1 / 20

On Individual and Aggregate TCP Performance

This paper investigates the performance of Transmission Control Protocol (TCP) in various network settings, focusing on both individual and aggregate behaviors under multiple competing flows. It evaluates TCP-Reno's mechanisms, such as slow start and congestion avoidance, and assesses throughput and goodput impacts based on network topology, including parameters like bandwidth and delay. Through extensive simulations, the study identifies performance discrepancies influenced by connection numbers and round-trip times, providing insights valuable for network provisioning. Future work aims to validate findings through Internet experiments.

akira
Télécharger la présentation

On Individual and Aggregate TCP Performance

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. On Individual and Aggregate TCP Performance Lili Qiu Yin Zhang Srinivasan Keshav Cornell University 7th International Conference on Network ProtocolsToronto, Canada, October 31 - November 3, 1999

  2. Talk Outline • Introduction & Motivation • Brief Overview of TCP-Reno • Related Work • Aggregate TCP performance • Summary and future work

  3. Introduction • Understanding TCP performance is critical • Our knowledge is still insufficient • TCP performance under many competing flows is not sufficiently explored • Unclear about the impact of network topology • Goal • Investigate both individual and aggregate TCP behavior under many competing connections • Investigate the impact of network topology • Approach • Extensive simulation

  4. Overview of TCP-Reno • Slow start • Congestion avoidance • Loss recovery • Time out • Fast retransmission

  5. Related Work • TCP analytical model [Padhye et al 98] • Models the throughput of an individual TCP conn • Our simulation evaluation shows the model is reasonably accurate • Doesn’t consider aggregate performance • TCP Behavior with many TCP Flows [Morris 97] • TCP-Taheo under RED dropping policy • Only studies the impact of different # conns • Doesn’t consider other network parameters

  6. Aggregate TCP Performance • Motivation • Useful for network provisioning • Goal • Aggregate TCP behavior for a large number of connections • Impact of network topology

  7. Network Model • Network parameters • Bandwidth • Prop delay • Buffer size • Total number of connections • TCP-Reno • Dropping tail • Notation • = bottleneck bandwidth * propagation delay • = + bottleneck buffer

  8. Simulation Methodology • Three sets of simulations • Same RTT • With random processing time • Two RTT’s

  9. Overview of Our Results

  10. Overview of Our Results (Cont’d)

  11. Simulation 1: With Same RTT • TCP exhibits wide range of behaviors depending on • Case 1: (Large Pipe) • Case 2: (Small Pipe) • Case 3: (Medium Pipe)

  12. Simulation 1: With Same RTTCase 1 ( ) • Global synchronization • Fair

  13. Simulation 1: With Same RTTCase 2 ( ) • Shut-off connections

  14. Simulation 1: With Same RTTCase 3: ( ) • Local synchronization

  15. Simulation 1: With Same RTTPerformance Results • Throughput • Close to 1 if buffer > Wopt or # conn is large • Goodput • Decreases with # conn • Decreasing rate depends on bottleneck bandwidth • Loss probability • Small # conn: Quadratic • Large # conn: Hyperbolic

  16. Simulation 2: With Random Processing Time • Case 1 ( ) • Global synchronization breaks down • Case 2 ( ) • Discrimination less severe • Fewer shut-off connections • Case 3 ( ) • Local synchronization disappears

  17. Simulation 2: With Random Processing TimePerformance Results • Aggregate Throughput • Aggregate Goodput • Loss Probability • Small # conns: linear increase • Large # conns: hyperbolic as before

  18. Simulation 3: Different RTTs • It’s well-known that TCP has bias against long roundtrip time connections • Goal: Quantify the discrimination • Simulation Topology:

  19. Simulation 3:Two RTT’s with Random Processing Time

  20. Summary and Future Work • Evaluate the analytical model for individual TCP connection • Study aggregate TCP performance • With same RTT • With random processing time • With two RTT’s and random processing time • Future directions • Use Internet experiments to verify the results • Further explore TCP performance under different RTT’s

More Related