1 / 22

Receiver Driven Bandwidth Sharing for TCP

Receiver Driven Bandwidth Sharing for TCP. Authors: Puneet Mehra, Avideh Zakor and Christophe De Vlesschouwer University of California Berkeley. Presented at: INFOCOM 2003 . Twenty-Second Annual Joint Conference of the IEEE Computer and Communications Societies. Overview of the Presentation.

Pat_Xavi
Télécharger la présentation

Receiver Driven Bandwidth Sharing for TCP

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. Receiver Driven Bandwidth Sharing for TCP Authors: Puneet Mehra, Avideh Zakor and Christophe De Vlesschouwer University of California Berkeley. Presented at: INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE Computer and Communications Societies.

  2. Overview of the Presentation • Motivation • Goals • Proposed Method • NS-2 Simulations • Conclusion

  3. Motivation • Most Internet traffic is TCP • HTTP, FTP, P2P, Multimedia streaming… • In many cases access links are bottleneck • Limited Bandwidth (B/W) eg: DSL/Cable < 1.5Mbps • User run many apps that compete for B/W • Problem: TCP shares bottleneck B/W according to RTT • Not fair to flows with large RTT • Doesn’t consider application needs or user prefs!

  4. Congestion Example: FTP Low RTT INTERNET P2P Video traffic Med. RTT High RTT

  5. Goals • Achieve full utilization of the receiver’s access link (bottleneck). • Satisfy user preferences: -priorities assigned to each flow. • Approach: limit throughput of low-priority flows to provide additional B/W for high-priority ones

  6. Overview of the Presentation • Motivation • Goals • Proposed Method • NS-2 Simulations • Conclusion

  7. System Overview User Preferences W1 & d1 R1 T1 TRAS Target Rate Allocation Sub-System FCS1 Flow Control System Sender1 . . . . . . Internet Wn & dn σ Tn FCSn Flow Control System σ Calculation Sub-System Rn R1 Sendern Rn For the receiverσ= system target bit-rate For the nth connectionWn= Advertised Windowdn = Delay in ACK packetsTn= Target RateRn = Measured Rate BWSSBandwidth Sharing System

  8. System Overview… • Band-Width Sharing System (BWSS) consists of: a) Flow Control System (FCS) b) Target Rate Allocation Sub-system (TRAS) c) σ Calculation Sub-system.

  9. Flow Control System For the nth connectionW = Advertised Windowd = Delay in ACK packetsR = Measured RateP = Packet size in bitsTi= Target Ratemi = minimum bandwidthwi= weight Measure Bit-rate and RTT R1 Adapt Receiver Window / ACK Delay Calculate Target Rate – Measured Rate W1 T1 d1 FCS1 Flow Control System

  10. Flow Control System… Ri < Ti : search for the smallest Wi to achieve (1- α )Ti =< Ri =< (1+ α )Ti If Ri > (1+α)*Ti then delay the ACKs as decreasing Wi is ineffective. Aim to minimize delay : otherwise results in unresponsiveness & instability in TCP flow.

  11. Example After fast recovery Receiver’s advertised window Window size limits the data rate : Max Window size = min (cwndmax, receiver’s adv. window) Slide borrowed from Dr. Nitin Vaidya’s TCP tutorial

  12. RTT and Bandwidth estimation • TCP timestamp option to estimate RTT. • Bandwidth estimation relies on exponentially weighted moving average R  α*R + (1-α)*Rø • Ø – bandwidth estimation period, tradeoff between accuracy of estimation and time for convergence.

  13. σ User Prefs. Tn Target Rate Allocation System T1 • Some apps need minimum guaranteed rate(video), others don’t (ftp) • User assigns each flow: • Priority (pi), minimum rate (mi) and weight (wi) • Bandwidth allocation algorithm: • Satisfy minimum rate in decreasing order of priority • Remaining B/W shared according to weight Prevents starvation of low priority connection

  14. σ – Calculation Subsystem R1 σ U = Σi Ri RN Goal: Choose σ to maximize link utilization. U = Σi Ri (σ) Approach: Iteratively increase/decrease σ and measure the impact on utilization σ < σideal implies under-utilization of the link. If σ > σideal , does it affect the system ?

  15. Overview of the Presentation • Motivation • Goals • Proposed Method • NS-2 Simulations • Conclusion

  16. Example of User Preferences Time 0: Min. Rate = 0 Kb/s weights = 1,2,3 for S0-S2 Priority -> S0 (max), S2(min) Time 300: Min Rate = 600 Kb/s TCP BWSS

  17. Network-Congestion Example Priorities: increasing from S0-S2 Min Rate: S0,S2 – 600Kb/s S1 – 100 Kb/s Time 400s to 1200s 700Kb/s Interfering TCP traffic S2 limited to 300Kb/s

  18. Multimedia Streaming Example • S0 – Ftp traffic. Low Priority • Min Rate = 700Kb/s • S1 – Streaming at 450Kb/s • High Priority • 300Kb/s UDP flow (400s-1000s)

  19. Overview of the Presentation • Motivation • Goals • Proposed Method • NS-2 Simulations • Conclusion

  20. Conclusion • BWSS allows user to allocate link B/W • Flexible B/W allocation model • Adapts to changing network conditions • No changes to TCP/senders/routers • Observation: - works only if desired rate is achievable under flow’s cwnd - What was receiver window advertisement actually designed for??

  21. Data1 win4 2 Ack1 win4 4 3 4 5 6 DATA3 ~ 6 win4 8 Ack6 win2 9 DATA10 ~11 win4 10 11 Observation: TCP window management sender receiver 1

  22. Questions??

More Related