1

• Methodology to characterize the performance of
  IEEE 802.11 nodes to be deployed in multi-hop
  environments

Marc Portoles Comeras, Andrey Krendzel, Josep
Mangues-Bafalluy Centre Tecnològic de
Telecomunicacions de Catalunya (CTTC) April 20,
2007
2
Outline

• Introduction
• A study case what fails?
• Accounting for errors in detecting other
  transmissions ?
• Measuring ?
• Experimental results
• Using the value in practice
• Conclusions

3
Introduction

• Motivation
• A large body of theoretical work on wireless mesh
  networking has still not been experimentally
  tested
• When going experimental, however, unexpected
  issues are prone to arise
• Previous work has shown that good theory can be
  too promptly discarded when the hardware being
  used is not correctly calibrated
• The objective here is to analyze possible
  inaccuracies in the wireless hardware that may
  distort measurements in experimental wireless
  mesh networking

4
A study case what fails?

• Saturation throughput study 802.11 leads to
  asymmetry
• Theoretical model prediction B receives 4-6 of
  time share
• Experimental resultsB is active more than 20
  of time!

• Whats failing?
• Is the model accurate?
• Does the hardware behave as expected?

--------------------------------------------------
--------------------------------------------------
---------------------------------------------- Cl
aude Chaudet et al. Study of the Impact of
Asymmetry and Carrier Sense Mechanism in IEEE
802.11 Multihops Networks through a Basic Case,
in proceeding MSWIN 2004, Venezia, Italy
5
Accounting for errors in detecting other
transmissions

• Active 802.11 node can be in any of three states
  
• active sending data
• idle not detecting activity in the medium
• stall detecting the medium busy or receiving
  data
• The maximum data that a backlogged station can
  send,

6
Accounting for errors in detecting other
transmissions

• Lets define a metric
• ? 1 - Prnode senses medium idle when it is
  being used
• Maximum data that a non-ideal station sends
• In case we could compute Taccu_stall exactly

7
Measuring ?

• A scenario to measure the metric ?
• Requirement Taccu_stall must be easy to compute
• Flow 1 and Flow 2 characteristics
• Independent, identical and multicast
• In this case ? Taccu_stall Tactive
• ? can be obtained through measuring the maximum
  transmission rate that Flow 2 achieves

8
Measuring ?

• Workload can affect the measure
• If the NUT is not able to handle the data rates
  that it is required to (Flow 1 Flow 2) the
  measure of metric ? is not reliable
• We need a method to detect whether the NUT is
  loosing data due to excessive workload

9
Measuring ?

• Workload can affect the measure
• A validation curve to detect excessive workload

10
Experimental results

• A corrrect measure of the metric without workload
  interference

11
Experimental results

• An example of workload interference in the measure

12
Using the value in practice

• Applying ? to the initial example

• There is a high probability (11 in our
  experiments) that B does not correctly detect
  transmissions from A and C
• Strong bias in the measure

• Possible solutions are
• Modify the model to include hardware inaccuracies
• Choose alternative hardware or correctly tune the
  one that is being used.

13
Conclusions

• This paper
• Shows how wireless hardware solutions may fail in
  detecting transmissions from other stations
• Proposes a novel metric to account for the
  probability that a station fails in detecting
  other transmissions
• Proposes a measurement methodology to obtain a
  value of the metric
• The study also draws attention on the importance
  of correctly characterizing hardware behavior
  before rising conclusions out of measurement
  observations.

14
Thanks for your kind attention!

• Questions?

Marc Portoles Comeras Centre Tecnològic de
Telecomunicacions de Catalunya (CTTC)
Barcelona Contact e-mail marc.portoles_at_cttc.es