Download presentation
Presentation is loading. Please wait.
1
BitVampire: P2P On-Demand Media Streaming in the Internet http://www.cs.ubc.ca/~liu/bitvampire http://www.cs.ubc.ca/~liu/bitvampire PhD Student: Xin Liu Supervisor: Dr. Son T. Vuong {liu, vuong}@cs.ubc.ca j Networking and Internet Computing (NICLab) University of British Columbia March 27, 2006
2
2 Outline 1.Introduction 2.System Design 3.Simulation 4.Prototype Implementation 5.Future Work
3
3 COOLSearch-BitVampire a novel technology for P2P content search and on-demand media streaming On-Demand Media Streaming received intensive consideration due to the proliferation of high bandwidth networks. Two main categories of existing on-demand media streaming systems: Server-based systems Proxy-based systems The cost of deploy and maintain servers/proxies could be very expensive. We propose a novel architecture that exploits the often under- utilized peers ’ resources to support large-scale on-demand media streaming services. 1. Introduction – Motivation
4
4 Basic Ideas/Contributions 1. Published video is split into segments and distributed to different peers (Media Segment Distribution MSD algorithm). 2. When watching a video, the requesting peer (receiver) searches the corresponding segments (COOLSearch), then 3. Aggregates bandwidths from multiple peers to stream the video (Multi-Source Scheduling MSS Algorithm).
5
5 2. System Design Example of watching a video P1 (Seg #0) P6 (Receiver) Playback bit rate: 500kbps P2 (Seg #0) P3 (Seg #0, #1) P4 (Seg #1, #2) P5 (Seg #2) 250kbps 200kbps 100kbps 300kbps 200kbps P6 wants to watch a video, it searches the Seg #0 and finds that P1, P2, P3 have Seg #0. P6 aggregates bandwidths from P1, P2 and P3 to stream the Seg #0. When streaming session of Seg #0 is almost over, P6 searches Seg #1. P6 aggregates bandwidths from P3 and P4 to stream Seg #1. When streaming session of Seg #1 is almost over, P6 searches Seg #2. P6 aggregates bandwidths from P4 and P5 to stream Seg #2.
6
6 COOLSearch (Category Overlay)
7
7 A simple category overlay network
8
8 Overlay Networks - Three Categories
9
9 Large-Scale Category Overlay Network
10
10 Need An Efficient Scheduling Algorithm To fully use the bandwidths from multiple supplying peers, we further divide (logically) each segment into equal sized blocks. Given a set of supplying peers {P 1, P 2, …, P M } and the bandwidths from these peers {Bw 1, Bw 2, …, Bw M }, the problem is how to assign blocks to these supplying peers to minimize the total downloading time, as well as to download the earlier blocks as soon as possible. Possible solutions: Round Robin (RR) Bandwidth Proportional (BP): Assigning blocks to suppliers in proportion to their contributed bandwidth
11
11 Multiple-Source Scheduling (MSS) Algorithm Scheduling: 1: for (i = 1; i ≤ num_suppliers; i++) 2: time[i] = 0; // time[i]: the earliest time for supplier[i] to start sending blocks; 3: curr_blk = 0; 4: while (curr_blk ≤ num_blks-1) { 5: for (i = 1; i ≤ num_suppliers; i++) 6: estimatedTime[i] = time[i] + blk_size / Bw[i]; 7: select k, where estimatedTime[k] = Min{estimatedTime[i]}; 8: assign blocks[curr_blk] to supplier[k]; 9: time[k] = time[k] + blk_size / Bw[k]; 10: curr_blk ++; 11: if (curr_blk > num_blks-1) 12: break; 13: }
12
12 Scheduling Algorithms Comparison Example: Assign 8 blocks to 3 suppliers using RR, BP, MSS respectively. The video bit rate: 512kbps Contributed bandwidth: P 1 : 320kbps P 2 : 128 kbps P 3 : 64kbps 6.4s 4.8s
13
13 Streaming Session Once schedule is generated, receiver sends it to suppliers. Then suppliers send the assigned blocks to receiver using UDP. Receiver maintains a ring buffer, Once receiver receives a block, it writes this block to the right position of the ring buffer. During the streaming session, supplier switching will happen if one supplier leaves/fails, or the incoming rate from a supplier is decreasing. To absorb the transient effects of supplier switching, we require receiver to buffer at lease S initBuff blocks before playback starts. The time to finish downloading these blocks is called initial buffering time.
