Understanding 802.11e Voice Behaviour
via Testbed Measurements and Modeling
Ian Dangerfield, David Malone, Doug Leith.
20 April 2007
1
Voice over WiFi
• Behaviour of voice over WiFi.
• Infrastructure mode 802.11(e).
• AP known to constrain capacity.
• Modeling/Simulation suggests solution.
• Buffering also a question.
• Simple on-off traffic model.
• Use test-bed to understand problem and solution.
2
Problem: When busy
MAC is per-packet ‘fair’
1.2
1
ratio of throughputs
0.8
0.6
0.4
0.2
0
1
2
3
4
5
6
wired link bandwidth (Mbs)
3
7
8
9
10
4
Delay Measurement
• Transmission not complete until MAC ACK.
• Hardware supports interrupt after ACK.
Interface TX
Queue
1. Driver notes
enqueue time.
Driver
4. Driver notes
completion time.
Driver TX
Discriptor
Queue
2. Hardware
contends until
ACK received
Packet transmitted
Hardware
ACK received
5
3. Hardware
interrupts
driver.
0.025
Fraction of Packets
0.02
0.015
0.01
0.005
0
900
1000
1100
1200
1300
Delay (seconds x 10-6)
6
1400
1500
1600
Throughput at AP, downlink (queue = 1, 3, 4 and 30)
35
Throughput (kbits/second)
30
25
20
15
10
Mean Offered Load
Model: Q=1
Model: Q=inf
Q=3
Q=4
Q=30
5
0
0
2
4
6
8
10
12
14
16
18
Number of stations (n)
Figure 1: Measured and modelled throughput at the AP.
Buffer size at the AP and the STAs is the same. Analytic
model also shown.
7
Throughput at one STA (queue = 1, 3, 4 and 30)
35
Throughput (kbits/second)
30
25
20
15
10
Mean Offered Load
Model Q=1
Model Q=inf
Q=3
Q=4
Q=30
5
0
0
2
4
6
8
10
12
14
16
18
Number of stations (n)
Figure 2: Measured and modelled throughput at a single
STA. The buffer size at the AP and the STAs is the same.
8
Throughput at AP, downlink (mixed buffer sizes)
35
Throughput (kbits/second)
30
25
20
15
10
Mean Offered Load
STA Q=4 AP Q=4
STA Q=4 AP Q=15
STA Q=30 AP Q=30
STA Q=4 AP Q=399
5
0
0
2
4
6
8
10
12
14
16
18
Number of stations (n)
Figure 3: Measured throughput at the AP. Results are
shown for various combinations of buffer sizes at the STAs
and the AP.
9
Throughput at one STA (mixed buffer sizes)
35
Throughput (kbits/second)
30
25
20
15
10
Mean Offered Load
STA Q=4 AP Q=4
STA Q=4 AP Q=15
STA Q=30 AP Q=30
STA Q=4 AP Q=399
5
0
0
2
4
6
8
10
12
14
16
18
Number of stations (n)
Figure 4: Measured throughput at a single STA. Results
are shown for various combinations of buffer sizes at the
STAs and the AP.
10
CDF AP TXOP 1, CWmin 31 (for 5, 10 and 15 STAs)
100
90
Percentage of packets
80
70
60
50
40
30
5 STAs, Q=4
5 STAs, Q=30
10 STAs, Q=4
10 STAs, Q=30
15 STAs, Q=4
15 STAs, Q=30
20
10
0
0
2000
4000
6000
8000
10000
MAC Delay (seconds x 10-6)
Figure 5: Cumulative distribution for delays at the access
point when there are 5, 10 and 15 stations with standard
MAC settings.
11
CDF STA TXOP 1, CWmin 31 (for 5, 10 and 15 STAs)
100
90
Percentage of packets
80
70
60
50
40
30
5 STAs, Q=4
5 STAs, Q=30
10 STAs, Q=4
10 STAs, Q=30
15 STAs, Q=4
15 STAs, Q=30
20
10
0
0
2000
4000
6000
8000
10000
MAC Delay (seconds x 10-6)
Figure 6: Cumulative distribution for delays at a station
when there are 5, 10 and 15 stations with standard MAC
settings.
12
Solution: 802.11e?
• 802.11e makes parameters tunable.
• CWmin: base range for backoff.
• TXOP: transmission duration.
• Simple solution: TXOP = n packets.
• Better solution: CWmin = 16, TXOP = n/2 packets?
13
1 Sta, CWmin 31
0.012
’1_sta_cw31_1000_col_per’ u 2:1
Fraction of total Packets
0.01
0.008
0.006
0.004
0.002
0
900
1000
1100
1200
1300
delay (seconds x 10-6)
14
1400
1500
1600
Throughput at AP, downlink (queue = 3, 4, n and 30)
35
Throughput (kbits/second)
30
25
20
15
10
Mean Offered Load
Q=3, TXOP n, CWmin 31
Q=4, TXOP n, CWmin 31
Model: Q=n, TXOP n, CWmin 31
Model: Q=inf, TXOP n, CWmin 31
Q=30, TXOP n, CWmin 31
5
0
0
2
4
6
8
10
12
14
16
18
Number of stations (n)
Figure 7: Throughput at the AP for prioritised voice, with
T XOP = n packets.
15
Throughput at AP, downlink (queue = 1, 3, 4 and 30)
35
Throughput (kbits/second)
30
25
20
15
10
Mean Offered Load
Q=3, TXOP n/2, CWmin 15
Q=4, TXOP n/2, CWmin 15
Model: QSTA=1, QAP=n/2, TXOP n/2, CWmin 15
Model: Q=inf, TXOP n/2, CWmin 15
Q=30, TXOP n/2, CWmin 15
5
0
0
2
4
6
8
10
12
14
16
18
Number of stations (n)
Figure 8: Throughput at the AP for prioritised voice,
T XOP = n/2, CWmin = 15.
