Packet Capture UDP Experiments May 2001 Packet Capture UDP Experiments Christopher Ware - TITR, University of Wollongong chris@titr.uow.edu.au Eryk Dutkiewicz - Motorola Australia Research Centre Eryk.Dutkiewicz@motorola.com C Ware - UoW, E Dutkiewicz - MARC
Month 2000 doc.: IEEE 802.11-01/xxx May 2001 Overview Capture experiments using TCP over hidden terminal connections illustrate a strong signal strength dependence (11-01/058) In all cases, the stronger connection is able to lock other connections out, capturing the channel TCP backoff and timeout exacerbates this problem C Ware - UoW, E Dutkiewicz - MARC John Doe, His Company
Overview ‘Capture’ occurs at two levels May 2001 Overview ‘Capture’ occurs at two levels Packet capture where a frame is ‘captured’ by the receiver in the presence of noise/other interference Channel capture where protocol timers interact to prevent stations from accessing the channel Hidden terminal TCP experiments exhibit channel capture, but the signal strength dependence indicates there are additional factors effecting this behaviour C Ware - UoW, E Dutkiewicz - MARC
May 2001 Scenario Repeat the previous experiments without TCP to remove the effect of retransmission timers and timeouts Two hidden terminals are used, each sending traffic to a common receiver each node floods the channel with UDP packets No retransmission, timeout, ACK’s etc Perform experiments with NIC’s from 2 manufacturers Use 802.11b 11Mbit/s PHY C Ware - UoW, E Dutkiewicz - MARC
Experiment Setup Nodes hidden from transmission range viewpoint May 2001 Experiment Setup Nodes hidden from transmission range viewpoint carrier sense may be available 802.11b, RTS/CTS used, (500 byte thresh) Send 10,000+ 512 byte UDP packets to host 2 Trace each transfer with tcpdump C Ware - UoW, E Dutkiewicz - MARC
Vendor A - Equal SNR SNR Con. A - 25dB Con. B - 25dB Connection A May 2001 Vendor A - Equal SNR SNR Con. A - 25dB Con. B - 25dB Connection A commences first C Ware - UoW, E Dutkiewicz - MARC
Vendor B - Equal SNR SNR Con. A - 25dB Con. B - 25dB Connection B May 2001 Vendor B - Equal SNR SNR Con. A - 25dB Con. B - 25dB Connection B commences first C Ware - UoW, E Dutkiewicz - MARC
Vendor A - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Stronger Host May 2001 Vendor A - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Stronger Host commences first C Ware - UoW, E Dutkiewicz - MARC
Vendor B - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Stronger Host May 2001 Vendor B - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Stronger Host commences first C Ware - UoW, E Dutkiewicz - MARC
Vendor A - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Weaker host May 2001 Vendor A - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Weaker host commences first C Ware - UoW, E Dutkiewicz - MARC
Vendor B - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Weaker host May 2001 Vendor B - Unequal SNR SNR Con. A - 25dB Con. B - 20dB Weaker host commences first C Ware - UoW, E Dutkiewicz - MARC
Results Summary In the equal case, sharing of the channel occurs Month 2000 doc.: IEEE 802.11-01/xxx May 2001 Results Summary In the equal case, sharing of the channel occurs In the unequal cases, the stronger host obtains significantly greater throughput, less errors etc. There is no channel capture due to transmission timers etc. Remove TCP and unfairness still present C Ware - UoW, E Dutkiewicz - MARC John Doe, His Company
May 2001 Questions / Issues Is it possible to address this issue within the current QoS draft? Current simulation models do not show this behaviour This is not a fundamental problem with the MAC relative fairness with equal signal power This manifestation of near/far effect greatly affects fairness properties, particularly when hidden terminals are present C Ware - UoW, E Dutkiewicz - MARC