1. ExScal Breakout Session Anish Arora ExScal Team June 2004

2. 2 Objective of Extreme Scaling To create a robust 10,000 node autonomous ad hoc sensor network in an operationally relevant environment for demonstrating the capabilities in the monitoring and protection of long linear structures Analysis:  Robust 10,000 node autonomous ad hoc sensor network  Not a toy demonstration, two orders of magnitude larger than before  Operationally relevant environment  Not a laboratory demonstration, not a simulation, it will be in a real system  Demonstrating the capabilities  Okay, not a real fieldable system, but having all the capabilities for the mission  Monitoring and Protection  The network detects, tracks, and classifies intruders  Long linear structures  Pipelines, borders, …

3. 3 Candidate applications Colombian Pipeline Kansas Pipeline Northern border control Ideal environmentals 2D detection of mounts and dismounts Kansas, but canopy precludes aerial techniques Rain forest –Mountains - Water Maximum environmental challenges

4. 4 Convoy protection 1D Convoy protection 2D Candidate applications IED Hide Site Detect anomalous driving patterns on road Detect anomalous activity along roadside

5. 5 Pipeline Pump Station Security Zone 20 km 1 km Detection and tracking of personnel Detection and tracking of vehicles Detection of unknowns ? Reaction Force Canonical problem and solution

6. 6 Outline Timeline Xsm schedule Workplan Metrics Site and logistics Deployment Monitoring & Management Visualization Simulation & Testbed Beyond extreme scaling: Other proposed ExScal experiments for December ? What to measure ?

7. 7 Checkpoint Status -Status on April 30 th 2004 (actual): -XSM prototype integration -Tier-2 Stargate relay -Tier-3 display -Topology: 12 XSM, double-line, 10m apart -Intruder model: Diesel golf-cart, diesel truck -Detection: Acoustic, infra-red and magnetometer -Classification: Acoustic-based -Integrated services: routing and time-sync at Tier-1 -Terrain and environment: fallow field, minor growth, wind -Packaging: Prototype wind shields for XSMs -Standalone demonstrations of reprogramming, localization

8. 8 ExScal is behind schedule! -Status on April 30 th 2004 (planned): -100 XSM Motes, 2 Stargates, Tier 3 visualization -Packaged nodes -Topology: 180x36m -Intruder model: Dismounts 2-5kmph, vehicles 5-40kmph Steady-state: 4 intruders in network Separation: 50m Frequency: 1-4/minute Straight-line motion with variable speed -Detection: Acoustic, infra-red, magnetometer -Classification: Dismounts, vehicles -Tracking: Continuous, within 5-10m -Integrated self-localization and reprogramming at Tier 1 -And we’re not here even today !

9. 9 Extreme Scaling Project Plan (Dec 2003) Activity Primary Performer System Analysis  Define Operational Scenario  Define Operational/Technical Rqmts.  Document System Architecture  Document Performance Metrics DARPA OSU System Design  Design System Hardware  Design System Software Crossbow OSU System Development  Enhance XSM Algorithms/Software  Enhance Relay Node Software  Enhance Display Unit Software Developers System Production  XSM Production  Relay Node Production  Display Unit Production Crossbow OSU System Integration  Integrate XSM / Relay Node / Display Unit Components OSU System Testing/Demonstration  System Component Testing  Preliminary Integration Testing  Final Integration Testing  System Demonstrations Developers OSU Program Management  Extreme Scaling Workshop  IPT Meetings  Status Conference Calls DARPA Dec. ’03 Jan. ’04 Feb. ’04 Mar. ’04 Apr. ’04 May ’04 Jun. ’04 Jul. ’04 Aug. ’04 Sep. ’04 Oct. ‘04 Build 1 Delivered Build 2 Delivered Build 3 Delivered Component Demo Integrated Demo #1 Ful dress rehearsal Integrated Demo #2 Final demo December 9, 2004

10. 10 Progress thus far weighed down by XSM design Main delay: XSM design, validation, and delivery First test units : xBow schedule slipped from Feb 25 th to March 25 th  then found driver & testing issues  and design flaws:  incorrect spec for PIR signal conditioning  mag circuits (circuit error, needed hardware patch)  no packaging yet, Packaging not dealt with, so wind issues  could only get signal chains working by April 25 th Following April IPT  joint specification revision, design review, schematics review, test framework development  contract issues

11. 11 Delivery schedule XSM: July 15: Test units August 6: 200 XSM September 3 : 1500 XSM end October : rest Stargates: expected late July

