Staffer Day Template

1
Information Theory for Mobile Ad-Hoc Networks
(ITMANET) The FLoWS Project
Thrust 3 Application Metrics and Network
Performance Asu Ozdaglar and Devavrat Shah
2
New Paradigms for Upper Bounds
Application Metrics and Network Performance
3
Optimizing Application Metrics and Network
Performance

• Objective
• Developing a framework for optimizing
  heterogeneous and dynamically varying application
  metrics and ensuring efficient operation of
  large-scale decentralized networks with uncertain
  capabilities and capacities
• Providing an interface between application
  metrics and network capabilities
• Focus on a direct involvement of the application
  in the network, defining services in terms of the
  function required rather than rates or other
  proxies
• Application and Network Metrics utility
  functions of users-applications, distortion,
  delay, network stability, energy
• We envision a universal algorithmic architecture
• Capable of balancing (or trading off) application
  requirements and network resources
• Adaptable to variations on the network and user
  side
• Operable in a decentralized manner, scalable
• Robust against non-cooperative behavior

Algorithmic Architecture for Optimizing
Application and Network Performance
4
Thrust Areas

• Optimization Methods for General Application
  Metrics
• Design cross-layer optimization algorithms
• Optimize general application metrics (e.g.,
  coupled performance measures, hard-delay
  constraints) subject to physical layer
  constraints
• Jointly optimize over application metrics and
  coding parameters
• Completely distributed and scalable
• Adapt to dynamics in channel characteristics and
  topology
• Stochastic Network Algorithms and Performance
  Analysis
• Design and seamless integration of queuing-level
  and flow-level network algorithms that yield
  desired performance
• Flow-level performance optimization and stable
  network operation
• Game-Theoretic Models and Multi-Agent Dynamics
• Network formation and resource allocation models
  for heterogeneous and non-cooperative users
• Dynamics and performance analysis

5
Thrust AchievementsOptimization Methods for
General Application Metrics

• Cross-Layer Optimization in Wireless Networks
  under Different Application Delay Metrics (Ng,
  Medard, Ozdaglar 08)
• Joint optimization of different user delay
  metrics, packet coding parameters, and physical
  layer parameters
• Motivation In Phase I, random linear network
  coding shown to achieve significant delay (mean
  completion time) gains in rateless transmission
  scenarios, such as file downloads (Eryilmaz,
  Medard, Ozdaglar 07)

6
Thrust AchievementsOptimization Methods for
General Application Metrics

• Cross-Layer Optimization in Wireless Networks
  under Different Application Delay Metrics (Ng,
  Medard, Ozdaglar 08)
• Consider a packet erasure channel with delayed
  acknowledgment feedback
• Introduce a class of different user delay metrics
  and provide an optimization formulation for a
  block-by-block coding scheme
• Illustrate the tradeoffs between different delay
  metrics
• Extend the framework to a multi-user environment
• Joint optimization of packet coding block size
  and power allocation can be formulated as a
  convex optimization problem (and therefore solved
  efficiently) for any convex user delay metric
• In existing literature, true for rate-based
  metrics only in the high SINR regime

7
Thrust AchievementsOptimization Methods for
General Application Metrics

• Wireless Network Utility Maximization(NUM) (Boyd,
  Goldsmith 08)
• Existing wireline NUM theory assumes fixed
  capacity links, steady state operation
• A new wireless NUM theory developed
• NUM/Adaptive Modulation Cross-layer rate and
  power allocation policies for several practical
  modulation schemes
• Dynamic NUM Multi-period model and distributed
  algorithm for dynamic network utility
  maximization with time-varying utilities and
  hard-delay requirements
• Stochastic NUM Optimal control policies in
  random environments Rate-Delay-Energy tradeoffs
  Distributed control policy developed based on
  model predictive control.

8
Thrust AchievementsOptimization Methods for
General Application Metrics

• Resource Allocation in Fading Multiple Access
  Channels (Parandehgheibi, Medard, Ozdaglar 08)
• Efficient resource allocation over the
  information theoretic capacity region of multiple
  access channel to maximize a general concave
  utility function of transmission rates
• Efficient algorithms that rely on optimization
  methods and rate-splitting idea
• Algorithms use channel state information (does
  not rely on queue-length information)
• Demonstrated superior rate of convergence
  performance for limited duration communication
  sessions

9
Thrust AchievementsStochastic Network Algorithms

• Performance Optimization for MaxWeight Policies
  (Meyn 08)
• Maxweight scheduling/routing policies have become
  popular in view of their throughput properties.
  However, these policies are inflexible with
  respect to performance (delay) improvement
• Extended maxweight using general Lyapunov
  functions
• Demonstrated excellent performance on practical
  topologies
• Distributed implementation for wireless models