16
Throughput at one STA (queue = 1, 3, 4 and 30)
40
Throughput (kbits/second)
35
30
25
20
15
Mean Offered Load
Q=3, TXOP n, CWmin 31
Q=4, TXOP n, CWmin 31
Model: QSTA=1, QAP=n, TXOP n, CWmin 31
Model: Q=inf, TXOP n, CWmin 31
Q=30, TXOP n, CWmin 31
10
5
0
0
2
4
6
8
10
12
14
16
18
Number of stations (n)
Figure 9: Throughput at a STA for prioritised voice, with
T XOP = n packets at the AP.
17
Mean delays at AP, queue = 3, 4 and 30
1800
Q=30, TXOP 1, CWmin 31
Q=4, TXOP 1, CWmin 31
Q=3, TXOP 1, CWmin 31
Q=4, TXOP n, CWmin 31
Q=3, TXOP n, CWmin 31
Q=30, TXOP n, CWmin 31
Q=3, TXOP n/2, CWmin 15
Q=4,TXOP n/2, CWmin 15
Q=30, TXOP n/2, CWmin 15
Mean delay (seconds x 10-6)
1600
1400
1200
1000
800
600
400
200
0
0
2
4
6
8
10
12
14
16
n, number of voice stations
Figure 10: Mean MAC delay at the AP. The group with
longer mean delays correspond to the experiments in which
the AP is not prioritised.
18
Mean delay at voice station, queue = 3, 4 and 30
8000
Q=30, TXOP 1, CWmin 31
Q=4, TXOP 1, CWmin 31
Q=3, TXOP 1, CWmin 31
Q=3, TXOP n, CWmin 31
Q=4, TXOP n, CWmin 31
Q=30, TXOP n, CWmin 31
Q=3, TXOP n/2, CWmin 15
Q=4, TXOP n/2, CWmin 15
Q=30, TXOP n/2, CWmin 15
Mean delay (seconds x 10-6)
7000
6000
5000
4000
3000
2000
1000
0
0
2
4
6
8
10
12
14
16
n, number of voice stations
Figure 11: Mean MAC delay a station. The mean interpacket arrival time at a STA is 10000µs.
19
CDF AP TXOP n/2, CWmin 15 (for 5, 10 and 15 STAs)
100
90
Percentage of packets
80
70
60
50
40
30
5 STAs, Q=4, TXOP n/2, CWmin 15
5 STAs, Q=30, TXOP n/2, CWmin 15
10 STAs, Q=4, TXOP n/2, CWmin 15
10 STAs, Q=30, TXOP n/2, CWmin 15
15 STAs, Q=4, TXOP n/2, CWmin 15
15 STAs, Q=30, TXOP n/2, CWmin 15
20
10
0
0
2000
4000
6000
8000
10000
MAC Delay (seconds x 10-6)
Figure 12: Cumulative distribution for delays at the AP
when there are 5, 10 and 15 stations and the AP is prioritised using T XOP = n/2 and CWmin = 15.
20
CDF STA TXOP 1, CWmin 31 (for 5, 10 and 15 STAs)
100
90
Percentage of packets
80
70
60
50
40
30
5 STAs, Q=4
5 STAs, Q=30
10 STAs, Q=4
10 STAs, Q=30
15 STAs, Q=4
15 STAs, Q=30
20
10
0
0
2000
4000
6000
8000
10000
MAC Delay (seconds x 10-6)
Figure 13: Cumulative distribution for delays at a station
when there are 5, 10 and 15 stations with standard MAC
settings.
21
CDF STA TXOP n, CWmin 31 (for 5, 10 and 15 STAs)
100
90
Percentage of packets
80
70
60
50
40
30
5 STAs, Q=4, TXOP n, CWmin 31
5 STAs, Q=30, TXOP n, CWmin 31
10 STAs, Q=4, TXOP n, CWmin 31
10 STAs, Q=30, TXOP n, CWmin 31
15 STAs, Q=4, TXOP n, CWmin 31
15 STAs, Q=30, TXOP n, CWmin 31
20
10
0
0
2000
4000
6000
8000
10000
MAC Delay (seconds x 10-6)
Figure 14: Cumulative distribution for delays at a station
when there are 5, 10 and 15 stations and the AP is prioritised using T XOP = n.
22
CDF STA TXOP n/2, CWmin 15 (for 5, 10 and 15 STAs)
100
90
Percentage of packets
80
70
60
50
40
30
5 STAs, Q=4, TXOP n/2, CWmin 15
5 STAs, Q=30, TXOP n/2, CWmin 15
10 STAs, Q=4, TXOP n/2, CWmin 15
10 STAs, Q=30, TXOP n/2, CWmin 15
15 STAs, Q=4, TXOP n/2, CWmin 15
15 STAs, Q=30, TXOP n/2, CWmin 15
20
10
0
0
2000
4000
6000
8000
10000
MAC Delay (seconds x 10-6)
Figure 15: Cumulative distribution for delays at a station
when there are 5, 10 and 15 stations when the AP is prioritised using T XOP = n/2 and CWmin = 15.
23
Conclusion
• Reproduced capacity problem.
• Buffering helps, TXOP helps more.
• Models are producing useful predictions.
• Burstiness for TXOP seems OK.
• Some interesting MAC/buffer tradeoffs.
24