Investigating Spurious TCP Timeouts in
802.11
David Malone, Douglas J. Leith,
Anshuman Aggarwal, Ian Dangerfield
17 June 2008
1
TCP over WLAN Woes
12 uploads
fixed 12 uploads
12 TCP Uploads to AP for 180 sec
12 TCP Uploads to AP for 180 sec
1200
600
500
800
Throughput(kbits/sec)
Throughput(kbits/sec)
1000
600
400
200
0
400
300
200
100
1
2
3
4
5
6
7
8
STA id
9
10
11
0
12
2
1
2
3
4
5
6
7
8
STA id
9
10
11
12
Spurious Timeouts?
• Believed that spurious timeouts a problem.
• Once AP fixed, don’t see many problems.
• Fully understand out fix.
• Investigate if there are spurious timeouts.
• What sort of tweaking needed?
• Office/hotspot usage.
3
Previous Work
• Gurtov (01): delay spikes (radio coverage, handover,
prio).
• Korhonen (01): Problems in early GSM/GPRS/. . .
• Fotiadis (05): Test delay spikes in 802.11.
• Vacirca (06) and Ricciato (07): Modern GPRS/3G.
Fixes: drive up RTO, Eifel Algorithm, F-RTO.
4
Testbed
STAs
AP
Dummynet
Uploads, WLAN Buffer 20+19, retries 11, fixed rate, long
preamble, fast ACKs.
5
Busted Case: 4 Uploads
120000
700
Observed RTT for monitored STA
rto
Observed RTT for monitored STA
rto
600
100000
500
RTT (ms)
RTT (ms)
80000
60000
400
300
40000
200
20000
100
0
0
0
20
40
60
time (s)
80
100
120
0
20
40
60
time (s)
RT O = srtt + max(200ms, 4rttvar)
Basically no MAC losses.
6
80
100
120
Busted Case: AP Queue
20
20
Observed Driver Queue
Observed Driver Queue
18
18
14
AP Driver Queue Occupancy
AP Driver Queue Occupancy
16
12
10
8
6
16
14
12
4
10
2
0
8
0
20
40
60
time (s)
80
100
120
0
Lost whole flights of ACKs.
Hard for Eifel/F-RTO to help.
7
20
40
60
time (s)
80
100
120
Fixed Case
500
18
Observed RTT for monitored STA
rto
Observed Driver Queue
450
16
400
14
AP Driver Queue Occupancy
350
RTT (ms)
300
250
200
150
12
10
8
6
4
100
2
50
0
0
0
20
40
60
time (s)
80
100
120
0
20
40
60
time (s)
80
100
120
No timeouts, still basically no MAC losses. Less variable?
8
Clearance
0.045
priotised overhead
unpriotised overhead
0.04
0.035
Observed Frequency
0.03
0.025
0.02
0.015
0.01
0.005
0
100
150
200
250
Clearance (ms)
300
350
Both safely positive.
Trade-off between STA and AP draintimes.
9
400
noprior
prior
160
10
drain time
scaled queue size
drain time
scaled queue size
9
140
8
120
7
drain time (ms)
drain time (ms)
100
80
60
6
5
4
3
40
2
20
AP
1
0
0
0
5
10
15
20
0
5
time (s)
10
15
20
time (s)
200
300
drain time
scaled queue size
drain time
scaled queue size
180
250
160
140
drain time (ms)
drain time (ms)
200
120
100
80
150
100
60
40
50
20
STA
0
0
0
5
10
15
20
time (s)
0
5
10
time (s)
10
15
20
Trying Harder
• Maybe 4 uploads not hard enough?
• Other sources of RTT varience?
• Mobility.
• Varying number of users.
• Varying physical rate.
11
More users
Change from 4 to 8, 12 and 16.
3000
3000
3000
Observed RTT for monitored STA
rto
Observed RTT for monitored STA
rto
2500
2500
2000
2000
2000
1500
RTT (ms)
2500
RTT (ms)
RTT (ms)
Observed RTT for monitored STA
rto
1500
1500
1000
1000
1000
500
500
500
0
100
120
140
160
time (s)
180
200
0
100
120
140
160
time (s)
180
200
0
100
120
140
160
180
200
time (s)
RTTs increase: 200ms, 400ms, 600ms and 900–1000ms.
Collision prob also rising: 18%, 35%, 47% and 54%.
12
More users: close up
3000
3000
3000
Observed RTT for monitored STA
rto
Observed RTT for monitored STA
rto
2500
2500
2000
2000
2000
1500
RTT (ms)
2500
RTT (ms)
RTT (ms)
Observed RTT for monitored STA
rto
1500
1500
1000
1000
1000
500
500
500
0
120
122
124
126
time (s)
128
130
0
120
122
124
126
128
130
0
120
122
time (s)
No timeouts (but one failed fast retransmit).
13
124
126
time (s)
128
130
Harsh Rate Control
Change from 11 to 5.5, 2 and 1.
3000
3000
3000
Observed RTT for monitored STA
rto
Observed RTT for monitored STA
rto
2500
2500
2000
2000
2000
1500
RTT (ms)
2500
RTT (ms)
RTT (ms)
Observed RTT for monitored STA
rto
1500
1500
1000
1000
1000
500
500
500
0
100
120
140
160
time (s)
180
200
0
100
120
140
160
180
200
0
100
time (s)
Sharp increase, in line with expectations.
14
120
140
160
time (s)
180
200
Close Up
3000
3000
3000
Observed RTT for monitored STA
rto
Observed RTT for monitored STA
rto
2500
2500
2000
2000
2000
1500
RTT (ms)
2500
RTT (ms)
RTT (ms)
Observed RTT for monitored STA
rto
1500
1500
1000
1000
1000
500
500
500
0
120
122
124
126
time (s)
128
130
0
120
122
124
126
128
130
0
120
time (s)
No timeouts even though RTT quite long.
15
122
124
126
time (s)
128
130
Why Uploads?
12 uploads
fixed 12 uploads
6 Uploads and 6 Downloads for 180 sec
6 Uploads and 6 Downloads
1200
700
600
1000
Throughput(kbits/sec)
Throughput(kbits/sec)
500
800
600
400
400
300
200
200
0
100
1
2
3
4
5
6
7
8
STA id
9
10
11
0
12
16
1
2
3
4
5
6
7
8
STA id
9
10
11
12
Conclusions
• Didn’t find spurious timeouts.
(even after trying hard.)
• Need to check up/down and request/response.
• Buffer size interesting.
• ARP over busy WLAN busted.
17