* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download slides - network systems lab @ sfu
Cracking of wireless networks wikipedia , lookup
Piggybacking (Internet access) wikipedia , lookup
Network tap wikipedia , lookup
Computer network wikipedia , lookup
Deep packet inspection wikipedia , lookup
Recursive InterNetwork Architecture (RINA) wikipedia , lookup
Airborne Networking wikipedia , lookup
Zero-configuration networking wikipedia , lookup
UniPro protocol stack wikipedia , lookup
Opportunities and Challenges of Peer-to-Peer Internet Video Broadcast J. Liu, S. G. Rao, B. Li and H. Zhang Presented by: Yan Ding Proc. of The IEEE, 2008 Outline Introduction Architectural choices – Router-based: IP Multicast – Non router-based: Overlay networks (CDNs, P2P) Peer-to-Peer Video Broadcast – Tree-based Construction (ESM) – Data-driven randomized Construction (CoolStreaming) Challenges – Technical issues and deployment challenges Conclusion 2/16 Introduction Internet Video Broadcast/Multicast – Real-time performance requirements: higher bandwidth and lower latency – More challenging than file download, voice over IP – Can use IP Multicast or Overlay networks IP Multicast – S. Deering and D. Cheriton, “Multicast routing in datagram internetworks and extended LANs”, ACM Transaction on Computer Systems, vo. 8, no. 2, pp. 85-110, May 1990. – Scalability concern, slow deployment, cannot support for higher level functionality Peer-to-Peer Multicast – Cost-effective and easy to deploy – Has potential to scale: peer severs as both server and client – Tree-based and Data-driven approaches 3/16 Architectural choice— Router-based IP Multicast Network-layer solution – Routers responsible for multicasting • Group membership: remember group members for each multicast session • Multicast routing: route data to members – Efficient bandwidth usage • network topology is best-known in network layer Figures are from Chu et al. “A case for end system multicast”,in Proc. ACM SIGMETRICS’00, June 2000. 4/16 IP Multicast Requires per-group state in routers – Maintenance produces high complexity, especially in core routers – Scalability concern – Violate end-to-end design principle: ‘stateless’ Slow deployment – Changes at infrastructural level – IP multicast is often disabled in routers Difficult to support higher layer functionality – E.g., Error, flow control and congestion control 5/16 Architectural choice— Non Router-based Overlay Network Application-layer solution – Multicast functionality in end systems – End system participate in multicast via an overlay structure – Overlay consists of application-layer links – Application-layer link is a logical link consisting of one or more links in underlying network Figures are from Chu et al. “A case for end system multicast”,in Proc. ACM SIGMETRICS’00, June 2000. 6/16 Overlay Network Performance penalties – Produce redundant traffic on physical link: multiple overlay edges use the same physical link – Increase latency: communication involves other end systems Fast deployment – Unicast IP service: all packets are transmitted as unicast packets – Requires end system to perform additional processing Enable higher layer functionality – Leverage unicast solutions – Exploit application-specific intelligence Used by both CDNs and pure P2P systems 7/16 P2P Multicast Tree-based – – – – – – Peers organized into trees for delivering data Parent-child relationships Push-based Maintain and repair tree when nodes join, leave or fail Failure of higher nodes disrupt many users Uplink bandwidth not utilized at leaves • Data can be divided and disseminated along multiple trees – Example: End System Multicast (ESM) • Y.-H. Chu, S. G.Rao, and H. Zhang, “A case for end system multicast”,in Proc. ACM SIGMETRICS’00, June 2000. 8/16 End System Multicast (ESM) Main idea – Construct a good overlay structure (mesh) – Construct spanning trees of the mesh, each rooted at the corresponding source Good overlay structure – Path quality: between any pair of members is comparable to that of the unicast path between them (e.g. delay, bandwidth) – Overhead control: each member has a limited number of neighbors Construct spanning trees – Use standard routing algorithms to construct per-source (reverse) shortest path spanning trees for data delivery 9/16 ESM (1/2) Maintain information – A list of members in the group (randomly chosen) – Path from source to itself Update information – Changes due to dynamics (node join, leave or fail) – Periodically exchange with its neighbors Dynamics – Member join • Contact the source and get a few active members • Request to be neighbors and select one as parent (parent selection) – Member leave or fail • Notify its neighbors before leaving • No refresh message received: member fail (need parent re-selection) 10/16 ESM (2/2) Parent selection – When to select: • node (say, A) newly join or current parent leave, fail or performs poorly – Who to select: • neighbors that have low delay or haven’t been probed – How to select: • Degree-saturated or not • Descendant or not • Performance (throughput, delay) Failure of higher nodes disrupt many users and uplink bandwidth not utilized at leaves • Each sub-stream delivered via one overlay tree • Robust to the failure of an ancestor and utilize bandwidth of all nodes • Need specialized coding algorithm on receiver 11/16 P2P Multicast Data-driven – Use availability of data rather than explicit structure to guide data flow – Pull-based • Avoid redundancy – Periodically exchange data availability with random partners and retrieve new data • Robust to node failures – Real time constraints • Scheduling problem – Example: CoolStreaming • X. Zhang, J. Liu, B. Li and T.-S. P. Yum, “DONet/CoolStreaming: A datadriven overlay network for live media streaming”, in Proc. INFOCOM’05, Miami, FL, USA, March 2005. 12/16 CoolStreaming Membership Manager – Maintain information • A list of members in the group – Update information • Periodically generate membership message • Distribute it using Scalable Gossip Membership Protocol (SGAM) Partnership Manager – Partners are members that have expected data segments – Exchange Buffer Map (BM) with partners – Buffer Map contains information of availability of segment Scheduling – Determine which segment should be abtained from which partner – Get segment from partner and provide availble segment to partner 13/16 Challenges (1/2) Tree-based vs. Data-driven – Tree-based: face instability and bandwidth under-utilization – Data-driven: suffer latency-overhead tradeoff – Hybrid: identification and position set of stable overlay nodes Incentives and fairness – Free riders exist in p2p system – Need incentive mechanism for real-time requirements 14/16 Challenges (2/2) Support heterogeneous receivers – Scalable Coding, Multiple Descriptive Coding (MDC) – Low efficiency, bandwidth penalty, need powerful receiver Coding at peers? – Network coding improve througput – Tradeoff between coding efficiency and startup delay Deployment – Interests conflicts between network and service providers – Internet triumphs on development of new service – Service providers relay on dedicated networks 15/16 Conclusion Introduce two Internet achitectural choices to support Video broadcasting – IP Multicast: implement in network layer (router) – Overlay: implement in application layer (end system) Reviews the state-of-the-art of P2P technologies and examine two broad approaches – Tree-based: construct trees to deliver data – Data-driven: use availability info to guide data flow Discuss technical challengings and deployment issues 16/16