Using PeertoPeer Data Routing for InfrastructureBased Wireless Networks

1

• Using Peer-to-Peer Data Routing for
  Infrastructure-Based Wireless Networks
• Sethuram Balaji Kodeswaran, Olga Ratsimor,
• Anupam Joshi, Tim Finin, Yelena Yesha
• ebiquity.umbc.edu

2
eBiquity Group Technical Roots
Web Services
DB
SemanticWeb
AI
IntelligentInformationSystems
Mobility
Trust
Networking Systems
Security
Assurance
PervasiveComputing
Privacy
3
Overview

• Introduction
• Motivation
• Network Model
• Typical Usage Scenario
• Numi Framework Overview
• Simulation Results
• Sample Application
• Conclusion and Future Work
• QA

4
Introduction

• Combining infrastructure and ad hoc communication
  to meet mobile users data needs
• Mobile devices offer unused resources to support
  the needs of their peers
• Infrastructure components monitor device mobility
  and orchestrate utilization of available excess
  resources
• Mobile devices use peer-to-peer interactions to
  obtain data carried by peers on their behalf

5
Motivation

• Widespread use of personal devices
• Devices with limited and varying capabilities
• Data-intensive services
• Improved ad-hoc capabilities
• Expensive cellular WAN connectivity
• Increased Popularity of HotSpots (commercial
  community networks)
• Starbucks, Borders, T-Mobile Hotspots,
  PersonalTelco, Seattle Wireless, Consume
• Islands of high speed network connectivity
  separated by areas of no network access

6
Network Model

• Landing Zones are islands of high speed cheap
  network connectivity around a Service Portal
  limited only by that portals wireless range
• Transit Zones are regions where there is no
  network connection
• In Landing Zones, mobile hosts communicate with
  service portals only
• In Transit Zones, mobile hosts communicate with
  each other

7
Network Components

• Service Portals
• Infostations offering high-speed network
  connectivity and hosting services that can be
  used by nearby mobile hosts
• Portals use their wireless capabilities to
  interact with mobile hosts that are in range
• Portals use wireline connectivity to communicate
  among themselves.
• Mobile Hosts
• Mobile hosts are wireless mobile devices that can
  communicate both with infrastructure and
  neighboring peer mobile hosts that are within
  range (ad hoc mode).
• Services
• Service Agents
• Service Data Units
• Service Data Volumes

8
Typical Usage Scenario
9
Network Component Interactions

• Node to Portal Interaction
• A mobile host, in a landing zone, can request a
  set of new services (upon user's request) or it
  can ask for additional data for currently running
  services (transparent to the user)
• Portal to Portal Interaction
• Portals notify neighbors of service/data provided
  to a passing by mobile node. This ensures that
  data that would be needed by mobile hosts are
  properly scheduled to be delivered to them
• Node to Node Interaction
• The Node-to-Node interaction is employed by a
  mobile host to obtain any additional data from
  another mobile hosts in transit zones

10
Numi Framework

• Heartbeat Generator Agent is responsible for
  broadcasting device presence messages
• Location Monitor Agent is responsible for
  identifying whether or not that device is
  currently in a landing zone or a transit zone
• Message Handler Agent is responsible for handling
  the messaging needs of the framework
• Logger Agent records every interaction that takes
  place on the local device

Numi Node Framework
Numi Task Scheduler
Node Heartbeat Generator
Location Monitor
Data Handler Agent
Message Handler
Node Service Manager
Logging Agent
Numi Portal Framework
Numi Task Scheduler
Portal Heartbeat Generator
Location Monitor
Data Handler Agent
Message Handler
Portal Service Manager
Logging Agent
Music Service Agent
11
Numi Framework

• Task Scheduler Agent is responsible for
  scheduling prescribed tasks at various times
• Data Handler Agent is used for transferring data
  volumes between MHs and between an MH and an SP
• Portal Service Agents run on top of our Numi
  platform on SPs and offer services to a user
• Node Service Agent runs within NUMI on an MH
  offering a service to the user
• Service Manager Agent is responsible for managing
  service agents on a platform

