Upload
buck-short
View
221
Download
0
Embed Size (px)
DESCRIPTION
A Comparison of RaDiO and CoDiO over IEEE WLANs May 25 th Motivation A huge success of video streaming over the Internet Requirements –A small buffer at the receiver –Small delay of user control and short preroll delay Research done in the network parts –Reservation based Routing Protocols : RSVP –QoS support : diffserv, interv –Priority queue at the network layer and the Link Layer Research done in the application parts –Frame reordering, ARQ, RaDiO, CoDiO
Citation preview
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 1
Jeonghun Noh Deepesh Jain
A Comparison of RaDiO and CoDiOover IEEE 802.11 WLANs
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 2
Outline
Introduction– Motivation, RaDiO vs. CoDiO, intuition behind CoDiO
Problem Identification : Dedicated Line vs. Shared WLAN
Channel Estimation Techniques
Simulation Results on dynamic wireline and WLANs
Conclusions
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 3
Motivation
A huge success of video streaming over the Internet
Requirements– A small buffer at the receiver– Small delay of user control and short preroll delay
Research done in the network parts– Reservation based Routing Protocols : RSVP– QoS support : diffserv, interv– Priority queue at the network layer and the Link Layer
Research done in the application parts– Frame reordering, ARQ, RaDiO, CoDiO
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 4
RaDiO vs CoDiO
RaDiO : An optimal transmission policy for each packet determined by minimizing D+λR– D : expected distortion according to the current policies– R : expected transmission rate
CoDiO : An optimal transmission policy for each packet determined by minimizing D+λΔ– D : expected distortion according to the current policies– Δ : expected congestion in the network (last hop)– Delay on the last hop is determined by the size of the queue
CoDiO achieves the same R-D performance yet reduces congestion
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 5
4
4
Why CoDiO performs better?Intuition behind CoDiO
CoDiO and RaDiO send virtually the same set of packetsBig difference comes from how to order them
time
2
1
3
time
1
3
2
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 6
Problem Identification
Why WLANs?– Becoming popular as the last hop– Provide more bandwidth than 3G/4G cellular system
Why hard?– Bandwidth fluctuation– Shared medium with multiple users– No guarantee of QoS service
Differences with the previous work– Available bandwidth is no longer fixed – Multiple users share the same wireless channel
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 7
consideration of WLANs
Channel Capacity fluctuates– mobile movement– interference from electronic devices– Other users sharing bandwidth
For CoDiO, Channel Capacity is important for delay estimation– Determines drain rate of packets at the
bottleneck link
timeQ
ueue
siz
e
Drain rate
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 8
Inputs to RaDiO/CoDiO
RaDiO/CoDiOVideo Info
Network BandwidthOptimal Scheduling Policy
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 9
Channel Estimation Technique 1
Packet-pair to estimate the bottleneck link capacity
tNSC packetssizepacket *)1(*_
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 10
Channel Estimation Technique 2
Loss Delay based algorithm (LDA)– measures current available channel bandwidth
Initial Bandwidth Estimate
Send Packets
Packet LostPacked received
late
Decrease Capacity Increase Capacity
NOYes
)*1(* RflossrateBB old )*)/1(*( max IfCBAIRBB old
B: BandwidthC: CapacityRf: Reduction factorIf: Increment factorAIR: Additive increment rate
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 11
Network Simulation Setup - Wireline
Network path composed of high bandwidth link and low bandwidth last hopChannel Capacity Fluctuating
Random cross traffic
last hop
Video traffic
Acknowledgements
High bandwidth links
PI
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 12
Network Simulation Setup - WLAN
Network path composed of high bandwidth link and wireless last hopChannel traffic varying
Random cross traffic
last hop
Video traffic
Acknowledgements
High bandwidth links
PI
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 13
Video / background traffic setupLayered video stream encoded with H.263+
Network experiments in ns-2 over a 2-link path
22Mbps of exponential cross traffic
T3 link Wireline / WLAN
Video traffic
Traffic over WLAN
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 14
Comparison results : Dynamic Wireline
Simulations using H.263+Rate : 10 fps ( 1 view every 100ms)Sequence : Foreman Layered Coding : BL32kbps,EL32kbpsSequence length : 24sPreroll delay : 600ms
Capacity change (50-34Kbps every sec)Capacity estimation using burst packetsOptimistic (Cap = Cap estimated)
20 22 24 26 28 30 32 3416
18
20
22
24
26Rate-Distortion Curve
Rate (Kbps)
PS
NR
(dB
)
RaDiOCoDiO
0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.2616
18
20
22
24
26PSNR-Congestion Curve
Delay (s)
PS
NR
(dB
)
RaDiOCoDiO
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 15
Comparison results : Dynamic Wireline
Simulations using H.263+Rate : 10 fps ( 1 view every 100ms)Sequence : Foreman Layered Coding : BL32kbps,EL32kbpsSequence length : 24sPreroll delay : 600ms
Capacity change (50-34Kbps every sec)Capacity estimation using burst packetsPessimistic (Cap = Cap estimated*f)
f<1
5 10 15 20 25 30 3510
15
20
25
30Rate-Distortion Curve
Rate (Kbps)
PS
NR
(dB
)
RaDiOCoDiO
0 0.02 0.04 0.06 0.08 0.1 0.1210
15
20
25
30PSNR-Congestion Curve
Delay (s)
PS
NR
(dB
)
RaDiOCoDiO
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 16
4
4
Why CoDiO performs worse?
