Improving Performance of Wireless Networks
description
Transcript of Improving Performance of Wireless Networks
![Page 1: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/1.jpg)
1
Improving Performance ofWireless Networks
Nitin Vaidya
Joint work with Fan Wu, Tae Hyun Kim, Jian Ni,Vijay Raman, R. Srikant
November 4, 2010
![Page 2: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/2.jpg)
2
What Makes Wireless Networks Interesting?
Many forms of diversity
•Time
•Route
•Antenna
•Spatial
•Channel
![Page 3: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/3.jpg)
3
Multi-Channel Environments
Available spectrum
2 3 4 … c
Spectrum divided into channels
1
![Page 4: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/4.jpg)
Multi-Channel Wireless Networks
Benefits of channelization
g Channel diversity •Gain variations
•Interference mitigation
g Channel access efficiency gain
4
![Page 5: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/5.jpg)
Recent Contributions onMulti-Channel Networks
g Incorporating opportunism in multi-channel networks
g Improving channel utilization
g Game theoretic approach for channel management
5
![Page 6: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/6.jpg)
Opportunistic Routing
![Page 7: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/7.jpg)
Opportunism
g Traditional routing: S R D
g But D may sometimes overheard S R transmission
g No need to forward such packets on R D
7
S R D
![Page 8: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/8.jpg)
Opportunism using MOREg Source sends linear combinations of packets in batches
g Forwarders keep all heard packets in a buffer
g Nodes transmit linear combinations of buffered packets
g Destination decodes once it receives enough combinations
S R D
P1
P2
P3
P1 P2 P3 =+ b + ca a,b,c
2,1,3
0,2,1
2,1,3
P1 P2 P3 =+ 1 + 32 2,1,3P1 P2 P3 =+ 2 + 10 0,2,1P1 P2 P3 =+ 0 + 23 3,0,2
3,0,2
=2 + 1 0,2,1 7,4,92,1,3 + 1 3,0,2
7,4,9
=2 + 2 0,2,1 1,6,62,1,3 - 1 3,0,2
1,6,6
P1
P2
P3
![Page 9: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/9.jpg)
Opportunism versus Concurrency
g For opportunistic scheme to work,nodes must be on the same channel
g Reduces concurrency
9
S R D
![Page 10: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/10.jpg)
Trade-Off
10
Advantages Disadvantages
Opportunism
Exploits broadcast nature
Reduces average # hops
Fewer transmissions
Higher contention
No multiple channel support
Multichannel
Concurrency
Lower contention
No opportunistic overhearing
Potentially longer routes
![Page 11: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/11.jpg)
Example
g Traditional Channel Assignment
S
A
D
0.25
0.5
B
0.250.75
C1
C1
C2
C2
C3
0.75
C3
C3
0.9
End-to-end throughput = 0.5
Loss probability
![Page 12: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/12.jpg)
“Opportunism-Aware” Channel Assignment
S
A
D
0.25
0.5
B
0.2
5
0.75
C1
C1
C1
C2
C2
0.75
C2
0.9
C1 C2
End-to-end throughput = 0.6475
![Page 13: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/13.jpg)
Our Contribution
g Take into account both opportunistic gains obtained by assigning identical channels to the nodes, as well as concurrency gains by assigning different channels
g Extended MORE to a multi-radio multi-channel (MRMC) environment
13
![Page 14: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/14.jpg)
Summary
g Opportunistic schemes can benefit in multi-channel environments
g Channel assignment needs to be opportunism-aware
g Proposed such an assignment scheme
14
![Page 15: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/15.jpg)
15
Packet Size-Dependent Channel Selection
![Page 16: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/16.jpg)
Channel Width
g Typically, channels are assumed identical width
g May benefit by varying channel widths
16
2 3 4 … c1
![Page 17: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/17.jpg)
Motivation
17
Rate-independent MAC overhead
L1 bitsDIFS
)/(Overhead
RLT
T
i
Header
L1 /R
L2 bitsDIFS Header
T
L2 /R
![Page 18: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/18.jpg)
MAC Overhead vs Packet Size
18
Packet size Li
T = 50μs; R = 54 Mbps )/(
OverheadRLT
T
i
![Page 19: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/19.jpg)
Current Approach
g Frame Aggregation (used in IEEE 802.11n)
Aggregate and send multiple packets in a single transmission opportunity
19
L1 bitsDIFS Header L2 bits L3 bits