Numi Node Framework
Numi Task Scheduler
Node Heartbeat Generator
Location Monitor
Data Handler Agent
Message Handler
Node Service Manager
Logging Agent
Numi Portal Framework
Numi Task Scheduler
Portal Heartbeat Generator
Location Monitor
Data Handler Agent
Message Handler
Portal Service Manager
Logging Agent
Music Service Agent
12
Simulation Environment

• Simulations conducted using GlomoSim
• 802.11 used as MAC protocol
• Geographical region considered is 10 sq
  Kilometers
• Node mobility assumed to be piece-wise linear
• Numi configuration
• Node presence interval 8 sec
• Node zone refresh interval 1 sec
• Portal presence interval 8 sec
• Portal zone refresh interval 1 sec
• Presence message validity 10 secs
• Data Query Manager runs every 100ms to identify
  next data segment needed
• Simulated behavior of an application that is
  non-real time and whose data needs are fairly
  predictable (example playlist, newspaper, eBooks,
  etc)

13
Simulation Experiments

• Compared the performance of the following
  schemes
• Data Hoarding (DH) Device tries to cache as much
  data as needed until it reaches the next portal
  on its route. If device runs out of data, a
  service disruption is assumed to occur
• Ad hoc Querying (AQ) Device tries to cache as
  much data as needed until it reaches the next
  portal on its route .If device runs out data,
  device queries its immediate neighborhood for
  next segment. If query fails, a service
  disruption is assumed to occur
• Ad hoc Querying with Peer Routing (AQPR) Similar
  to above scheme but in addition, devices offer
  their excess capacity to the network portals to
  use for transporting data segments for other peers

14
Service Disruption vs Number of Portals
Device capacity lt 1MB Number of nodes 100 Speed lt
20 mt/sec Data segment size lt .2 MB
15
Service Disruption vs Nodes in Network
Device capacity lt 1MB Number of portals 3 Speed lt
20 mt/sec Data segment size lt .2 MB
16
Service Disruption vs Node Speed
Device capacity lt 1MB Number of nodes 100 Number
of portals 5 Data segment size lt .2 MB
17
Service Disruption vs Total Universe of Responses
Device capacity lt 1MB Number of nodes 100 Number
of portals 5 Speed lt 20 mt/sec Data segment size
lt .2 MB
18
Service Disruption vs Device Memory
Number of nodes 100 Number of portals 5 Speed lt
20 mt/sec Data segment size lt .2 MB
19
Prototype Application

• A Music Jukebox application that allows a user to
  listen to his favorite MP3 playlist on his PDA
• Implemented in Java (Personal Java)
• Network Model Used
• 3 PCs equipped with 802.11b wireless LAN cards
  acting as Service Portals
• 3 iPAQs equipped with 802.11b wireless LAN cards
  acting as Mobile Hosts
• Web Browser on iPAQ used as User Interface
• Service Portals run Tomcat Servlet Engine for
  accepting users requests

20
Mobility Coordinator

• We have developed an additional simulation
  component called the Mobility Coordinator
• Control messages can be sent to any mobile host
  to change its current cell ID
• Only messages tagged with the same cell ID that a
  mobile device belongs to are accepted. Others are
  dropped

21
Conclusion and Future Work

• Utilize a combination of infrastructure and ad
  hoc communication to provide uninterrupted
  services to a user
• Mobile devices offer up excess capacity to the
  network for supporting peers
• Ongoing work
• Group and Multi-hop scheduling
• Route deviations and its effects
• Optimal Portal Placement strategies
• Security and Privacy Issues

22
ebiquity.umbc.edu
23
Typical Usage Scenario
Portal 2
P 5
P 6
P 3
P 4
Portal 1