Idle Sense : An Optimal Access Method for High Throughput and - - PowerPoint PPT Presentation

idle sense
SMART_READER_LITE
LIVE PREVIEW

Idle Sense : An Optimal Access Method for High Throughput and - - PowerPoint PPT Presentation

Idle Sense : An Optimal Access Method for High Throughput and Fairness in Rate Diverse Wireless LANs Martin Heusse , Franck Rousseau, Romaric Guillier, Andrzej Duda LSR-IMAG Laboratory Grenoble Logiciels Systmes Rseaux Outline


slide-1
SLIDE 1

Logiciels Systèmes Réseaux

LSR-IMAG Laboratory Grenoble

Idle Sense:

An Optimal Access Method for High Throughput and Fairness in Rate Diverse Wireless LANs

Martin Heusse, Franck Rousseau,

Romaric Guillier, Andrzej Duda

slide-2
SLIDE 2

2

Outline

  • 802.11 DCF principles and shortcomings
  • Towards a better access method
  • Idle Sense principles and properties
  • Performance evaluation
slide-3
SLIDE 3

3

802.11 DCF in a nutshell

Data + ACK Data

DIFS

Data

Data + ACK

Host # 2 1 3 … … … Elapsed backoff Residual backoff collision Medium busy exponential backoff

CW 2 x CW

time

slide-4
SLIDE 4

4

Known shortcomings of DCF

  • Under optimal throughput for N > 4
  • Hosts are too aggressive ⇒ collisions
  • CW too small, not enough time spent in

contention

  • Exponential backoff
  • Good short term fairness for N=2,

degrades for larger N

  • Performance anomaly in rate diverse cells
  • Slow host limits the throughput of faster

hosts

slide-5
SLIDE 5

5

Known shortcomings of DCF

  • Contention control in DCF
  • “Bad day” effect

° If a host looses frames due to bad transmission conditions, it performs frequent exponential backoffs ° Increased CW lowers the transmission attempt probability

  • Physical capture effect
slide-6
SLIDE 6

6

Physical layer capture effect

  • The stronger signal in a collision may be

successfully received

  • It causes long term unfairness
  • Farther host has a greater average

contention window

(Kochut et al., ICNP'04)

slide-7
SLIDE 7

7

Towards a better access method

  • Keep good aspects of DCF
  • No explicit information exchange
  • Keep backoff procedure: random backoff
  • Modifications
  • No exponential backoff

° make hosts use similar values of CW ⇒ fairness

  • Adapt CW to varying traffic conditions

° more hosts, bigger CW; less hosts smaller CW ° do not change CW upon frame loss

slide-8
SLIDE 8

8

Idle Sense

  • Observe the number of idle slots
  • Channel load indicator
  • Control CW
  • Adjust CW to the current state
  • Optimal operation in all conditions

° What is the optimal CW? ° How it relates to the number of idle slots?

slide-9
SLIDE 9

Optimal CW

50 100 150 200 250 300 50 100 150 2 4 6 8 10 CW Cost

Cost function: Proportion of time spent in collisions or contention Minimizing the cost ⇒ Maximizing throughput

(Calì et al., Transactions on Networking, 2000)

slide-10
SLIDE 10

10 20 30 40 50 200 400 600 N

  • ptimal CW

10 20 30 40 50 3.5 4.5 5.5 N ni

Optimal CW

CW proportional to N n̄i: average number of idle slots between transmission attempts n̄i converges quickly n̄i

target

slide-11
SLIDE 11

11

Idle Sense

  • Hosts track n̄i and make it converge to the

target value

  • Each host estimates n̄i
  • Rises/Lowers CW when n̄i too small/big

compared to n̄i

target

  • Adjusting CW is done according to AIMD

⇒ all hosts converge to a similar value of CW

slide-12
SLIDE 12

Example

1 1 1 1 2 3 2

8 6 5 4 5 2

3

1

ni

CW1=28

(α= 1/1.2, ε=0.01, ntrans=3, n̄i target = 5.7)

CW2=22 CW1=25 n̂i = 6.3 n̂i = 6.3 CW1=30 n̂i = 3.7 CW2=26 n̂i = 3.7 CW3=30 n̂i = 2

CW2=25

Host 3 joins

CW3=25 2 active hosts DIFS

slide-13
SLIDE 13

13

Properties

  • Contention control independent of frame

loss detection

  • No “bad day” effect
  • Solves the physical layer capture effect
  • Short term fair
  • Fixes performance anomaly
  • Time fairness achieved by scaling CW

according to the transmission rate

  • Hidden terminal problem: use RTS/CTS
  • No hardware modification required
slide-14
SLIDE 14

14

Properties:

Channel adaptation

  • With Idle Sense, the collision probability Pc is

known and bounded (after convergence)

  • Frame loss probability Perr ≈ 1- Pc - Pok

° Pok can be observed

  • Provides a new means for setting the right

transmission rate

  • Change rate when Perr exceeds a given

threshold

  • May be combined with SNR measurements
slide-15
SLIDE 15

15

Performance evaluation

  • Throughput
  • Fairness: Jain index
  • Convergence speed
  • Time fairness
slide-16
SLIDE 16

Throughput

0.05 0.1 0.15 0.2 0.25 0.3 0.35 20 40 60 80 100 120 140 160 180 200 Throughput (Mb/s) N 802.11b Idle Sense

slide-17
SLIDE 17

Fairness

50 hosts

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 10 20 30 40 50 Jain index Normalized window size Idle Sense 802.11b DCF Slow Decrease AOB

slide-18
SLIDE 18

Convergence speed

50 100 150 200 250 1000 2000 3000 4000 5000 6000 7000 CW size Transmission number Host staying for the whole run Host staying for a small number of transmissions

slide-19
SLIDE 19

19

Time fairness

200 400 600 800 1000 Idle Sense + Time fairness 802.11b Time spent transmitting (s) Host transmitting at 11Mbps Host transmitting at 1Mbps

slide-20
SLIDE 20

20

Conclusions

  • Performance gains
  • Addresses many issues in wireless LANs
  • Main property: it uncouples frame loss and

contention control

  • Enables other improvements
  • eg. give more weight to the access point