Joint Energy Management and Resource Allocation in Rechargable Sensor Networks

Similar documents
Towards Achieving Perpetual Operation in Rechargeable Sensor Networks

Parametric fractional imputation for missing data analysis. Jae Kwang Kim Survey Working Group Seminar March 29, 2010

Queueing Networks II Network Performance

Stat 543 Exam 2 Spring 2016

Concepts for Wireless Ad Hoc

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud

Equilibrium with Complete Markets. Instructor: Dmytro Hryshko

Stat 543 Exam 2 Spring 2016

Lecture 21: Numerical methods for pricing American type derivatives

Outline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1]

Lecture Randomized Load Balancing strategies and their analysis. Probability concepts include, counting, the union bound, and Chernoff bounds.

( ) ( ) ( ) ( ) STOCHASTIC SIMULATION FOR BLOCKED DATA. Monte Carlo simulation Rejection sampling Importance sampling Markov chain Monte Carlo

CIE4801 Transportation and spatial modelling Trip distribution

Primer on High-Order Moment Estimators

Lab session: numerical simulations of sponateous polarization

Zhongping Jiang Tengfei Liu

Lecture Notes on Linear Regression

Portfolios with Trading Constraints and Payout Restrictions

OPTIMAL COMBINATION OF FOURTH ORDER STATISTICS FOR NON-CIRCULAR SOURCE SEPARATION. Christophe De Luigi and Eric Moreau

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA

A Computational Viewpoint on Classical Density Functional Theory

Basically, if you have a dummy dependent variable you will be estimating a probability.

IN THE. C. G. Cassandras

An Experiment/Some Intuition (Fall 2006): Lecture 18 The EM Algorithm heads coin 1 tails coin 2 Overview Maximum Likelihood Estimation

Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan

DOWNLINK CELL ASSOCIATION OPTIMIZATION FOR HETEROGENEOUS NETWORKS VIA DUAL COORDINATE DESCENT. Kaiming Shen and Wei Yu

Pricing and Resource Allocation Game Theoretic Models

Priority Queuing with Finite Buffer Size and Randomized Push-out Mechanism

EM and Structure Learning

Report on Image warping

Suggested solutions for the exam in SF2863 Systems Engineering. June 12,

risk and uncertainty assessment

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)

Checking Pairwise Relationships. Lecture 19 Biostatistics 666

Amplification and Relaxation of Electron Spin Polarization in Semiconductor Devices

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

RELIABILITY ASSESSMENT

Hidden Markov Models

Variability-Driven Module Selection with Joint Design Time Optimization and Post-Silicon Tuning

Chapter 3. Two-Variable Regression Model: The Problem of Estimation

En Route Traffic Optimization to Reduce Environmental Impact

Estimation: Part 2. Chapter GREG estimation

Cross-layer Optimization of Correlated Data Gathering in Wireless Sensor Networks

Dynamic Programming. Lecture 13 (5/31/2017)

power systems eehlaboratory

CS 798: Homework Assignment 2 (Probability)

A Link Transmission Model for Air Traffic Flow Prediction and Optimization

Motion Perception Under Uncertainty. Hongjing Lu Department of Psychology University of Hong Kong

Notes on Frequency Estimation in Data Streams

Efficient Optimal Control for a Unitary Operation under Dissipative Evolution

Amiri s Supply Chain Model. System Engineering b Department of Mathematics and Statistics c Odette School of Business

CIS526: Machine Learning Lecture 3 (Sept 16, 2003) Linear Regression. Preparation help: Xiaoying Huang. x 1 θ 1 output... θ M x M

Newton s Method for One - Dimensional Optimization - Theory

A Simple Inventory System

High resolution entropy stable scheme for shallow water equations

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

Fast Multi-Channel Gibbs-Sampling for Low-Overhead Distributed Resource Allocation in OFDMA Cellular Networks

Planning and Scheduling to Minimize Makespan & Tardiness. John Hooker Carnegie Mellon University September 2006

Outline for today. Markov chain Monte Carlo. Example: spatial statistics (Christensen and Waagepetersen 2001)

J. Parallel Distrib. Comput.

Random Partitions of Samples

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14

Maximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models

PROBLEM SET 7 GENERAL EQUILIBRIUM

Cognitive Access Algorithms For Multiple Access Channels

SIO 224. m(r) =(ρ(r),k s (r),µ(r))

