1. DTN Outdoor Mobile Environment UMass: Mark Corner
DTN Outdoor Mobile Environment UMass: Mark Corner Brian Neil Levine GaTech: Mostafa Ammar Ellen Zegura

2. Overview / Phase II Goals
Prototyping
DTN/Mesh Synergy
Traces
Throwbox power management
DTNRG .net implementation
Deliverables

3. Prototyping: DieselNet
40 equipped buses, routes spanning 150 sq. mi.
Town center (4 sq.mi.) is hub of network.
Each bus:
577Mhz, 256MB ram, b radio b AP, GPS, 40GB drive, XTend radio, GPRS.
Custom software

4. Transfer Opportunities
Red dots: bus-to-bus transfers during 1-month.
Each transfer we record duration, data, location, speed and direction.
Amherst downtown and UMass:
4 sq mi

5. Creating the DOME Cisco 1500 mesh APs Existing 40 buses
Diversity:
Mobility (scheduled / unscheduled)
Nodes (mobile / stationary)
Power (grid / diesel / solar)
Radios ( / XTend / GRPS)
Storage (40GB down to 32MB)
Cisco 1500 mesh APs
Existing 40 buses
PC104 boxes in vehicles
Throwboxes

6. Prototyping: Throwbox
Multi-platform Device:
TelosB Mote (sensor)
900 MHz XTend radio
8 Mhz microcontroller
Stargate
802.11b CF card
400Mhz PXA255 Xscale
64 MB ram, 32MB store
Java 1.3
all DieselNet code
AA rechargeable batteries / solar power

7. Prototyping: Throwbox

8. Trace collection Traces of DieselNet are on the web:
Bytes transferred
Inter-transfer opportunity time
Traces of DieselNet are on the web:
Used in [Burgess-Infocom06], [Jun-Chants06]
To be added: throwbox data, mesh synergy

9. Throwbox Design low-power radio will remain idle unnecessarily.
High power consumption leads to short lifetimes.
Intermittent power is more efficient, but may not improve performance.
An idle, low-power hailing radio that wakes a data radio will save energy. Tx Rx
Idle
Sleep
WaveLan PC card
1.33
0.96
0.84
0.0664
Chipcon CC1000
0.08
0.02
Usual assumption: dense networks
DTNs: contacts are infrequent
low-power radio will remain idle unnecessarily.

10. Throwbox Design
Solution: use model of node mobility to avoid wasted idle periods.
low-power radios are very efficient at duty-cycling.
We can optimize wakeup periods to discover enough contacts to send given traffic load.
After optimizing for both radios, we see a gain in energy efficiency over a single radio.
Hyewon Jun, Mostafa Ammar, Mark Corner, Ellen Zegura. Hierarchical Power Management in Disruption Tolerant Networks with Traffic-Aware Optimization. In Proc. ACM SIGCOMM Wkshp on Challenged Networks (ACM CHANTS). September, 2006.

11. New Approach
Short-range radios waste energy when only a brief contact is available.
By definition, contact opportunities are short-shrifted.
New tiered platform:
tier-0: Mote (8Hmz, 10k,20mW) and XTend radio (~1.5km)
tier-1: Stargate (400MHz, 64MB,2.5W) and b (~150m)
Tier-0 searches for peers and predicts mobility.
Tier-1 manages data transfers (and routing).
Range Rx Tx
Sleep
XTend
1000m
0.36 W
1.0 W
10 mW
802.11
150m
1.2 W
50 mW

12. Approach

13. Mobility Prediction Buses transmit: pos, dir, speed. Box predicts:
Will bus reach data-range before tier-1 can be woken: D1/v > T?
Length of time in range: (P)(D2/v)
P is prob node enters data-range given cells to traverse.
Statistics kept on each cell.
Markovian assumption allows simple calculation.

14. Scheduling is NP-Hard Each contact incurs a fixed cost to wake tier-1.
Most efficient strategy: wake for largest opportunities.
Scheduling corresponds to the knapsack problem:
choose items to carry s.t. (∑weight ≤ capacity) and ∑value is maximized.
{c1...cn } events.
each event ci has
total energy cost ei (weight)
bytes transferred di (value)
Energy constraint P∙t (capacity)
Solution is subset of events that
maximizes bytes transferred ∑di
such that (∑ei ≤ P∙t)

15. Ignored or skipped events
Token Bucket Approach
Tokens arrive as per average power constraint.
Q: take this event, next event, or both?
Algorithm:
1. Estimate the size & energy cost of event.
Ignore if insufficient tokens.
2. Compute expected tokens generated in time for next event.
3. If sufficient tokens for both events, current event is taken.
Otherwise take event if larger than mean.
new tokens
Events
Battery capacity ?
Taken events
Ignored or skipped events

16. Preliminary results Fully functional prototype.
Solar cell produced 65mW over 24 hours (expected 85mW).
Solar cell slightly larger than the box will run perpetually.

17. Routing performance
Three boxes deployed in UMass DieselNet allowed trace-driven evaluations.

18. .Net DTNRG Implementation
Goals:
100% compliant with bundle spec
As compatible as possible with reference implementation (but spec takes precedence)
Cross-platform, to include handhelds
Status:
90% complete C# implementation in .NET
Windows on laptops
Compact Framework on handhelds
Mono on linux
Missing data store and automatic discovery
Can demo (but not here) laptop1-PDA-laptop2
Release date target: October 2006 for version 0.9
Personnel: Jon Olson, Kevin Webb (GT)

19. Deliverables (Part 2)
New throwbox power-management power management algorithms.
A throwbox platform prototype built from COTS components.
Data from a long-running DTN testbed.
Implementation of DTN specification in .Net
Mechanisms for synergistic networks consisting of DTN and mesh regions.
Implementation of applications that leverage the routing-related services.