Title: Multipath Routing of Multimedia Data Over Ad Hoc Wireless Networks
1Multipath Routing of Multimedia Data Over Ad Hoc
Wireless Networks
2Outline
- Ad hoc wireless networks
- Routing for ad hoc wireless networks
- Source routing and multi-path routing
- Example with Dynamic Source Routing (DSR)
- Congestion optimized stream routing
- Complete/practical solution
- Comparison with a heuristic scheme
- Video distortion model
- Experimental results
- Network congestion
- Two applications data download video streaming
- Conclusion
3Wireless Ad Hoc Networks
- Collection of wireless nodes with no
infrastructure - Every node can be source, destination or relay
- Many applications search and rescue, disaster
areas
4Ad Hoc Wireless Network Model
15-node network
If nodes are not limited in their transmissions,
we get the following formulas
- Assumptions
- Static nodes
- Bandwidth 2.2 MHz
- Interference limited network
- Every node has global information
-
5Existing Routing Algorithms
- Routing for wireless ad hoc networks
- DSDV
- AODV
- DSR
- TORA
- Optimization and routing
- Flow assignment
- Resource allocation
Perkins and Bhagwat, 1994
Perkins and Royer, 1999
Johnson and Maltz, 1996
Park and Corson, 1997
Kleinrock, 1976, Bertsekas Gallager, 1987
Xiao, Johansson and Boyd, 2002
6DSR example
- DSR allows to discover and maintain routes on ad
hoc wireless networks
- Example route from node 1 to node 5
- If no route is cached the route discovery
protocol is initiated - route request broadcast
- intermediate nodes append their address and
re-broadcast the reply - reply is sent back by the first node which knows
how to reach the destination
7Congestion Optimized Stream Routing
- Congestion may be estimated by the average
queuing delay for a packet on the network - Average delay over a link for the M/M/1 model
- So minimizing the congestion results in the
following problem
i
Cross traffic Fij
Optimal fij ?
j
- Subject to
- rate constraints at the source and destination
- flow conservation, flow positivity
- capacity constraints Fij fij lt Cij
8Illustration
Source
i
Cross traffic Fij
Optimal fij ?
j
Destination
9Solution Example
- Streaming 100 kbps from node 1 to node 5
10Solution Example
- Properties of the solution
- Diversity of paths
- Complex
- Linear convergence
- To lower the complexity
- Predetermine a set of paths
- Optimal flow partition among the paths
- Instantaneous convergence
Example over 3 paths 1 2 10 4 5 1 3 7 6 5
1 2 9 8 15 5
11Which Set of Paths ?
- First method
- Solve the initial problem
- Extract the routes carrying the most traffic
recursively - Partition over the k best
- Practical method based on DSR
- Discover multiple routes through route
request/reply broadcasts - Add the link state to the nodes address
information - Partition over these routes
12Comparison to Load Balancing
- Heuristic scheme for traffic partitions
- Load balancing among bottleneck links on each
path - Oblivious to joint links, length of path etc
3-path routing
6-path routing
13Video Distortion Model Encoder
- Encoder distortion
- MSE measure for distortion
- D0, ? and R0 are estimated via regression
Stuhlmuller 2000
14Video Distortion Model Packet Loss
- Prand is random packet loss rate
- M/M/1 model for delay
- ? is related to coding structure
- C is the maximum rate supported by the routes
- Ttarget is determined from empirical data
15Video Distortion Model Combined
Encoder MSE
Trans. MSE
- Decode video quality is limited by encoder
performance at low rate, and network congestion
at high rates
16Network Simulation Setup
- Two application scenarios
- Data download
- Live video streaming
- NS-2 configuration
- 15 static nodes, routing from node 1 to node 5
- No random packet loss, no propagation delay
- UDP connection, source routing
- M/M/1 model for data and cross traffic
- Constant Bit Rate (CBR) traffic for video
- http//www.isi.edu/nsnam/ns/
17Data Download Results Congestion
Traffic model M/M/1 No. of samples 600,000
18Video Streaming Results Congestion
Sequence Foreman QCIF Sequence length 250
f. Codec H.26L TML 8.5 Frame rate 30
fps Playout deadline 500 ms Packetization 1
f./packet Traffic model CBR No. of
realizations 400 No random loss
19Video Streaming Results End to End Delay
average end to end delay
90th percentile end to end delay
Traffic model CBR No. of realizations 400
20Video Streaming Results RD Performance
Sequence Foreman QCIF Sequence length 250
f. Codec H.26L TML 8.5 Frame rate 30
fps Playout deadline 500 ms Packetization 1
f./packet Traffic model CBR Number of
realizations 400 Packet loss rate 0
21Video Streaming Results Sequence
Comparison of the operating point for different
number of paths
Foreman QCIF Sequence
1 path 80 kbps, PSNR 32.5 dB
3 paths 187 kbps, PSNR 36.2 dB
6 paths 278 kbps, PSNR 38 dB
22Video Streaming Results Sequence
Comparison of the operating point for the
heuristic and proposed schemes
6 paths heuristic 187 kbps, PSNR 36.2 dB
6 paths optimal 278 kbps, PSNR 38 dB
23Video Streaming Results GOP length
Sequence Foreman QCIF Sequence length 250
f. Codec H.26L TML 8.5 Frame rate 30
fps Playout deadline 350 ms Packetization 1
f./packet Traffic model CBR Number of
realizations 400 Packet loss rate 1
24Conclusions
- Congestion optimized stream routing
- Convex optimization formulation
- Efficient utilization of limited bandwidth
resource - Outperforms heuristic load balancing
- Video distortion model
- Combines the influence of encoder distortion with
packet delay - Predict and compare behaviors of different coding
schemes - Network simulation
- Applied to data download and live video streaming
- Demonstrated the advantage of the proposed
routing scheme - Verified the video distortion model