overhead Multiple packets to amortize overhead
![Page 20: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/20.jpg)
Packet Size-Dependent Channel Widths
20
g Partition a channel into narrow and wide sub-channels
g Use narrow sub-channel for short packets
g Use wide sub-channel for long packets
![Page 21: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/21.jpg)
Proof-of-Concept
g Consider a node (A) communicating withmultiple other nodes
21
A
![Page 22: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/22.jpg)
Proposed Approach
22
1Clients estimate ownshort packet load,and inform node A
Node A estimates aggregate short packet load2
Node A determines partition {BWS, BWL}3
Clients use BWS for short
& BWL for long packets4
![Page 23: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/23.jpg)
Summary
g Channel width selection based on packet size distribution
g Can perform better than frame aggregation
g Ideas can be extended to arbitrary networks
23
![Page 24: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/24.jpg)
CSMA with Imperfect Carrier Sensing
![Page 25: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/25.jpg)
Carrier Sensing (CS)
g Not perfect
g With narrower channels and mobility,fading can be an issue
g What happens to network performance whenCS is imperfect ?
25
![Page 26: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/26.jpg)
Throughput-Optimal Schedulers
g A scheduler is throughput-optimal ifit can serve all schedulable traffic
g Throughput-optimal scheduler byTassiulas-Ephremides’92
•Schedule =
•Computationally complex and centralized solution
![Page 27: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/27.jpg)
Related Work
g Continuous-time CSMA-like algorithm by Jiang-Walrand’08
g Discrete-time CSMA by Ni-Srikant’09
![Page 28: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/28.jpg)
Our Contribution:Preemptive CSMA
g Discrete-time medium accessg Per-packet scheduling decisiong Data packet collisions modeledg Non-zero carrier sense time
Analysis for
g Perfect carrier sensingg Imperfect carrier sensing
![Page 29: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/29.jpg)
Model
g Link-centric model
i Transmission rate is normalized to 1
i One-hop traffic
g Conflict graph to model interference
![Page 30: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/30.jpg)
Medium Access Model
Last α-duration of each time slot for carrier sense
![Page 31: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/31.jpg)
Preemptive CSMA
g Two access probabilities: ai and pi
Carrier sense
u(t): preemptionx(t): transmission scheduleCi: set of conflict links of i
ACK reception
![Page 32: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/32.jpg)
Performance Analysis
g Schedule evolution: discrete-time reversible Markov chain
Stationary distribution
iCu : set of conflicting links of links in u
iWhen pi = 1 - =
exp{wi(qi)} -1exp{wi(qi)}
1exp{wi(qi)}
![Page 33: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/33.jpg)
Throughput-Optimality
g Preemptive CSMA is throughput-optimal
i When access probabilities are
• 0 < aLB ≤ ai ≤ aUB < 1
• pi = 1 - 1/exp{wi(qi)} where wi is a strict concave function with wi(0) = 0
i Proof relies on time-scale separation
•At each time slot, the Markov chain in the steady state
![Page 34: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/34.jpg)
Carrier Sense Failure
g i.i.d. failure events over time slots and links
g Two types of carrier sense failures
•False positive– No activity, but busy state sensed– False positive with probability η
•False negative:– Activity, but idle state sensed– False negative with probability γ
![Page 35: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/35.jpg)
Carrier Sense Failure:Main Result
g By choosing small enough access probability, possible to stabilize arbitrarily large fraction of capacity region
Proof complexity:Markov chain is no longer reversibleUse perturbation theory for Markov chains
![Page 36: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/36.jpg)
Summary
Preemptive CSMA
gGood performance achievable despite imperfect carrier sensing
gSmall access probability overcomes the effect of carrier sensing failures
![Page 37: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/37.jpg)
Where are we now ?
37
![Page 38: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/38.jpg)
38
What Makes Wireless Networks Interesting?