10
Thrust AchievementsGame-Theoretic Models and
Algorithms

• Local Dynamics for Topology Formation (Johari 08)
• Efficient topology formation for routing in
  wireless networks with decentralized cost
  structures
• Existing work on dynamics for topology formation
  requires global information
• Developed decentralized dynamics for topology
  formation with general cost metrics
• Competitive Scheduling in Wireless Collision
  Channels with Correlated Channel State (Candogan,
  Menache, Ozdaglar, Parrilo 08)
• Competitive scheduling models allow the
  flexibility to incorporate different user
  objectives, but focus on users with independent
  channel models
• Developed distributed convergent dynamics and
  equilibrium characterization for competitive
  scheduling with correlated channels, which model
  joint fading effects

11
Inter-Thrust Achievement

• Capacity region of a large wireless network
  (Niesen, Gupta, Shah 08)
• Existing scaling results provide one-dimensional
  characterization of the capacity region in the
  large network limit
• Characterization of dimensional capacity
  region!
• Approach
• Scaling results for networks with random (regular
  enough) node placement and arbitrary demand
• Developed a coding scheme independent of demand
  requirement, effectively achieving network and
  physical layer separation
• More in Shahs focus talk

12
Achievements Overview
Optimization Theory Distributed efficient
algorithms for resource allocation
Boyd Efficient methods for large scale network
utility maximization
Goldsmith Layered broadcast source-channel coding
Medard, Ozdaglar Cross-Layer optimization for
different application delay metrics and
block-by-block coding schemes
Medard, Shah Distributed functional compression
Boyd, Goldsmith Wireless network utility
maximization (dynamic user metrics, random
environments and adaptive modulation )
Medard, Ozdaglar Efficient resource allocation
in non-fading and fading MAC channels using
optimization methods and rate-splitting
Ozdaglar Distributed optimization algorithms for
general metrics and with quantized information
Goldsmith, Johari Game-theoretic model for
cognitive radio design with incomplete channel
information
Shah Capacity region characterization through
scaling for arbitrary node placement and
arbitrary demand
Johari Local dynamics for topology formation
Shah Low complexity throughput and delay
efficient scheduling
Ozdaglar Competitive scheduling in collision
channels with correlated channel states
Meyn Generalized Max-Weight policies with
performance optim- distributed implementations
Game Theory New resource allocation paradigm that
focuses on hetereogeneity and competition
Stochastic Network Analysis Flow-based models and
queuing dynamics
13
Thrust Synergies An Example
Combinatorial algorithms for upper bounds
Shah Capacity region characterization through
scaling for arbitrary node placement and
arbitrary demand
Thrust 1 Upper Bounds
(C,D,E) optimal solution of
Boyd, Goldsmith Wireless network utility
maximization (dynamic user metrics, varying
environments, adaptive/cooperative coding)
Thrust 3 Application Metrics and Network
Performance

• T3 solves this problem
• Using distributed algorithms
• Considering stochastic changes, physical layer
  constraints and micro-level considerations
• Modeling information structures (may lead to
  changes in the performance region)

Medard, Ozdaglar Cross-Layer optimization for
different application delay metrics and
block-by-block coding schemes
Capacity
Delay
(C,D,E)
Thrust 2 Layerless Dynamic Networks
Meyn Generalized Max-Weight policies with
performance optim- distributed implementations
Energy
Algorithmic constraints and sensitivity analysis
may change the dimension of performance region
14
Next Steps

• Multi-period dynamic NUM for optimally
  trading-off metrics such as delay, rate,
  admission costs
• Layers of bipartite graphs as a model for the
  network and resource allocation using scheduling
  and distributed optimization across layers
• Decentralized implementations for generalized
  maxweight policies
• Game-theoretic models for collision channels with
  partial channel state-correlation
• Multicast capacity scaling

15
Roadmap for meeting Phase 2 Goals

• Evolve results in all thrust areas to examine
  more complex models, robustness/security, more
  challenging dynamics, and larger networks.
• Wireless NUM for time-varying user metrics and
  dynamic network conditions
• Demonstrate synergies between thrust areas
  compare and tighten upper bounds and
  achievability results for specific models and
  metrics apply generalized theory of distortion
  and utility based on performance regions
  developed in Thrusts 1-2.
• Joint optimization of coding, delay metrics, and
  power allocation
• Joint optimization of rate and power over
  information-theoretic capacity region of fading
  multiple access channels
• Characterization of capacity region in the large
  network limit
• achievability through coding schemes independent
  of traffic demand
• Demonstrate that key synergies between
  information theory, network theory, and
  optimization/control lead to at least an order of
  magnitude performance gain for key metrics.
• Delay gains of network coding
• Rate-reliability tradeoffs and performance gains
  for wireless NUM
• Performance improvement for generalized Maxweight
  policies