ashu sabharwalrice university capacity and fairness in multihop wireless backhaul networks ashu...
Post on 02-Jan-2016
218 Views
Preview:
TRANSCRIPT
Ashu Sabharwal Rice University
Capacity and Fairness in Multihop Wireless Backhaul Networks
Ashu SabharwalECE, Rice University
Ashu Sabharwal Rice University
Wireless Utopia:Mobile Broadband
• WiFi Hot-spots– Reasonable speeds – Expensive + poor coverage low subscriber rates,
failing companies,…
• 3G– Ubiquitous, allows mobility but low data rates– Expensive to deploy slow deployments
• Major costs– Wired connection to backbone– Spectral fees– Uneasy “on-demand” growth
Ashu Sabharwal Rice University
Transit Access Points:Multi-hop Backbone
• Few wires– Most TAPs multi-hop to wired gateways– Add wires to TAPs as demand grows
• Use both licensed and unlicensed spectrum– Licensed spectrum: protected, allows guarantees– Unlicensed spectrum: free, more, less interference
outdoors
Multiple radios& MIMO
Ashu Sabharwal Rice University
Major Challenges
• High information density around wires– Capacity per gateway log(n)
• Service quality transparent to user location– Users close to wire can win big– TCP on RTT time-scale, too slow
Ashu Sabharwal Rice University
Characteristics of TAP Networks
• No mobility in backbone– TAPs don’t move static topology
• Slow variability can be used at all time-scales– Physical layer can use fast feedback – Medium access could be topology aware– Qos routing can be reliably done
Opportunity for optimization based on topologyvia feedback at multiple time-scales
Ashu Sabharwal Rice University
Outline
• Opportunistic Cooperative Relaying [Sadeghi,Chawathe,Khoshnevis,Sabharwal]
– Route diversity– Cooperative PHY– OCR
• TAP Fairness [Gambiroza,Sadeghi,Knightly]
– Performance of current protocols– Inter-TAP fairness model
• Rice TAP Testbed
Ashu Sabharwal Rice University
Multi-hop Networks
• Multiple routes to destination– Many routes exist to destination– Route quality function of time
• Coherence time – Time for which channel SNR remains constant– For low mobility channels, several packets long
Route diversity
0
1
2
3
Ashu Sabharwal Rice University
Cooperative PHY
• Why use only one route every time ?– Carrier sense will shut off many TAPs– Use their power and antenna resources
0
1
2
3
Ashu Sabharwal Rice University
Cooperative PHY
• Send packet(s) to other TAPs
0
1
2
3
Ashu Sabharwal Rice University
Cooperative PHY
• Send packet(s) to other TAPs• All TAPs together “forward” the packet
– Acts like a 3M x M antenna system (in above picture)– Simplest form of network coding
0
1
2
3
Ashu Sabharwal Rice University
Throughput Gains
• Rule: Choose best “k-out-of-m” routes leading to minimum total delay
• Substantial gains for moderate network size
Maximum Available Routes
Th
roughput
(Mbit
s/s)
~60%
~70%
Ashu Sabharwal Rice University
Challenges in Realizing Route Diversity
• Quality of routes unknown– Use of a route depends on its current condition– Thus, routes have to measured before every use
• Multiple TAP coordination– Medium access has to coordinate multiple TAPs
• Knowledge of routes– Many routes exist– Which subset to actively monitor ?
Ashu Sabharwal Rice University
Opportunistic Cooperative Relaying
• 4-way multi-node handshake– Allows source (TAP 0) to know all channel qualities– AND coordinate participating TAPs– TAP 0 chooses the smallest delay route
• Multi-hop MAC– Forwarded packets do not contend again– Slot reservation ensures safe passage to destination
Ashu Sabharwal Rice University
Throughput Performance
• Throughput gains (20-30%) outweigh spatial reuse loss
• 2-4 routes give max gain due to handshake overhead
Distance from source (d)
Th
roughput
(Mbit
s/s)
0 12
3
200 m
d
2-hop 802.11
2-route OCR3-route OCR
4-route OCR
Ashu Sabharwal Rice University
Outline
• Opportunistic Cooperative Relaying [Sadeghi,Chawathe,Khoshnevis,Sabharwal]
– Route diversity– Cooperative PHY– OCR
• TAP Fairness [Gambiroza,Sadeghi,Knightly]
– Performance of current protocols– Inter-TAP fairness model
• Rice TAP Testbed
Ashu Sabharwal Rice University
Unfairness in Current Protocol
• IEEE 802.11, 5 MUs/TAP • TAP1 completely starved
– Same for TCP– Caused mainly by information assymetry
• In general, closest to the wire TAP wins
TAP1 TAP2 TAP3
TA(1)
TAP4
TA(2)TA(3)
MU1 MUn1 MUn4MU1MUn3MU1MUn2MU1
Internet
...
...
...
...
0
399
518
917
0
400
800
1200
TAP1 TAP2 TAP3 Total
Goodput [kb/sec]
UDP/CSMA
Ashu Sabharwal Rice University
Inter-TAP Fairness
• Ingress Aggregation– Flows originating from a TAP treated as one– TAPs implement inter-flow fairness
• Temporal fairness– Different links have different throughputs– Throughput fairness hurts good links
• Removal of Spatial Bias– Equal temporal share not sufficient– More hop flows get lesser bandwidth
Ashu Sabharwal Rice University
Throughput with Temporal Fairness
• Temporal Fairness– Equal time shares to all flows– Flow receives 1/F of the throughput of the case it was the
only flow
• Shares: 18%, 21%, 61%
• Increase in number of hops decrease in throughput
TAP1 TAP2 TAP3
TA(1)
TAP4
TA(2)TA(3)
Internet
∑=
=fh
i i
f
CF
T
1
1
1
20Mbps 5Mbps 10Mbps
Ashu Sabharwal Rice University
Removing Spatial Bias
• Spatial Bias Removal (SBR)– Find the bottleneck link of each flow– Share of all flows traversing bottleneck equal
• SBR+Temporal Fair = Equal temporal share in bottleneck links
• SBR + Throughput Fair = Equal throughput for all flows regardless of their paths
Ashu Sabharwal Rice University
Throughput Comparisons
20Mbps 5Mbps 10Mbps
Example
Ashu Sabharwal Rice University
Outline
• Opportunistic Cooperative Relaying [Sadeghi,Chawathe,Khoshnevis,Sabharwal]
– Route diversity– Cooperative PHY– OCR
• TAP Fairness [Gambiroza,Sadeghi,Knightly]
– Performance of current protocols– Inter-TAP fairness model
• Rice TAP Testbed
Ashu Sabharwal Rice University
TAP Hardware Design
• Platform for new PHY + Protocol Design• Generous compute resources
– High-end FPGAs with fast interconnects– Simulink GUI environment for development
• 2.4 GHz ISM band radios– 4x4 MIMO system
• Open-source design– Both hardware and software
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
Ashu Sabharwal Rice University
TAP Testbed Goals
• Prototype network on and around Rice campus• Measurement studies from channel conditions
to traffic patterns
Ashu Sabharwal Rice University
Summary
• Transit Access Points– WiFi “footprint” is dismal– 3G too slow and too expensive– Removing wires is the key for economic viability
• Challenges– Enabling high capacity backbone– Multi-hop fairness
top related