Clustering through Mixture Models

I + HH H N 0 M T H = UΣV H = [U 1 U 2 ] 0 0 E S. X if X 0 0 if X < 0 (X) + = = M T 1 + N 0. r p + 1

Robust mixture modeling using multivariate skew t distributions

Lecture 4. Instructor: Haipeng Luo

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that

Computational issues surrounding the management of an ecological food web

Space of ML Problems. CSE 473: Artificial Intelligence. Parameter Estimation and Bayesian Networks. Learning Topics

Durban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications

Testing for seasonal unit roots in heterogeneous panels

An Integrated Asset Allocation and Path Planning Method to to Search for a Moving Target in in a Dynamic Environment

Efficient, General Point Cloud Registration with Kernel Feature Maps

1 The Mistake Bound Model

Department of Computer Science Artificial Intelligence Research Laboratory. Iowa State University MACHINE LEARNING

Statistical tables are provided Two Hours UNIVERSITY OF MANCHESTER. Date: Wednesday 4 th June 2008 Time: 1400 to 1600

Global Optimization of Truss. Structure Design INFORMS J. N. Hooker. Tallys Yunes. Slide 1

Statistical Evaluation of WATFLOOD

Hashing. Alexandra Stefan

Probabilistic Graphical Models

Grover s Algorithm + Quantum Zeno Effect + Vaidman

Markov Chain Monte Carlo Lecture 6

Appendix B: Resampling Algorithms

ECTRI FEHRL FERSI Young Researchers Seminar 2015

Notes prepared by Prof Mrs) M.J. Gholba Class M.Sc Part(I) Information Technology

BIO Lab 2: TWO-LEVEL NORMAL MODELS with school children popularity data

Finite Mixture Models and Expectation Maximization. Most slides are from: Dr. Mario Figueiredo, Dr. Anil Jain and Dr. Rong Jin

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 3, MARCH

Externalities in wireless communication: A public goods solution approach to power allocation. by Shrutivandana Sharma

Continuous vs. Discrete Goods

Ensemble Methods: Boosting

Chapter Newton s Method

Dynamic Systems on Graphs

Queuing system theory

Chapter 4: Root Finding

Transcription:

Jont Energy Management and Resource Allocaton n Rechargable Sensor Networks Ren-Shou Lu, Prasun Snha and C. Emre Koksal Department of CSE and ECE The Oho State Unversty

Envronmental Energy Harvestng Many technologes avalable for harvestng energy n dfferent forms, etc.

Energy Management - Dfferent Paradgm Classcal data Wth replenshment data P n = (t) P out transcever P =ρ (t) n P out (t) transcever B(t) B(t) sensor nfo sensor nfo lfetme: < purpose: max R lfetme utlty(t) dt lm T max 1 T utlty(t) dt undesrable: empty battery empty or full battery challenge: extend lfetme perpetual operaton wth varable P n R T Q: Energy management a straghtforward extenson or fundamentally dfferent? A: Close to the latter.

Network Utlty Maxmzaton r k f j f j j s P n =ρ ( t) (t) B node P out (t) Σj f j Σ j transcever f j +r sensor r max r,f j P log r P subject to j f j P j f j + r, s flow balance P n X λ (tx) j f j + X λ (rx) j f j +λ (sn) r energy conservaton j j {z } f Π fxed xmt and recv power achevable rate regon But P n s nether a constant nor perfectly known Battery s fnte

Dynamc Problem τ ρ (t) π ( e) π long term tme Opton E n Issue (1) ρ (t) complexty/convergence overhead (2) π long-term fnte battery extended (3) ˆπ(e) = 1 τ P eτ t=(e 1)τ+1 ˆρ (t) perods of dscharge fndng optmal tradeoff between overhead and dscharge probablty

Overhead and Dscharge Rate -5-1 Computed utlty -15-2 -25-3 -35-4 -45 QuckFx Standard dual-based algorthm -5 1 2 3 4 5 6 7 8 9 1 Iteratons Too slow a convergence n general networks. Need to choose τ 1 Hgh battery dscharge rate Need to sacrfce performance for perpetual operaton Convergence: Assume/generate DAGs Dscharge rate: SnapIt

Resource Allocaton over DAGs w j : fracton of node traffc over lnk j z k (w): fracton of node traffc over node k z k (w) = w j w jk w j z k j k s rewrte problem n terms of z and w The structure of the DAG enables effcent solutons QuckFx Algorthm Dual decomposton subgradent-based dstrbuted soluton Effcent jont updates explotng DAG structure

