Transcript Document
On Accurate Measurement of Link
Quality in Multi-hop Wireless
Mesh Networks
Kyu-Han Kim and Kang G. Shin
Real-Time Computing Laboratory
Department of EECS, The University of Michigan
September 25, 2006
ACM MobiCom
Accurate Measurement of Link Quality
Routing protocols
Quality-of-Service
Fault diagnosis
Channel assignment
Internet
Gateway 1
Gateway 2
Accuracy
Efficiency
Asymmetry
Feasibility
Focus of this work
Present a novel link-quality measurement framework
Show potential benefits of the framework
`
2
Outline
Limitations
Approach
Evaluation
Conclusion
3
Limitations
Broadcast-based Active Probing (BAP)
▪ Based on inexpensive broadcast
▪ Easy to implement at all layers
LAB S AB S BA
A
B
Different PHY settings [Aguayo04]
Bidirectional measurements
BAP
A
Data
B
SBA=0.6
LAB = LBA = 0.54
A
SAB=0.9
B
ACK
LAB= 0.9
4
Limitations
Unicast-based Active Probing
▪ Same PHY settings as data transmissions
▪ Unidirectional measurement (LAB≠ LBA)
A
B
Capacity overhead (i.e., O(n) vs. O(1) )
Blind to underlying retransmission at MAC
Self-monitoring data frame transmission
▪ Reduce probing overheads
▪ Use unicast and unidirectional results
Require active probing for probing idle links
Blind to underlying retransmission at MAC
5
Outline
Limitations
Approach
Evaluation
Conclusion
6
EAR: Efficient and Accurate link-quality monitoR
EAR
▪ exploits existing traffic by adaptive selection of passive,
active or cooperative measurement scheme
uses unicast packets and derives unidirectional results
▪
▪ is easily deployable and places itself at a network layer and
a device driver for cross-layer interactions
Mesh Router
Inner EAR or iEAR
IP
EAR
Device driver
Outer EAR or oEAR
MAC / PHY
7
EAR Design and Operations
Tegg ≥ Pthresh
MeasureMeasureTcrss ≥ Cthresh Cooperative
period (i)
Cycle (i)
Tcrss < Pthresh Tcrss ≥ Cthresh
UpdateActiveperiod (i)
Time
Tcrss ≤ Cthresh
Techniques
Passive
Cooperative
Active
Routing-table
Manager
Link State
Table
Passive
Tegg < Pthresh
Task
Processor
Outgoing traffic
Incoming traffic
oEAR MAC
Tegg ≥ Pthresh
Task
Timers
iEAR
Distributed measurement
Hybrid techniques
Unicast-based results
Cross-layer interaction
Management Information Base at MAC
Data frame transmission results
Link quality of interest
Link capacity: Data transmission rate
Delivery ratio: d = NS/NT
8
Measurement Techniques (1)
Passive scheme
A
B
C
Monitoring at a device driver
Interaction with MAC’s MIB
Obtaining transmission results
Link-state table at B
Time
Links Scheme Ratio Data rate
BA Passive 0.9 11 Mbps
9
Measurement Techniques (2)
Cooperative scheme
A
B
C
Selective overhearing
Overhearing cross traffic
Reporting overhearing results
Time
Link-state table at B
Links Scheme Ratio Data rate
BA Passive 0.9 11 Mbps
BC
Coop
0.9 11 Mbps
10
Measurement Techniques (3)
Active scheme
A
B
C
Minimizing probe overheads
Adaptive active probing timer (ET)
Using a cooperation technique
Link-state table at B
Time
Links Scheme Ratio Data rate
BA
Active
0.9 11 Mbps
BC Active-Co 0.9 11 Mbps
ET=rand[0,W]
P
W=2
W=4
W=1
P
P
P
P
Cycle
11
Outline
Limitations
Approach
Evaluation
Conclusion
12
Performance Evaluation
Implementation
▪ Linux kernel-2.4.20 (Netfilter and Orinoco device driver)
▪ ETX and ETT routing metrics
N
W
E
▪ BAP for comparison
N4 N8
S
Testbed
▪
▪
▪
▪
Corridor
2nd floor of EECS Building
N10
10 mesh nodes
IEEE 802.11b PCMCIA
N5
N2
Other public networks (802.11b/g)
N9
N7
N1
Offices
N6
N3
Evaluation Metrics
▪ Accuracy, asymmetry-awareness, and efficiency
13
Characteristics of Link Asymmetry
Link asymmetry is common
diff =| SF– SB |
duration
Wireless link-quality has different degrees of quality
asymmetry with different amounts of asymmetry duration
14
Accuracy
Comparison between BAP and EAR
▪ BAP: 10.2% error
▪ EAR: 1.6% error
SN1N2
N1
N2
LN1N2
EAR reduces measurement error from 4 to 20 times,
compared to BAP, and provides unidirectional results
15
Asymmetry Awareness
EAR improves end-to-end throughput
N
W
E
N4
S
N8
Corridor
BAP
N5
N10
N9
EAR
N7
N1
N2
Offices
N6
N3
Benefits
▪ Goodput improvement
▪ 12.9~35.2% (1-hop), 114% (3-hop)
▪ Thanks mainly to unidirectional measurements of EAR
EAR helps routing protocols identify/use asymmetric links
16
Efficiency
Probing overheads
▪ Large number of neighboring
▪
▪
nodes in 200m x 200m
No egress/cross traffic
Thanks to cooperation and
exponential back-off timers
Use of data traffic for measurements
13 times more
measurement
traffic than BAP
owing to hybrid
approach
17
Outline
Limitations
Approach
Evaluation
Conclusion
18
Conclusion
EAR solves problems of varying and asymmetric wireless
link-quality in wireless mesh networks
EAR is a hybrid measurement framework that efficiently
and accurately measures wireless link quality
EAR’s link-asymmetry-awareness improves end-to-end
throughput by up to two times
EAR is useful for wireless network protocols, such as
routing, QoS support and network diagnosis
Remaining Issues
▪ Measurement of other QoS parameters (e.g., latency)
▪ Extension for MANETs
19
Any questions?
Thank You !
Contact:
Kyu-Han Kim ([email protected])
Real-Time Computing Laboratory
(http://kabru.eecs.umich.edu)
20