Intelligent Interface Switching among Heterogeneous Wireless

Networks for Vehicular Communications

Kevin C. Lee, Xiaoqing Zhu, Lillian Dai, Jiang Zhu, Sateesh Addepalli, Rong Pan and Flavio Bonomi

Advanced Architecture & Research

OUTLINE

The future with smart vehicles Intelligent interface switching Problem formulation

System modelOptimization objectiveSDP-based policy

Performance evaluation Conclusions

PROBLEM OF TODAY’S VEHICLES• Safety

• 40,000 deaths/year • 6,000,000

crashes/year • Leading cause of

death for ages 4 to 34

• Mobility• 4,200,000,000 hours of

travel delay• $80,000,000,000 cost of

urban congestion

• Environment• 2,900,000,000 gallons of

wasted fuel

THE FUTURE WITH SMART VEHICLES

Internet

servers

network controllers

…… road-side-uits

(RSUs)

clients

CHALLENGES OF MULTI-HOMING

network controllers

in-vehicle devices

on-board unit

WiFi

WiFi

multiple 3G connections (AT&T, Verizon, Sprint, etc.)

WiMax

KEY COMPONENTS OF AN OBU

transport optimization

mobility manager

interface manager

WiFi 3G WiMax 11p

video streaming

file transfer

safetymessaging

application prioritization

…

user preference specificatio

n

reward-to-gocurrent reward

expected future reward

STOCHASTIC DYNAMIC PROGRAMMING (SDP): AN INTRODUCTION

P(s’|s,a)State: s

Action: a

Reward: r(s, a)

State: s’

…

…

…

SYSTEM STATES

Time

3GWiFi

…Rate

DECISION VARIABLES

REWARD FUNCTIONS

data over 3G:

data over WiFi:

monetary:

performance:

overhead:

OPTIMIZATION OBJECTIVE

user-specified weighting factors

SIMULATION EVALUATION IN QUALNET

AP AP AP3G

25mph

BG CBR traffic

BG CBR traffic

BG CBR traffic

FTP

Server

TRADEOFF ACHIEVED BY ALL POLICIES

PERFORMANCE VS. MONETARY REWARDS

SDP-BASED POLICY VS. HEURISTICS

Varying importance of performance Varying importance of overhead

CONCLUSIONS

A novel scheme for policy-driven optimized interface switching Balances multiple objectives: monetary cost, data

transfer efficiency, and low switching overhead Can be flexibly customized to reflect user preferences Accounts for characteristics of both current and future

interface connection quality Consistently outperforms greedy heuristic schemes

Future work: integrate with intelligent vehicular routing schemes