wlan interface management on mobile devices · wlan interface management on mobile devices hossein...
Post on 01-Aug-2020
1 Views
Preview:
TRANSCRIPT
WLAN Interface Management on Mobile Devices
Hossein Falaki
Master’s Thesis
July 25, 2008
WaterlooUniversity of
Hossein Falaki
WaterlooUniversity of
Motivation
2
Smartphones are proliferating
Multiple Network Interfaces:
Bluetooth
EDGE or 3G
WiFi
Hossein Falaki
WaterlooUniversity of
Motivation
Higher bandwidth
Good energy trade-off
Free
3
Advantages of WiFi
Hossein Falaki
WaterlooUniversity of
Problem
WLAN interfaces consume considerable energy in idle mode
WLAN scanning is highly energy consuming
To discover a WiFi opportunity the WLAN interface should be “up” and “scanning”
4
What is a good strategy for turning the WLAN NIC on and scanning?
Hossein Falaki
WaterlooUniversity of
WLAN Scanning
Passive Scanning:
The interface listens for periodic AP beacons on each channel
Active Scanning:
On each channel the interface sends a broadcast probe request, and waits for probe responses
5
Hossein Falaki
WaterlooUniversity of
Thesis
For background/delay-tolerant applications, static scanning works better than expected due to the power-law distribution.
Context hints can be used through a cache to help interface management.
User-initiated WLAN scans do not appear to incur significant costs.
6
Hossein Falaki
WaterlooUniversity of
Outline
Modeling
Heuristic Strategies
Measurements
Evaluation
Conclusions
7
Hossein Falaki
WaterlooUniversity of
Definitions
1. Medium
2. Availability block
3. Interface states
4. Schedule, T-connected schedule
5. Strategy8
0
20
40
60
80
100
0 200 400 600 800 1000 1200 1400
Access P
oin
t ID
Time (minute)
WIFi
Hossein Falaki
WaterlooUniversity of
Optimal Strategy
Optimal T-Connected schedule
Optimal strategy
Future knowledge assumption
9
Hossein Falaki
WaterlooUniversity of
Greedy
Sort the blocks according to length
Start filling the schedule with the longest blocks
The NIC is off between blocks
10
If blocks are “far apart,” the greedy algorithm finds the optimal schedule:
G
S
...
...g_i
s_i
...
...
Hossein Falaki
WaterlooUniversity of
Dynamic Programming
11
If some blocks are “too close,” it is better not to turn off the NIC.
Time
T1 T2 T3
T2,3
Hossein Falaki
WaterlooUniversity of
Outline
Modeling
Heuristic Strategies
Measurements
Evaluation
Conclusions
12
Hossein Falaki
WaterlooUniversity of
Heuristic Strategies
Naive
Static
Exponential Back-off
Bounded Exponential Back-off
13
Hossein Falaki
WaterlooUniversity of
Naive Scanning
14
Considerable number of scans
Almost zero missed opportunity
Hossein Falaki
WaterlooUniversity of
Static Scanning
15
0
100
200
300
400
500
600
700
800
0 500 1000 1500 0
2
4
6
8
10
12
Nu
mb
er
of
Sca
ns
Mis
se
d o
pp
ort
un
ity (
ho
ur)
Scan interval (seconds)
Number of ScansMissed Opportunity
Hossein Falaki
WaterlooUniversity of
Exponential Back-off
16
Hossein Falaki
WaterlooUniversity of
Bounded EB
17
If availability rate is “too low,” the number of back-offs should be bounded.
Hossein Falaki
WaterlooUniversity of
Outline
Modeling
Heuristic Strategies
Measurements
Evaluation
Conclusions
18
Hossein Falaki
WaterlooUniversity of
Measurements
19
User Behavior
Dynamic Environment
Scheduler
Keyb
oard
/Screen
NIC
1
NIC
2
NIC
3
Hossein Falaki
WaterlooUniversity of
Wireless MeasurementsSix iPhones scanned WiFi and GSM every minute for five weeks
Similar to the Rice measurement (10 WM), with fewer missed samples
20
-200
-150
-100
-50
0
50
0 200 400 600 800 1000 1200 1400 1600
GS
M o
r W
iFi ID
Time (minute)
GSMWiFi
Hossein Falaki
WaterlooUniversity of
Waterloo Dataset
21
0
500
1000
1500
2000
2500
3000
User 6
User 5
User 4
User 3
User 2
User 1
Number of GSM and WiFi IDs visited by Waterloo users
GSM IDsBSSIDs
3070 GSM and 5709 WiFi unique IDsAvg. availability rate: 0.62Avg. missing samples/day: 66
Hossein Falaki
WaterlooUniversity of
Rice Dataset
22
200
300
400
500
600
700
800
900
1000
1100
User 10
User 9
User 8
User 7
User 6
User 5
User 4
User 3
User 2
User 1
Number of GSM and WiFi IDs visited by Rice users
GSM IDsESSIDs
2806 GSM and 3907 WiFi unique IDsAvg. availability rate: 0.48Avg. missing samples/day: 147
Hossein Falaki
WaterlooUniversity of
Block Length
23
0
500
1000
1500
2000
2500
3000
3500
4000
0 1000 2000 3000 4000 5000
Fre
qu
en
cy
Length (seconds)