Many forms of diversity
•Time
•Route
•Antenna
•Spatial
•Channel
![Page 39: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/39.jpg)
Wireless Diversity
g This project has furthered our understanding of approaches to wireless diversity using suitable protocols
g We now have a better understanding ofcross-layer protocol design
39
![Page 40: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/40.jpg)
What Remains?
g Physical layer community has also been making significant progress
– Interference alignment– Cooperation– Security
g Need to incorporate these ideas intothe protocol stack
40
Natural continuationof DAWN MURI
![Page 41: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/41.jpg)
What Remains?
41
HigherLayers
UnicastMulticast
PhysicalLayer
DistributedApplications
![Page 42: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/42.jpg)
What Remains?
Much attention to
g Moving bits betweennodes in the network
•throughput
•delay, jitter
•packet loss
g Cross layer ~ Layers 1-2-3
42
HigherLayers
UnicastMulticast
PhysicalLayer
DistributedApplications
![Page 43: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/43.jpg)
What Remains?
g Not as much attention to semantics ofdistributed applications
g How to exploitapplication-awareness ?
43
HigherLayers
UnicastMulticast
PhysicalLayer
DistributedApplications
![Page 44: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/44.jpg)
44
HigherLayers
UnicastMulticast
PhysicalLayer
DistributedApplications
DistributedPrimitives
Wireless Network-AwareDistributed Primitives
![Page 45: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/45.jpg)
Example primitives:gOrdered group communicationgConsensusgAggregationgSynchronizationgCoordination
45
HigherLayers
UnicastMulticast
PhysicalLayer
DistributedApplications
DistributedPrimitives
Wireless Network-AwareDistributed Primitives
![Page 46: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/46.jpg)
Example primitives:gOrdered group communicationgConsensusgAggregationgSynchronizationgCoordination
Network-awarenessgWireless capacity regiongDiversitygBroadcast capabilitygEnergy constraints
46
HigherLayers
UnicastMulticast
PhysicalLayer
DistributedApplications
DistributedPrimitives
Wireless Network-AwareDistributed Primitives
![Page 47: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/47.jpg)
Past Work on Middleware
g Similar motivation
g But optimized for wired networkswith high capacityand more benign characteristics
47
![Page 48: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/48.jpg)
Wireless Network-AwareDistributed Primitives
g Wired algorithms not efficientg Do not exploit wireless capabilities
Many (new) fundamental problems open
48
![Page 49: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/49.jpg)
Distributed Algorithms & Networking
g Overlapping scope
g But cultures differ
49
Communications / Networking
Distributed Algorithms
![Page 50: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/50.jpg)
50
DistributedAlgorithms
Black box networks
Emphasis onorder complexity
Emphasis on “exact”performance metrics
Constants matter
Communications / Networking
![Page 51: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/51.jpg)
51
DistributedAlgorithms
Black box networks
Emphasis onorder complexity
Emphasis on “exact”performance metrics
Constants matter
Information transfer(typically “raw” info)
Communications / Networking
![Page 52: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/52.jpg)
52
DistributedAlgorithms
Computationaffects communication
Emphasis on “exact”performance metrics
Constants matter
Information transfer(typically “raw” info)
Communications / Networking
Black box networks
Emphasis onorder complexity
![Page 53: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/53.jpg)
Picture from Wikipedia
Beneficial to bring together researchers inwireless networking & distributed algorithms
Wireless Network-AwareDistributed Primitives
![Page 54: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/54.jpg)
54
Thanks!
![Page 55: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/55.jpg)
55
Thanks!
![Page 56: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/56.jpg)
56
Thanks!
![Page 57: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/57.jpg)
57
Thanks!
![Page 58: Improving Performance of Wireless Networks](https://reader035.fdocuments.us/reader035/viewer/2022062500/568151d6550346895dc00ed7/html5/thumbnails/58.jpg)
Scheduling Example
A
B
C
PROBE
ACK
DATA
PROBE
ACK
DATA
PROBE
ACK
DATA DATAPROBE
PROBE
Access by aA
Access by aB
Access by aB
Access by pB
Sensed busy by Link A &
C
Preempted by Link
B
Sensed idle by
Link A & C
Preempted by Link A
& CConflict
graph forlinks A, B, C