An Energy-Efficient Architecture for DTN Throwboxes

1
An Energy-Efficient Architecture for DTN
Throwboxes

• Nilanjan Banerjee, Mark Corner, Brian N. Levine

University of Massachusetts, Amherst
http//prisms.cs.umass.edu/dome
2
What are Disruption Tolerant Networks ?

• DTNs are sparse networks with low node density
• Transfer data through intermittent contacts
• Nodes are largely disconnected
• Come naturally from the applications they support
• Wildlife tracking
• Underwater exploration and monitoring
• Or from fragility and failures in the network
  itself
• Major natural disasters
• Jamming and Noise
• Power Failure

3
Examples of DTN
UMass DieselNet Burgess et al. Infocom 06
4
Limitations of Mobile DTNs

• Do you have enough capacity in your DTN?
• what can you do about it?
• Most influential factor in DTN performance?
• the frequency and number of contact opportunities
  
• How can we increase contacts?
• more mobile nodes
• change the mobility pattern of nodes
• mobility patterns inherent to a particular
  network

5
Observation
Place a relay and create a virtual contact
Route A
Route B
6
Solution Throwboxes

• Throwboxes stationary battery powered relays
• has radios and storage
• cheap, small, easy to deploy
• solar powerperpetual operation
• Challenges
• where do we place these boxes ? Wenrui et al.
  Mass 06
• make them ultra low power for perpetual
  operation

7
Solution Throwboxes

• Throwboxes stationary battery powered relays
• has radios and storage
• cheap, small, easy to deploy
• solar powerperpetual operation
• Challenges
• where do we place these boxes ? Wenrui et al.
  Mass 06
• make them ultra low power for perpetual
  operation

8
Outline

• Design Goals
• Throwbox Architecture
• Mobility Prediction Engine
• Lifetime Scheduler
• Throwbox Prototype and Deployment
• Experimental Results
• Power Savings
• Routing Performance
• Conclusions

9
Throwbox Design Goals

• Small form factor, portable and cheap
• Can be placed practically anywhere in the
  network
• Design should be general
• Applicable to wide variety of DTNs
• Should not use prior information about
  mobility patterns
• Run perpetually on solar panels of the size of
  the box
• Translates to a small average power constraint
• Optimization goal maximize the number of packets
  forwarded
• for now, purely a local metric, not end-to-end
  delivery

10
Present Approches

• Use PSM on the 802.11 card Anand et. al
  MobiCom 2005
• Neighbor Discovery Cost is huge (gt 95 of
  total energy cost)
• Huge idle cost of the platform hosting the
  card
• Wasted Energy due to wakeups on brief contacts
• Use Dual Radio platforms Jun et. al Chants
  2006
• Huge idle cost of the platform hosting the
  radios
• Short range radio cannot detect a large
  number of contacts

Energy consumption too high for perpetual
operation !
11
Our Approach Tiered Architecture

• Neighbor Discovery
• DTNs are sparse discovery is extremely
  expensive
• Energy is wasted when waking the platform
• Data Transfer
• Requires a powerful WiFi Radio
• High power platform
• Tier-0 (low power) search peers , decide tier-1
  wakeups
• Tier-1 (high power) data transfers and routing

12
Overview
13
Mobility Measurement and Prediction

• Buses transmit pos, dir, and speed.
• Throwbox predicts
• if bus will reach data-range before tier-1 can be
  woken?
• length of time in range

• Track the probability the node enters data-range
  given series of cells it must traverse
• Statistics kept on each cell
• Markovian assumption allows simple calculation

14
Scheduling

• Each contact incurs fixed cost to wake tier-1
  platform.
• Most efficient strategy wake for largest
  contacts
• saves energy, but mostly designed to limit power
• 0-1 Knapsack problem reduces to this scheduling
  problem
• choose items to carry s.t. (?weight capacity)
  and maximizes ?value.
• C1 ... Cn events, each has
• total energy cost ei (weight), bytes transferred
  di (value)
• Energy constraint P t (capacity)
• Solution is subset of events s.t. (?ei Pt) and
  maximizes ?di

15
Token Bucket Approach

• Take this event, next event, or both?
• Token rate average power constraint
• Estimate the size energy cost
• ignore if insufficient tokens
• Compute tokens generated till next event
• based on tracking inter-arrival times
• If sufficient tokens for both events
• take current event
• If current event larger than next connection take
  it
• otherwise wait for next one

16
Prototyping Throwbox

• TelosB Mote (sensor)
• 900 MHz XTend radio
• 8 MHz microcontroller
• Stargate
• 802.11b CF card
• 400MHz PXA255 Xscale
• All DieselNet code
• Rechargeable cells, solar power, energy
  monitoring
• some custom hardware

17
Experimental Setup

• How effective is our energy management design?
• compare with single platform periodic wake up
  (PSM)
• Two-platform with mobility prediction (WoW)
• Can we really run it on solar-power?
• At reduced consumption does it still help?
• use the successful delivery metric
• Use trace-based simulation and deployment
• equipped 40 busses with XTend radios
• placed three Throwboxes for several weeks
• record contact opportunities with buses (both
  radios)

18
Throwbox Placement
Throwbox deployed on bikes in UMassDieselNet
19
Power Savings (equivalent transfers)

• 20x less power than periodic
  wakeup
• 5x less power than just mobility
  prediction

20
Routing performance

• Throwbox at 80mW equivalent to best case.

21
Conclusions

• Placing relays in DTNs can produce huge
  performance boost
• Motivates studies on adding Meshes or
  Infostations to DTN
• Tiered Architecture can produce substantial
  energy savings
• Can lead to 31 times less energy
  consumption
• Need for systems to adapt to variable
  solar power
• Multi-radio systems are energy efficient in
  sparse networks
• Need for more efficient use of the XTend
  channel
• Low bitrate radio can be used to gather packet
  info
• Need to integrate power management into routing

22
An Energy-Efficient Architecture for DTN
Throwboxes

• Nilanjan Banerjee, Mark Corner, Brian N. Levine

University of Massachusetts, Amherst
http//prisms.cs.umass.edu/dome
23
Throwbox Vs Infostations

• Infostations are hotspots connected to the
  Internet.
• Throwboxes are untethered routers in a DTN.
• Infostations build to provide mobile users with
  data of interest.
• Throwboxes act as routers to improve capacity of
  DTNs.
• Infostations designed with the motivation of
  providing always available service for urgent
  messages in cellular networks.
• Throwboxes designed with the motivation to
  engineer large number of contacts in disruption
  tolerant networks.

24
Energy performance

• Need larger cell, but perpetual operation
  possible
• Unanswered questions about solar variation