1. Delay Efficient Wireless Networking
Darpa IT-Manet Project meeting -- Nov. 9, 2006 l1
S1(t) {ON, OFF} l2
S2(t) lN
SN(t)
Michael J. Neely
University of Southern California

2. Outline: My research area -- Stochastic Network Optimization
Recent DARPA IT-MANET results since Summer 2006:
a. Order Optimal Delay for Opportunistic Scheduling
[Allerton Sept. 2006]
b. Non-Equilibrium Capacity and Delay analysis
for mobile ad-hoc networks
[USC CSI Tech Report Oct. 2006]
c. Simulations for Delay Enhanced Diversity Backpressure
Routing (E-DIVBAR)

3. Background on Stochastic Network Optimization:
-Wireless Networks
-Time Varying Channels
-Adaptive Transmission Rates
-Mobility
-Dynamic Power Allocation, Routing, Scheduling for
Maximum Throughput [2003, 2005]
-Fairness and Utility Optimization [2003, 2005]
-Energy Minimization, Average Energy Constraints [2005, 2006]
-General Utility Optimization [2006]
Online Book: (NOW Publishers)
L. Georgiadis, M. J. Neely, L. Tassiulas, “Resource Allocation
And Cross-Layer Control in Wireless Networks,” Foundations
And Trends in Networking, Vol. 1, No. 1, pp , 2006.

4. Cross-Layer Networking1 2 3
Node 1
Node 2
Transport
Layer
Transport
Layer
Network
Layer
Network
Layer
scheduling
Multi-Access
And
PHY Layers
Multi-Access
And
PHY Layers
modulation
and coding

5. Cross-Layer Networking1 2 3
Node 1
Node 2
Transport
Layer
Transport
Layer
optimize
Exogenous green data
Network
Layer
Network
Layer
scheduling
Multi-Access
And
PHY Layers
Multi-Access
And
PHY Layers
R1(c)(t)
node 1
modulation
and coding

6. Cross-Layer Networking1 2 3
Node 1
Node 2
Transport
Layer
Transport
Layer
optimize
Networking,
Multiple Access,
PHY Layers
Network
Layer
Network
Layer
scheduling
Multi-Access
And
PHY Layers
Multi-Access
And
PHY Layers
optimize
optimize
modulation
and coding

7. Cross-Layer Networking1 2 3
Node 1
Node 2
Transport
Layer
Transport
Layer
optimize
Networking,
Multiple Access,
PHY Layers
Network
Layer
Network
Layer
scheduling
Multi-Access
And
PHY Layers
Multi-Access
And
PHY Layers
modulation
and coding

8. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1 l2

9. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1 l2

10. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1
Av. Delay l2
shrinking radius

11. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1
Av. Delay l2
shrinking radius

12. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1
Av. Delay l2
shrinking radius

13. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1
Av. Delay l2
shrinking radius

14. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1
Av. Delay l2
shrinking radius

15. Example: Utility Optimization (Utility/Delay Tradeoffs)1 2 3 4 5 6 7 8 9
l93
l91
l48
l42
Un(c)(t)
Rn(c)(t)
ln(c)
sensor network
wired network
wireless
A general
heterogeneous
network l1
Av. Delay l2
shrinking radius

16. Pav X(t) Example: Virtual Power Queues P(t)
Average Power Constraint:
A wireless device
within a network
P(t)
t=1 t 1
Pav
lim
virtual power queue
P(t)
X(t)
Pav
X(t+1) = max[X(t) - Pav, 0] + P(t)
Stabilizing the virtual power queue ==>
Ensures Avg. Power Constraints Satisfied

17. Optimal Tradeoff Theory:
Capacity and Delay Tradeoffs for Ad-Hoc Mobile
Networks [CISS March 2003, IT June 2005]:
-Optimal Tradeoffs via Redundant Packet Transfers
Energy/Delay Tradeoffs for Multi-User Wireless
Downlinks [INFOCOM March 2006]:
-Extends Berry-Gallager Square Root Law to
Multi-User systems
Utility/Delay Tradeoffs for Wireless Networks
[INFOCOM March 2006, JSAC August 2006]:
-Establishes a “super-fast” Logarithmic Tradeoff Law.

18. Recent Results:
Order Optimal Delay for Opportunistic Scheduling in
Multi-User Wireless Uplinks and Downlinks [Allerton Sept. 2006] l1
S1(t) {ON, OFF} l2
S2(t)
prev. bound
New bound (LCG) lN
SN(t)
Simulation (LCG)
Previous Result: Avg. Delay <= O(N)
New Result: Avg. Delay = O(1) (independent of N)
[Longest Connected Group Algorithm (LCG)]

19. 2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Recent Results:
2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Wireless Mesh Networks [CSI Tech Report, Oct. 2006]
L(1) 1 2 8 7 4 6 3 5
input
rate 9
Arbitrary (perhaps non-ergodic) user mobility.
Channel States iid given the current topology.
Instantaneous Capacity Region L(t)
(assuming users maintain current locations, i.e., fixed topology)

20. 2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Recent Results:
2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Wireless Mesh Networks [CSI Tech Report, Oct. 2006]
L(1) 1 2
L(2) 8 7 4 6 3 5
input
rate 9
Arbitrary (perhaps non-ergodic) user mobility.
Channel States iid given the current topology.
Instantaneous Capacity Region L(t)
(assuming users maintain current locations, i.e., fixed topology)

21. 2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Recent Results:
2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Wireless Mesh Networks [CSI Tech Report, Oct. 2006]
L(1) 1 2
L(2) 8 7 4
L(3) 6 3 5
input
rate 9
Arbitrary (perhaps non-ergodic) user mobility.
Channel States iid given the current topology.
Instantaneous Capacity Region L(t)
(assuming users maintain current locations, i.e., fixed topology)

22. 2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Recent Results:
2. Non-Equilibrium Capacity and Delay Analysis for Ad-Hoc
Wireless Mesh Networks [CSI Tech Report, Oct. 2006]
L(1) 1 2
L(2) 8 7 4
L(3) 6 3 5
input
rate 9
Arbitrary (perhaps non-ergodic) user mobility.
If input rates within intersection of all L(t): Achieve full throughput
with end-to-end average delay that is independent of the timescales
of the user mobility process. (can also treat case when rates are not
within this intersection)

23. Halfway through the simulation, node 0 moves (non-ergodically)
Communiation Pairs: , , … ,
Halfway through the simulation, node 0 moves (non-ergodically)
from its initial location to its final location. Node 9 takes a Markov
Random walk. 1 2 8 7 4 6 3 5 9
Full throughput is maintained throughout, with noticeable delay
increase (at “new equilibrium”), but which is independent of mobility
timescales.

24. More simulation results for the mesh network:
Flow control included (determined by parameter V)

25. Recent Results:
3. Delay Enhanced Diversity Backpressure Routing (E-DIVBAR) [CSI Tech Report, Oct. 2006] 3 2
error 1
broadcasting
(conference version: DIVBAR: CISS March 2006)

26. “Upside Down Layering”: Perform Routing after Transmission
Transport
Layer
Network
Multi-Access
And
PHY Layers
Node 1
Node 2 1 2 3
scheduling
modulation
and coding
Networking,
Multiple Access,
optimize

27. ExOR
[Biswas, Morris 05]
DIVBAR, E-DIVBAR
[Neely, Urgaonkar
2006]

28. DIVBAR almost
doubles throughput
over ExOR.
E-DIVBAR achieves
capacity (as DIVBAR)
but also gets delay
that is better than both
ExOR and DIVBAR.