Explicit Transport Error Notification (ETEN) for Error-Prone Wireless and Satellite Networks

Rajesh Krishnan, Mark Allman, Craig Partridge, and James P.G. Sterbenz,
Explicit Transport Error Notification for Error-Prone Wireless and Satellite Networks,
BBN Technical Report 8333, 7 February 2002 (revised 22 March 2002).
[ PDF ]

ABSTRACT

Wireless and satellite networks have non-negligible error rates that can significantly influence TCP performance because TCP considers every packet loss as an indicator of congestion, and thus throttles the packet transmission rate. Explicit transport error notification (ETEN) mechanisms can aid TCP in distinguishing packets that are lost due to congestion from ones that are lost due to corruption. If TCP can retransmit a packet lost due to corruption without needlessly reducing the transmission rate, a performance benefit can be realized.

In this study we propose two types of ETEN mechanisms:

  1. per-packet mechanisms that notify endpoints of each detected corruption
  2. cumulative mechanisms that notify endpoints of aggregate corruption statistics

We have implemented the proposed mechanisms in the ns-2 simulator. We present simulation results on performance gains achievable for TCP Reno and TCP SACK, using ETEN mechanisms over a wide range of bit error rates and traffic conditions. We compare TCP Reno and TCP SACK enhanced with ETEN mechanisms against TCP Westwood, which uses a bandwidth estimation strategy in place of the traditional AIMD congestion avoidance algorithm. We discuss two issues related to the practical deployment of ETEN mechanisms: corruption detection mechanisms (and their co-operation with ETEN-based recovery in the transport layer) and security aspects. We include recommendations for further work. Our conclusions from this study are:

  1. per-packet ETEN mechanisms offer substantial gains in bulk TCP goodput in the absence of congestion; however, in the presence of congestion TCP congestion avoidance mechanisms dominate resulting in insignificant gains from ETEN
  2. the proposed per-packet mechanisms provide useful upper bounds on performance that can be used to evaluate future proposals of per-packet and cumulative ETEN techniques
  3. per-packet mechanisms present significant challenges to practical implementation by providing a new opportunity to exploit Internet security vulnerabilities and by requiring intermediate nodes to reliably extract information from the headers of corrupted packets
  4. cumulative ETEN techniques are more attractive to implementation; however, the particular mechanism we evaluated did not realize the potential gains of per-packet techniques
  5. future work in this area should focus on alternative cumulative ETEN mechanisms, accurate loss inference at endpoints to avoid tracking congestion losses at every hop, interactions with forward error correction, and cross-layer co-operation for ETEN

Keywords

Explicit transport error notification, wireless and satellite networks, TCP performance, congestion, corruption, bit errors.

Cross-layer interaction and optimization optimisation, knobs and dials.

Outline

  1. INTRODUCTION
  2. ERROR NOTIFICATION AND RESPONSE MECHANISMS
    1. Sufficient Information Available about Endpoints
      1. Oracle ETEN
      2. Backward ETEN
      3. Forward ETEN
    2. Insufficient Information Available about Endpoints – Cumulative ETEN
    3. Sender Response to ETEN
  3. PERFORMANCE OF PER-PACKET ETEN MECHANISMS
    1. Statistical Limitations of this Study
    2. Simulation Software Description
    3. Baseline – no Cross Traffic over a Single-Hop Topology
    4. Multi-Hop Topology with no Cross Traffic
    5. Multi-Hop Topology with UDP Cross Traffic
    6. Multi-Hop Topology with Competing TCP flows
    7. Comparison of ETEN to TCP Westwood
  4. CUMULATIVE ETEN STRATEGIES AND PERFORMANCE
    1. Transporting Corruption Survival Probability Estimates
    2. Computing the per-Link Packet Corruption Survival Probability at each Router per-Link
    3. Transport Endpoint Strategies to use CETEN to Discern Congestion from Corruption
    4. Cumulative ETEN Performance with UDP Cross Traffic
    5. Cumulative ETEN Performance with TCP Cross Traffic
    6. Discussion of CETEN Simulations
  5. COMBINING CETEN STRATEGIES WITH DYNAMIC FEC ADJUSTMENT
  6. MECHANISMS TO DETECT CORRUPTED PACKETS
    1. Link-Layer Mechanisms
    2. IP Checksum
    3. TCP Checksum
    4. IPsec
  7. ETEN SECURITY ISSUES AND IMPLICATIONS
    1. Potential Security Vulnerabilities due to ETEN
    2. Interoperability with Existing Security Mechanisms
  8. CONCLUSIONS AND FUTURE WORK
    1. Salient Conclusions
    2. Presentations, Reports, Publications, and Software
    3. Recommendations for Future Work
  9. REFERENCES
  10. FULL SET OF SIMULATION PLOTS
  11. ACKNOWLEDGMENTS

Last updated 16 November 2003 – Valid XHTML 1.1 – Lynx inspected – W3C AAA Conformance
©2003 James P.G. Sterbenz <jpgs@sterbenz.org>