1 / 14

Analysis of Different Models for Optimal P2P Content Distribution Network

Analysis of Different Models for Optimal P2P Content Distribution Network. Subhamoy Ghosh sghosh@cc.hut.fi. Agenda. Introduction The Next – Step Problem Analysis of the Next – Step Problem Chunk – based CDS Analysis of Chunk – based CDS Discussion – Other Models: Merits/Demerits

lev-weber
Download Presentation

Analysis of Different Models for Optimal P2P Content Distribution Network

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Analysis of Different Models for Optimal P2P Content Distribution Network Subhamoy Ghosh sghosh@cc.hut.fi

  2. Agenda • Introduction • The Next – Step Problem • Analysis of the Next – Step Problem • Chunk – based CDS • Analysis of Chunk – based CDS • Discussion – Other Models: Merits/Demerits • Conclusion • References

  3. Introduction • P2P systems – • designed to share computer resources by direct exchange, w/o any central authority. • Application overlay network on top of conventional protocols. • Fault-tolerant, reliable, scalable massive content distribution system. • Only possible option to manage flash crowds. • Content Distribution – • Has been the main focus of P2P research. • Responsibility is spread amongst all downloader(s). • Clients contribute resources of the system as a function of their upload capacities.

  4. Introduction • Performance perceived by user – • Time taken to fulfill request for a particular file. • Download time as the dominant factor. • Generic trend for file dissemination – • Partition the file to N equal parts. • User can download either from server or from a peer who has the requested chunk.

  5. The Next-Step Problem • Source node s divides a file into k equal parts to disseminate amongst n peers. • Assumptions: • Communications are synchronous. • Ignores link latencies • File reaches the destination in the same round, as the departure. • Graph G = (V,E) with |V| = n, W:VxV--->R • File is divided into k-parts such that: • Next – Step problem calculates the actions of each node, represented by Action Matrix:

  6. The Next-Step Problem • Max-Product Belief Propagation Network – • Calculates the MAP estimates of the nodes. • Provides a vector of node beliefs and selects the action with the highest probability. • Algorithm converges, when a node receives duplicate packets for two consecutive rounds from all nodes. • With the Next-Step Solution and a pre-defined Cost function, the problem reduces to:

  7. Analysis of Next-Step Problem

  8. Optimizing Chunk-Based CDS • Provide a network model independent lower – bound on the time taken for file dissemination in a chunk – based CDS. • Makespan: Time taken to disseminate M parts of a file to I peers in a centralized scenario. • Assumptions: • Peers upload only 1 file in each round. • Upload capacities are considered to be equal. • Lower bound on minimum number of rounds for any CDS is: • Delay Stretch: Captures the delay in download of a peer, for sharing chunks with other peers. Maximum delaystretch is calculated as:

  9. Analysis of Chunk – based CDS Scatter plots of data with M = 10 and M =50 [8]. • 100 independent simulations with N = 2, 4, …215. • In each case of M = 1 – 5, 8,10, 15, 20 ,50 fits a linear model.

  10. Discussion – Other Models: Merits/Demerits • Maximize peer – bandwidth utilization: • P2P is scalable since bandwidth increases with increase in participants. • Models focus on i) peer selection strategy – self-organization, ii) chunk selection strategy - duplication, iii) network degree – choose in-degree/out-degree of client based on upload/download capacity. • Uniform Workload Distribution: • Fairness in workload minimizes average download time. • Incentive – driven P2P Networks: • Minimizes download time subject to budget constraint. • Network Coding based P2P CDS: • Improves delay in transmission time, and makes network more robust and adaptive. • But almost all models suffer from the churning problem, and network coding models do not perform well, in case multiple nodes fail to reach every client.

  11. Conclusion • Loopy BP algorithm - • Approximate solution in case of cycled graphs, • Currently the experiment is terminated after a user-defined threshold of operations. • The authors in [1] consider to use gradient – descent algorithm here instead, to overcome this fluctuation in local optimum. • Pre-emptive operations in belief propagation - • improve on the computational delay and the final response on the choice of nodes. • The generic file dissemination approach - • equal service capacity for all peers, and the model can be extended to consider the case of selfish peers. • However it provides a model independent performance benchmark, that can be used to compare the dissemination times of different overlay networks.

  12. References • Bickson Danny, Dolev Danny, Weiss Yales, Efficient Peer-to-Peer Content Distribution Network, School of Computer Sc. & Engg, Hebrew University of Jerusalem. • Zheng Xiaoying, Cho Chunglae, Xia Ye, Optimal Peer-to-Peer Techniques for Massive Content Distribution, IEEE INFOCOMM (2008). • Felber Pascal, Biersack W. Ernst, Cooperative Content Distribution: Scalability Through Self-Organization, LNCS 3460, pp 343-357 (2005). • Li Jin, Chou A. Philip, Zhang Cha, MutualCast: An Efficient Mechanism for Content Distribution in a P2P Network, ACM SIGCOMM Asia ‘04, Beijing, China (2004). • Hamra Al Anwar, Felber A. Pascal, Design Choices for Content Distribution in P2P Networks, ACM SIGCOMM, Vol 35, pp 29-40 (2005). • Tewari Saurabh, Kleinrock Leonard, On Fairness, Optimal Download Performance and Proportional Replication in P2P Networks, LNCS 3462, pp 709 – 717, IFIP (2005). • Adler Micah, Kumar Rakesh, Ross Keith, Rubenstein Dan, Suel Torston, Yao D. David, Optimal Peer Selection for P2P Downloading and Streaming, IEEE INFOCOMM, Vol. 3, pp 1538 – 1549 (2005). • Kangasharju Jakko, Kangasharju Jussi, An Optimal basis for Efficient P2P Content Distribution Algorithms, 15th . Intl. Conf. on Computer Communications and Networks, Washington D.C. (2006).

  13. References • Mudinger Jochen, Weber Richard, Weiss Gideon, Optimal Scheduling of Peer-to-Peer File Dissemination, Mathematical Performance Modeling and Analysis (MAMA) (2006). • Loginova Oksana, Lu Haibin, Wang X. Henry, Peer-to-Peer Networks: A Mechanism Design Approach, JEL Classification, Working Papers, Dept. of Economics, University of Missouri (2007). • Androutsellis – Theotokis Stephanos, Spinellis Diomedis, A Survey of Peer-to-Peer Content Distribution Technologies, ACM Computing Surveys, Vol. 36, pp 335 – 371, (2004). • Small Tara, Li Baochun, Liang Ben, On Optimal Peer-to-Peer Topology Construction with Maximum Peer Bandwidth Contributions, Biennial Symposium on Communications, Kingston, Canada (2006).

  14. Thanks to all !

More Related