14
14 3. Simulation We conducted extensive simulation on large, Internet-like topologies. In average, the topologies consists of 2050 routers (generated by GT-ITM), and a total of 3010 end hosts (peers).GT-ITM
15
15 Simulation Results The simulation results demonstrate: Based on the usual, low-cost PCs, our proposed architecture can support large-scale on-demand media streaming services in a dynamic heterogeneous Peer-to-Peer network. Our segment distribution algorithm can achieve a fast system streaming capacity amplification, where system streaming capacity is defined as the number of media streaming sessions that can be served by the system concurrently. The proposed scheduling algorithm can achieve small initial buffering time.
16
16 4. Prototype Implementation We implemented a functional prototype using Java and Java Media Framework (JMF). The prototype is totally platform independent.JMF In the prototype, control packets are sent using TCP, streaming packets are sent using UDP. In total, the prototype implementation consists of 109 Java files, 21 packages, 133 classes, and more than 22,000 lines code (About 11,800 lines code in COOLSearch, and rest in BitVampire).
17
17 Prototype System Architecture Communication Component Peers on Network GUI Controller Service AbstractionPeer Abstraction Peer Adaptor Peer Clusters Category Table Content Index Table Neighbour Agents List Category Overlay Category Overlay Search Service Media Service Indexing Service Streaming ReceiverStreaming Supplier Application layer Abstraction layer Service layer Core layer
18
18 Prototype Snapshot
19
19 Possible Business Model Internet Cinema (or Internet DVD Player) users can register with the server of the Cinema Service Provider: Pay per view Pay monthly subscription fees Receive advertisements which can be made adaptive and intelligent. The deployment will be relatively simple, inexpensive, and incremental.
20
20 5. Future Work We plan to deploy our prototype on the PlanetLab (an Internet test-bed) in the near future.PlanetLab Currently we deliver the video in full quality. Some adaptive streaming techniques (layer coding, priority drop, etc.) could be used to improve the quality of service. An incentive mechanism to encourage peers to contribute their resources.
21
21 Thesis Xin Liu, “BitVampire: A Cost-Effective Architecture for On-Demand Media Streaming in Heterogeneous P2P Networks”, MSc Thesis, Computer Science, UBC, Nov. 2005. Xin Liu, Son T. Vuong, “BitVampire: A Cost-Effective Peer-to-Peer Architecture for Large-Scale On-Demand Media Streaming”, invited paper for the Journal of Multimedia (JMM), to appear. Xin Liu, Jun Wang, Son T. Vuong, “A Peer-to-Peer Framework for Cost- Effective On-Demand Media Streaming,” in the Proceedings of 3 rd IEEE Consumer Communications and Networking Conference (CCNC’06), Las Vegas, NV, Jan. 2006. Xin Liu, Son T. Vuong, “Supporting Low-Cost Video-on-Demand in Heterogeneous Peer-to-Peer Networks,” in the Proceedings of 7 th IEEE International Symposium on Multimedia (ISM’05), Irvine, CA, Dec. 2005. Xin Liu, Jun Wang and Son T. Vuong, “A Category Overlay Infrastructure for Peer-to-Peer Content Search,” in the Proceedings of APDCM'05 (in conjunction with IPDPS’05), Denver, CO, Apr. 2005. Journal & Conference Papers Patent In application process…
22
22 http://www.cs.ubc.ca/~liu/bitvampire Thanks!
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.