Protocols for Fast Delivery of Large Data Volumes
Download as PPTX, PDF1 like218 views
The document discusses protocols for fast delivery of large data volumes, focusing on applications such as high-definition video streaming, sensor networks, and data transfers between clouds. It explores various methods like parallel TCP and application-aware overlay networks to improve bandwidth and throughput while addressing latency issues and packet loss. Furthermore, it highlights the pros and cons of these techniques and presents alternative protocols like XTP and RBUDP for efficient data transmission.
Protocols for Fast Delivery of Large Data Volumes
- 1. Protocols for Fast Delivery of Large Data Volumes CS4482 High Performance Networking Dilum Bandara Dilum.Bandara@uom.lk Some slides extracted from Dr. Dan Massey’s CS557 class at Colorado State University
- 2. High Data Volume Applications High definition video streaming Ultra high-definition video Sensor networks Video surveillance Radar networks Array of Radio telescopes Data transfer between grids/clouds Data transfer from CERN Virtual reality Holographic 3D display Some applications require ordered delivery others don’t 2 Source: NAIC/Arecibo Obs/NSF Source: www.idrshare.com
- 3. Latency Bandwidth Tradeoff Bandwidth is increasing 10-100 Gbps networks Latency is not reducing Speed-of-light limitation Small transfers are latency limited telnet, ssh, chat messages, small file transfer Large transfers are still bandwidth limited Bulk transfer of files 3
- 4. Bulk Transfer 4 Source: N. Tolia et al., “An Architecture for Internet Data Transfer,” NSDI ‘06, 2006
- 5. TCP Design Assumptions Low physical link error rates Packet loss = congestion signal No packet reordering at network (IP) level Packet reordering = congestion signal These design assumptions are challenged today Parallel networking hardware Packet reordering Dedicated links or reservations No congestion Bulk transfers Streaming not needed 5
- 6. Parallel TCP Create many parallel TCP connections to aggregate bandwidth 6 Source: www.codeproject.com/Articles/28788/Distributed-Computing-in-Small-and-Medium-Sized-Of
- 7. Parallel TCP (Cont.) 7 . . . 1 2 N . . . i Bottleneck capacity c AIMD connection Xi = send rate Source: E. Altman, “Parallel TCP Sockets: Simple Model, Throughput and Validation,” IEEE Infocom, 2006
- 8. Parallel TCP – Pros & Cons Pros Aggregated bandwidth More resilient to network layer packet losses Only one stream would may experience timeout More aggressive behavior Slow-start is faster, k x MSS Recovery is faster compared to single stream with giant window Only one stream may experience loss Multiplicative decrease is effectively 1/(2k) rather than 1/2 Can work around max TCP buffer size limitations k x buffer size 8
- 9. Parallel TCP – Pros & Cons (Cont.) Cons Ideally, each connection should use a different path Exploits TCP’s fairness Can become unfair to other flows Requires changes to applications to support parallel streams May perform worse if loss is due to congestion May add to congestion Selecting optimum buffer size & number of streams is hard 9
- 10. Parallel TCP Performance 10 Source: www.codeproject.com/Articles/28788/ Distributed-Computing-in-Small-and-Medium- Sized-Of
- 11. Scalable TCP (SCTP) Modifies congestion control algorithm Each packet loss decreases congestion window by a factor of 1/8 instead of Standard TCP's ½ congestion window When packet loss stops, rate is ramped by adding one packet every 100 successful ACKs Standard TCP – increase by inverse of congestion window very large windows take a long time to recover 11
- 12. TCP friendly Rate Adaptation Based On Loss (TRABOL) UDP Fast, best effort, insensitive to congestion TCP Slow, reliable, sensitive to congestion TRABOL Fast, best effort, sensitive to congestion Depends on end application/user feedback End application/user specifies 2 rates Target Rate (TR) Minimum Rate (MR) Congestion control is similar to TCP AIMD 12
- 13. TRABOL (Cont.) 13 Source: A. Trimmer et al., “Performance of High- Bandwidth TRABOL Protocol for Radar Data Streaming,” IEEE Region 5 TPS Conference, 2006.
- 15. Application aWare Overlay Networks (AWON) Packets are marked based on application requirements Drop packets in an application-aware manner Multicast nodes send aggregated requests to source nodes 15 Source: T. Banka et al., “An Architecture and a Programming Interface for Application-Aware Data Dissemination Using Overlay Networks,” COMSWARE '07, 2007.
- 17. Content-Based Packet Marking 17 ADU – Application Data Unit
- 18. On-The-Fly Data Selection 18 • Compensation for lost packets • Select a packet from a higher rate
- 19. AWON Performance 19 Measurements on PlanetLab
- 20. Other Solutions 20 Source: http://cloudcomputingseminar.wordpress.com/2012/06/16/unit-2-grid-computing/ • XTP – Xpress Transport Protocol - Fast & light weight • RBUDP – Reliable Blast UDP - high-bandwidth, reliable • Tsunami – Improvement of RBUDP
Editor's Notes
- #20: (At source, at multicast node)