RaDiO sends packets aggressively at the beginningCoDiO distributes packets over time to reduce congestion
time
2
1
3
time
1
3
2
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 17
Comparison results : WLAN
Simulations using H.263+Rate : 10 fps ( 1 view every 100ms)Sequence : Foreman Layered Coding : BL32kbps,EL32kbpsSequence length : 24sPreroll delay : 600ms
Capacity = 130KbpsBandwidth = 60KbpsNo Traffic
0 5 10 15 20 25 30 35 40 45 5010
15
20
25
30Rate-Distortion Curve
Rate (Kbps)
PS
NR
(dB
)
RaDiOCoDiO
10 20 30 40 50 60 70 80 9010
15
20
25
30PSNR-Congestion Curve
Avg Queue Length (Bytes)
PS
NR
(dB
)
RaDiOCoDiO
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 18
Comparison results : WLAN
Simulations using H.263+Rate : 10 fps ( 1 view every 100ms)Sequence : Foreman Layered Coding : BL32kbps,EL32kbpsSequence length : 24sPreroll delay : 600ms
Capacity = 130KbpsBandwidth = 60KbpsTraffic = 64Kbps
5 10 15 20 25 30 35 40 4510
15
20
25
30Rate-Distortion Curve
Rate (Kbps)
PS
NR
(dB
)
RaDiOCoDiO
0 200 400 600 800 1000 1200 1400 1600 1800 200010
15
20
25
30PSNR-Congestion Curve
Avg Queue Length (Bytes)
PS
NR
(dB
)
RaDiOCoDiO
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 19
Comparison results : WLAN with LDA
Simulations using H.263+Rate : 10 fps ( 1 view every 100ms)Sequence : Foreman Layered Coding : BL32kbps,EL32kbpsSequence length : 24sPreroll delay : 600ms
Capacity = 130KbpsTraffic = 80Kbps (on/off every sec)Bandwidth estimation using LDA
0 10 20 30 40 50 60 7010
15
20
25
30Rate-Distortion Curve
Rate (Kbps)
PS
NR
(dB
)
CoDiO With LDACoDiO Without LDA
0 500 1000 1500 2000 2500 3000 350010
15
20
25
30PSNR-Congestion Curve
Avg Queue Length (Bytes)
PS
NR
(dB
)
CoDiO With LDACoDiO Without LDA
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 20
Conclusion
Summary– R-D curve for RaDiO & CoDiO similar (WLAN)– RaDiO aggressive creating more congestion (WLAN)– CoDiO reduces congestion (WLAN)– CoDiO more sensitive to available bandwidth– Need a good bandwidth estimation technique like LDA
Future work– Design a stochastic model of delay estimation– Analyze the effect of varying Capacity to transmission Rate– More general network configurations
A Comparison of RaDiO and CoDiO over IEEE 802.11 WLANs May 25th 2004 21
Acknowledgement
Professor Girod– Valuable discussion and feedback on intermediate results
Eric Setton– Wonderful code and guidance to a meaningful project result– Allowed us to use his slides for the presentation
Anonymous peer reviewers– Help us to identify the project scope (not too ambitious!)
All the valuable questions and feedback so far!