Histogram of the length of the availability blocks
Waterloo
1
10
100
1000
10000
1000 10000
Fre
quency
Length (seconds)
Log-log histogram of length of the availability blocks
Hossein Falaki
WaterlooUniversity of
Block Length
24
0
500
1000
1500
2000
2500
3000
3500
4000
0 1000 2000 3000 4000 5000
Fre
qu
en
cy
Length (seconds)
Histogram of the length of the availability blocks
Rice
1
10
100
1000
1000 10000
Fre
quency
Length (seconds)
Log-log histogram of length of the availability blocks
Hossein Falaki
WaterlooUniversity of
User MeasurementsTwo BlackBerrys logged user interaction times for about three weeks
25
0
200
400
600
800
1000
1200
0 100 200 300 400 500
Fre
quency
Length (seconds)
Histogram of the length of user activity times
0
100
200
300
400
500
600
0 500 1000 1500 2000
Fre
quency
Length (seconds)
Histogram of the length of inactivity times
Activity Inactivity
Hossein Falaki
WaterlooUniversity of
Outline
Modeling
Heuristic Strategies
Measurements
Evaluation
Conclusions
26
Hossein Falaki
WaterlooUniversity of
Evaluations
Performance metrics:
Number of scans
Missed opportunity
Configurable parameters:
Scanning interval
Maximum back-off
27
Hossein Falaki
WaterlooUniversity of
Comparing Strategies
28
Different Strategies
0
100
200
300
400
500
600
700
800
0 5 10 15 20 25
Num
ber
of S
cans
Time (hour)
NaiveOptimal
EBStatic
Hossein Falaki
WaterlooUniversity of
Number of Scans
29
0
100
200
300
400
500
600
User 6
User 5
User 4
User 3
User 2
User 1
OptimalEB
Static
0
50
100
150
200
250
300
350
400
450
User 10
User 9
User 8
User 7
User 6
User 5
User 4
User 3
User 2
User 1
OptimalEB
Static
Waterloo Rice
Hossein Falaki
WaterlooUniversity of
Missed Opportunity
30
0
0.2
0.4
0.6
0.8
1
Use
r 6
Use
r 5
Use
r 4
Use
r 3
Use
r 2
Use
r 1
EB
Static
0
0.2
0.4
0.6
0.8
1
Use
r 10
Use
r 9
Use
r 8
Use
r 7
Use
r 6
Use
r 5
Use
r 4
Use
r 3
Use
r 2
Use
r 1
EB
Static
Waterloo Rice
Hossein Falaki
WaterlooUniversity of
Simulation Results
Static scanning performs well
Low missed opportunity
Consistently low number of scans
Exponential Back-off performs fewer scans for some users, but with very high missed opportunity
31
Hossein Falaki
WaterlooUniversity of
Discussion
32
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 20 40 60 80 100
To
tal le
ng
th r
atio
Rank (percent)
Rank-size CDF of availability blocks
Waterloo
Hossein Falaki
WaterlooUniversity of
Discussion
33
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Ava
ilab
ility
0 50
100 150 200 250 300
# B
locks
100 200 300 400 500
Rice 10
Rice 9
Rice 8
Rice 7
Rice 6
Rice 5
Rice 4
Rice 3
Rice 2
Rice 1
Wat 6
Wat 5
Wat 4
Wat 3
Wat 2
Wat 1
# S
ca
ns
Hossein Falaki
WaterlooUniversity of
Tuning Static Scanning
34
40
60
80
100
120
140
160
180
200
100 400 700 1000 1300 1600 0
20
40
60
80
100
Num
ber
of S
cans
Mis
sed o
pport
unity (
perc
ent)
Scanning interval (seconds)
User 2
Number of scansMissed Opportunity
Sample Waterloo user:
Hossein Falaki
WaterlooUniversity of
Tuning Static Scanning 2
35
0
50
100
150
200
250
300
350
400
450
500
100 400 700 1000 1300 1600 0
20
40
60
80
100
Nu
mb
er
of
Sca
ns
Mis
se
d o
pp
ort
un
ity (
pe
rce
nt)
Scanning interval (seconds)
User 8
Number of scansMissed Opportunity
Sample Rice user:
Hossein Falaki
WaterlooUniversity of
Caching Scan Results
36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
User 6
User 5
User 4
User 3
User 2
User 1
Hit RatioFalse Positive
False Negative
Hossein Falaki
WaterlooUniversity of
Interactive Processes
37
0
50
100
150
200
250
300
350
User 6
User 5
User 4
User 3
User 2
User 1
Static Scanning with 300 Seconds Interval
No User InteractionUser Interactions
Hossein Falaki
WaterlooUniversity of
Outline
Modeling
Heuristic Strategies
Measurements
Evaluation
Conclusions
38
Hossein Falaki
WaterlooUniversity of
ConclusionsSeparate delay-tolerant and interactive processes
For delay-tolerant applications use static scanning with the largest possible scanning interval
For interactive processes use an aggressive scanning strategy
Use context hints to avoid unnecessary scans
39
Hossein Falaki
WaterlooUniversity of
Future Work
Considering usability of access points
Improving caching
Making interactive scans smarter
Management of multiple NICs
Collaborative scheduling
40
WLAN Interface Management on Mobile Devices
Hossein Falaki
Master’s Thesis
July 25, 2008
WaterlooUniversity of
top related