion Control Service Performance Measurements Network Measurements Scalable Bandwidth Monitoring The network traffic in P2P live streaming is headed towards multiple network destinations (nodes) and is composed of small media blocks. P2P Congestion Control is the functionality that controls the injection rate of this new type of traffic in a stable and efficient way. FIA 2014 University of Patras - Greece 10
Functionalities of Social aware Live Streaming IV Media Distribution Graph Neighbor List Distributed Block Transmission Scheduler Auxiliary Network Resources Distr. Bandwidth Control Algorithm Video Player Video Stream P2P Congestion Control Service Performance Measurements Network Measurements Scalable Bandwidth Monitoring Scalable Bandwidth Management is the functionality that monitors, in a scalable, accurate and dynamic way, the total available upload bandwidth of the peers that participate in each MDG (streaming object). FIA 2014 University of Patras - Greece 11
Functionalities of social aware LS V Media Distribution Graph Neighbor List Distributed Block Transmission Scheduler Auxiliary Network Resources Distr. Bandwidth Control Algorithm Video Player Video Stream P2P Congestion Control Service Performance Measurements Network Measurements Scalable Bandwidth Monitoring Social aware Bandwidth Provisioning System is the functionality that, in case that there is a deficit in the upload bandwidth in an MDG, determines the set of nodes that will enhance the bandwidth resources of this MDG. Additionally, it determines the set of nodes with which each node will interact. FIA 2014 University of Patras - Greece 12
Social aware Media Distribution Graph -Introduction The objective of the MDG is to determine the set of nodes with which each node exchanges media blocks. GOAL is to implement a self-managed graph that fast and without and centralized coordination is able to: 1. Exploit heterogeneous available bandwidth from various nodes. 2. Distributes the stream fast. 3. Adapts dynamically to peer arrivals and departures. 4. Exploits bandwidth resources from social helpers. 5. Connect each node (user and/or server) with nodes close to the underlying network. In order to achieve these we need: 1. A graph structure 2. Distributed, fault tolerant and fast algorithms that guarantee this structure in dynamic conditions. FIA 2014 University of Patras -Greece 13
Social aware Media Distribution Graph Intra DOMA FIA 2014 University of Patras -Greece 14
Social aware Media Distribution Graph Inter DOMA FIA 2014 University of Patras -Greece 15
Social aware Media Distribution Graph DOMA Objectives Requirements that DOMA has to fulfill: Fast Execution (avoid long handshakes and acks) Low Overhead (minimum message exchanges) Tolerant to packet losses (recovery states) Tolerant to graceful and ungraceful peer departures (timers) Parallel execution of multiple DOMA simultaneously from multiple nodes (locks) We are exploiting theory of distributed algorithms to achieve all these! FIA 2014 University of Patras -Greece 16
P2P congestion control Problem statement Challenge: Design of a congestion control suitable for network traffic that: 1.is non persistent 2.is heading to multiple network locations Sender peer B5 B4 ack2 Bottleneck network point -u(t) B2 d2 Receiver peer 2 B1 d1 ack1 Receiver peer 1 Other dynamic traffic ack3 B3 d3 Receiver peer 3 1. Di Delay of network path I 2. u(t) Dynamic of bottleneck network point 3. Bi P2P video block i 4. acki Acknowledge data from receiver i FIA 2014 University of Patras -Greece 17
P2P congestion control Objectives Will be able to optimally exploit the available bandwidth of Home Gateways. Will be stable and robust under time-varying delays and dynamic underlying network traffic. Will play fair, in terms of network resources that they allocate, to the other traditional TCP flows that exist in the underlying network. Will avoid unnecessary high latencies and retransmissions FIA 2014 University of Patras -Greece 18
P2P congestion control Architecture Packet Fragmentation Engine Fragments P2P media blocks to UDP packets. Distributed receiver feedback Metrics from each receiver that facilitate the estimators at each sender of: upload capacity, idle resources, path delays. Other traffic detector Detects other (i.e. TCP traffic) and informs the sending controller in case that other traffic exists. Sending Controller The major part (control algorithm) that determines the packet injection rate according to the output of the 2 previous components. FIA 2014 University of Patras -Greece 19
Scalable bandwidth monitoring and social aware bandwidth provision Problem statement A way to monitor and control the total bandwidth resources that each Media Distribution Graph (channel) has. Dynamic offered bandwidth: sum of upload of participating peers Dynamic requested bandwidth: Number of participating users multiplied with playback rate FIA 2014 University of Patras - Greece 20
Scalable bandwidth monitoring and social aware bandwidth provision Objectives Bandwidth monitoring: Scalable that doesn t depend at all in a central entity. Accurate in order to allow us to have a stable control algorithm. Very dynamic in order to be adaptable to peer arrivals and departures and to sudden bandwidth changes. Bandwidth provision algorithm: Stable in order to ensure QoS. Efficient in terms of the bandwidth resources that remain idle. Robust to tolerate inaccuracies in the measurements of the monitoring algorithm. University of Patras - Greece 21
Scalable bandwidth monitoring and social aware bandwidth provision Scalable bandwidth monitoring is a TOTALLY distributed protocol that has four steps: 1. Each normal peer selects dynamically its monitoring peer (through information that MDG provides). 2. Sends to it dynamically information relevant with its: upload, idle resources, hit ratio. 3. Each monitoring peer according to a control algorithm calculates the deficit or the surplus of resources. 4. Each social helper allocates/ dis-allocates bandwidth resources according to the control algorithm. FIA 2014 University of Patras -Greece 22
Scalable bandwidth monitoring and social aware bandwidth provision Architecture Peer monitoring process: Each peer derives the metrics that required from the control algorithm. Data mining protocol: Through a network protocol the metrics are gathered in the control points. Bandwidth control architecture: The control algorithm is executed and the components which concerned about the output are triggered. Dynamic playback rate adaptation: Playback rate is adapted according to the output of the control algorithm and the media encoding protocol. Social aware bandwidth discovery : A database where socially correlated peers are gathered. Bandwidth provisioning: Bandwidth is allocated. FIA 1st 2014 EC Review, Brussels, March 5, 2014 University of Patras -Greece 23
System s behavior in a nutshell FIA 2014 University of Patras -Greece 24
Thank you!!! Questions? Comments? Critique? Proposals? Contact info Nikolaos Efthymiopoulos: nefthymiop@ece.upatras.gr Spyros Denazis: sdena@upatras.gr FIA 2014 University of Patras - Greece 25