QuckFx - Dual Decomposton Approach Notes: MWM scheduler: weght s a combned battery/data queue state

QuckFx - Dual Decomposton Approach Notes: Two phases of QuckFx teratons: 1. Aggregate prces: parents chldren, update r 2. Aggregate traffc: chldren parents, update prces

QuckFx - Dual Decomposton Approach Notes: -5-1 Computed utlty -15-2 -25-3 -35-4 -45 QuckFx Standard dual-based algorthm -5 1 2 3 4 5 6 7 8 9 1 Iteratons

SnapIt - Localzed Energy Management π (e+1) π (e+2) Cumulatve energy msmatch: P e π ( e) e τ ( e+1)τ ( e+2)τ h Peτ tme t=(e 1)τ+1 ρ (t) ˆπ (e) Issue - fnte battery sze causes: Unbased estmator battery drft hgh dscharge rate Based estmator very hgh dscharge rate or neffcent replenshment Soluton: Adaptvely control drft

SnapIt - Localzed Energy Management B(t) M/2 r δ B(t) > M/2 r + δ δ M M/2 B (t) δ u u u d utlty δ r* δ r Q1: What performance s lost due to r δ? Q2: How much s dscharge probablty reduced wth δ drft? Theorem: If the varance, σ 2 ρ var 1 Pτ Q τ Q t=1 ρ (t) s bounded and the utlty functon s the log functon, U( ) = log( ), then, gven any β 1, SnapIt acheves p SnapIt and Ū Ū SnapIt log M = Θ M wth the choce of δ = βσ2 ρ log M λ (sn) M. Optmal utlty & low dscharge rate possble smultaneously. (M ) = O(M β )

Smulatons - Parameters A 67-node testbed wth topology created based on an actual local testbed Rechargng profles based on real solar radaton measurements from NREL λ (sn) = 15µW, λ (tx) = 63µW, λ (rx) = 69µW, α =.1, δ =.1r τ = 1 hour and 1 teraton every 5 mnutes

Smulatons - Sum Rate and Network Utlty Sum rate Network utlty Cumulatve full/empty tme Intal battery - hgh Sum of data rates at snk [pkt/s] 12 1 8 6 4 2 QuckFx w/o SnapIt QUckFx w/ SnapIt Instantaneous opt 6: 1: 14: 18: Tme Network utlty 2 1-1 -2 QuckFx w/o SnapIt QuckFx w/ SnapIt Instantaneous opt -3 6: 1: 14: 18: Tme Cumulatve battery full tme [s] 18 16 14 12 1 8 6 4 2 Node 1 w/o SnapIt Node 1 w/ SnapIt Node 2 w/o SnapIt Node 2 w/ SnapIt 6: 1: 14: 18: Tme Intal battery - low Sum of data rates at snk [pkt/s] 12 1 8 6 4 2 QuckFx w/o SnapIt QUckFx w/ SnapIt Instantaneous opt 6: 1: 14: 18: Tme Network utlty 2 1-1 -2 QuckFx w/o SnapIt QUckFx w/ SnapIt Instantaneous opt -3 6: 1: 14: 18: Tme Cumulatve down tme [s] 3 25 2 15 1 5 Node 1 w/o SnapIt Node 1 w/ SnapIt Node 2 w/o SnapIt Node 2 w/ SnapIt 6: 1: 14: 18: Tme

Smulatons - Sunny/Cloudy Day vs. IFRC Sum rate Network utlty Sunny day Sum of data rates at snk 6 5 4 3 2 1 QuckFx w/ SnapIt IFRC Network utlty -5-1 -15-2 -25-3 -35 QuckFx w/ SnapIt IFRC 6: 1: 14: 18: Tme -4 6: 1: 14: 18: Tme Cloudy day Sum of data rates at snk 6 5 4 3 2 1 QuckFx w/ SnapIt IFRC Network utlty -5-1 -15-2 -25-3 -35 QuckFx w/ SnapIt IFRC 6: 1: 14: 18: Tme -4 6: 1: 14: 18: Tme

Summary and Future Work QuckFx + SnapIt to acheve optmal network utlty Addressed convergence ssue assocated wth varable replenshment Addressed (fnte) battery dscharge rate due to varable replenshment Can we mprove the convergence ssue wthout the DAG assumpton/constructon? Can we combne battery control wth data queue control to acheve low dscharge and data buffer overflow rates?