Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 33

An Experimental Study of ARQ

Protocol in 802.11b Wireless LAN

Chung Ho Nam, Soung C. Liew, C


heng Peng Fu
Outline
 Introduction
 Background
 Details of experiments
 Conclusion
Introduction
 802.11b is becoming hot
 Previous studies are based on
simulations and old wlan technology
 Experimental approach was used in this
paper
Introduction
 2 main issues:
 Effectiveness of ARQ in 802.11b
 3 Experiments
 Loss behavior of 802.11b network

 2 more:
 Self-collision of TCP traffic
 Changes in collision rates due to

movements of mobile hosts


What is ARQ?
 Automatic Repeat reQuest
 Error control in data transmission
 3 common techniques:
 Stop and Wait
 Go back N
 Selective Repeat
 802.11b employs “Stop and Wait” at
link-layer
ARQ in 802.11b
 Stop and Wait
ARQ in 802.11b
 Stop and Wait
ARQ in 802.11b
 Not 100% reliable
 Only try at most (1 + 3) times in current
commercial products
 e.g. Orinoco, Buffalo

 Algorithm
 Skip
Why?
 Transmission error in air
 Channel noise
 Channel interference
 Packet collisions
 No direct collision detection in 802.11b
 TCP or UDP on top of 802.11b
 Misinterpret packet loss as congestion
loss (TCP)
Set-up for Experiments
Statistics from AP Manager
 Multiple Transmit Retry Count (MTRC)
 >= 2 retransmission attempts (successful)
 Transmit Retry Count (TRC)
 >= 1 retransmission attempts (successful)
 Transmit Failed Count (TFC)
 Still failed after 3 retransmission attempts
Experiment (I)
 Is ARQ adequate to compensate for packet lo
ss due to transmission error?
 Server: NetProbe
 Client: NetProbe
 Protocol: UDP
 Packet size: 1460 bytes
 Data rate: 500kbps
 Number of packets: 100,000
 Distance: 5m (best), 15m (good),
25m (marginal)
Result of experiment (I)
Link quality Best Good Marginal
TRC 661 2322 18073
MTRC 16 1172 4030
TFC 10 264 760
Loss rate 0.01% 0.26% 0.76%

Loss rate = TFC / Total number of UDP packets


Result of experiment (I)
 Total number of unsuccessful transmission attempts
at layer 2
= Packets requiring exactly one retransmission to
succeed
+ 2 * Packets requiring exactly two retransmissions to
succeed
+ 3 * Packets requiring exactly three retransmissions to
succeed
+ 4 * Packets that fail to be transmitted successfully
(i.e., TFC)
Result of experiment (I)
= Packets requiring one or more retransmissions to
succeed (i.e., TRC) 1+2+3

+ Packets requiring two or more retransmissions to


succeed (i.e., MTRC) 2+3

+ 1 * Packets requiring exactly three retransmissions


to succeed 3
4+4+4+4
+ 4 * TFC

Let  be this
Result of experiment (I)
TRC  MTRC    4 * TFC
Ploss 
num _ of _ packets  TRC    MTRC  4 * TFC

TRC  MTRC  4 * TFC



num _ of _ packets  TRC  MTRC  4 * TFC

Link quality Best Good Marginal


TRC+MTRC+4*TFC 717 4550 25143
Lower bound on loss rate 0.71% 4.35% 20.09%
Result of experiment (I)
Experiment (II)
 Is ARQ adequate to compensate for packet
loss due to packet collisions?
 Two-way traffic
 Protocol: UDP
 Packet size: 1460 bytes
 Data rate: 500kbps
 Number of packets: 100,000
 Distance: 5m (best), 15m (good),
25m (marginal)
Result of experiment (II)
Link quality Best Good Marginal
TRC 1371 6951 16565
MTRC 231 3110 6030
TFC 7 1022 2038
Loss rate 0.007% 1.02% 2.04%
TRC 661 2322 18073
MTRC 16 1172 4030
TFC 10 264 760
Loss rate 0.01% 0.26% 0.76%
Result of experiment (II)
TRC  MTRC  4 * TFC
num _ of _ packets  TRC  MTRC  4 * TFC

Link quality Best Good Marginal


TRC+MTRC+4*TFC 1630 14149 30747
Lower bound on loss rate 1.60% 12.40% 23.52%

Link quality Best Good Marginal


TRC+MTRC+4*TFC 717 4550 25143
Lower bound on loss rate 0.71% 4.35% 20.09%
Result of experiment (II)
TCP throughput VS Loss Prob.
BW=1.6Mbps, RT T =120ms, buffer size=12

200
Reno
Throughput(kBytes/s)

180
160
140
120
100
80
60
40
20
0
1.00E-05 1.00E-04 1.00E-03 1.00E-02 1.00E-01 1.00E+00

Packet loss rate


Implication of Experiment (II)
 Collision Probability Increases with
Distance
Interaction between transmission errors and
collision errors
 But “noise” and “interference” should be
independent
 More time is needed for transmission when
the data rate is low
 Wireless medium is much more occupied
 Probability of packet loss is higher
The Poor Pulling Down the Rich
 e.g.

 When A moves farther away from AP,


packet duration increases
 Less airtime for B
 Both A and B degrade
Experiment (III)
 Interactions between link-layer ARQ in
WLAN and transport-layer ARQ in TCP?
 Suiteserver and Suiteclient (like FTP)
 Protocol: TCP
 Packet size: 1460 bytes
 Data rate: Not restricted
 Total data size = 1460*100,000 bytes
 Distance: 5m (best), 15m (good),
25m (marginal)
Result of experiment (III)

Link quality Best Best Good Good Margina Marginal


l
TRC 5804 5872 6563 7751 20747 16850
MTRC 891 785 1095 953 6546 4569
TFC 17 10 149 101 1588 698
Throughput 4883.4 4458.4 3579.9 4010.0 747.7 2315.4
(kbps)
> 5% of packets were retransmitted
Result of experiment (III)
 Recall: link-layer packets and ACKs DO
NOT collide
 TCP ACKs are regular link-layer packets
 Collide with packets from AP
 More experiments need to be done
Packet Loss Pattern in 802.11b
 Random Loss VS Bursty Loss
 Let p = total packet loss / total packet transmitted
= the overall loss rate of a link
 If packets losses are random, then the gap between t

wo successive packet losses follows the following pro


bability distribution:
 P = P [j next packets transmitted successfully befor
j
e the next loss] = p (1-p) j 
 In particular, if loss is random, P = p, whereas if los
0
s is bursty, P > p.
Experimental Evaluation
 Server: NetProbe
 Client: NetProbe
 Protocol: UDP
 Packet size: 1460 bytes
 Data rate: 500kbps
 Number of packets: 100,000
 Distance: 25m (marginal)
 p=0.0676

IDEAL

y=p(1-p)x
 p=0.0676

REAL
Conclusions
 One-way traffic
 No collision
 ARQ is effective in keeping the loss rate to
below 0.01
 Two-way traffic
 Collisions exist
 Loss rate may rise beyond 0.01
Conclusions
 Two possible improvements
 Strengthen the link-layer ARQ
 Modify TCP
 Two issues to be studied
 Self-collision in TCP
 Movements of clients
 Packet loss in 802.11b is not random

You might also like