12. 12 Work-plan (locations, dates, and goals) OSU 4/30/04 OSU 7/31/04 OSU 8/30/04 OSU 9/30/04 WPAFB 10/30/04 Eglin-1 11/16/04 Eglin-Final 12/14/04 XSM prototype PIR, Magnetic sensors Acoustic detection & classification Tier-1 Routing, Time-sync; Tier 2 Relay Stand-alone re- programming, localization Redesigned XSM validated XSM Enclosure All partner components integrated Small- topology tier- 2 network 100 XSM stress-test: variability and fault tolerance 1000 XSM simulation 150 Stargate Tier-2 expts. 400 XSM: Confidence in localization & reprogrammi ng (400XSM) Alternative topology test Deployment by OSU, with AFRL parti- cipation Remote deployment (1000 XSM, 1 km) 275 Stargate expts. Confidence in application Site logistics Dry run at complete scale (10, 000 XSM, 10 km x 500m) Final demo

13. 13 Current focus -Integrated baseline -3 signal chains -Routing, time-sync, localization, reprogramming, visualization -Tier 2 scale -Application refinement: -Multi-modal fusion -Probabilistic finer-grained classification, -Based on harmonic analysis, trajectory analysis -Power management -System monitoring and management -Deployment and logistics

14. 14 Outline Site and logistics Deployment Timeline Xsm schedule Workplan Metrics Monitoring & Management Visualization Simulation & Testbed Beyond extreme scaling: Other proposed ExScal experiments for December ? What to measure ?

15. 15 Metrics -Coverage: 10km x 500m -Sensors: 9500 XSM, 275 Stargates -Deployment: < 1 day -Startup: Desired 2 min - Current target 10 min -Shutdown: 1 min -Collection: < 1 day: 10-20 personnel + vehicles -Reprogramming: Desired 10 min for 128KB

16. 16 Metrics (contd.) -Reliable communication range: - Tier-1: < 30m - Tier-2: < 500m - Tier-2-Tier-3: < 5 km -Detection: P FA : < 0.005 - Range: >10m -Classification: P FA : < 0.01 for vehicle, - < 0.04 for dismount -Latency: Detection: < 2s Classification: < 10s -Accuracy: Tracking: - continuous within thick region, - bounded-box in thin region

17. 17 Metrics (contd.) -Robustness: -random failure: < 10% -byzantine failure: < 3% -network: loss of 1 Stargate per 1000 nodes -Endurance: Tier 1, Tier-2: 14 days at 10hrs/day, 10 events/day -Time synchronization: Within 1ms -Visualization: Correlated with landmarks

18. 18 Outline Site and logistics Deployment Timeline Xsm schedule Workplan Metrics Monitoring & Management Visualization Simulation & Testbed Beyond extreme scaling: Other proposed ExScal experiments for December ? What to measure ?

19. 19 Management Handle mode switching between application phases  e.g. data transfer from one mode to next (e.g. node coordinates & potential parents after localization phase) Scheduler services to allow exclusive access to mote resources for better performance  e.g. suspend application during reprogramming Managing RAM usage across different services  localization + sensing application > 4K RAM

20. 20 Management Tuning application parameters  Defined interfaces to allow components to expose tunable parameters  Piggyback parameter updates on routing control messages  Network friendly  Reliable and self stabilizing  Use tools e.g. config (UCB) to automate code generation, component wirings, etc

21. 21 Monitoring Detecting system-level deadlocks and livelocks  Watchdog timer  Periodically post a task and check if it executes Monitoring component liveness through heartbeats Grenade timer  Forced reset to allow access to protected code  Recover from incorrectly downloaded, corrupted or misbehaving programs Monitoring system and component level invariants

22. 22 Outline Site and logistics Deployment Timeline Xsm schedule Workplan Metrics Monitoring & Management Visualization Simulation & Testbed Beyond extreme scaling: Other proposed ExScal experiments for December ? What to measure ?

23. 23 Visualization Bird’s eye view of the entire test area Close-up 3D view of a selectable portion of the test area, with ability to rotate, zoom and pan across the display Real time display of network topology, target locations, tracks etc. Ability to freeze, playback, etc. Display of diagnostic/informational parameters

24. 24 2D view of the entire test area showing Tier 2 nodes Close up 3D view with Tier 1 and Tier 2 nodes Motes detecting targets Mote topology info Target with red trail Tier 2 nodes Toggle switches for viewing various features on demand Control Panel for Start/Stop, Playback, etc. Summary info about the experiment Text messages for diagnostics

25. 25 Outline Site and logistics Deployment Timeline Xsm schedule Workplan Metrics Monitoring & Management Visualization Simulation & Testbed Beyond extreme scaling: Other proposed ExScal experiments for December ? What to measure ?

26. 26 Kansei testbed Experience  progressive increase in scale :  from 10  100  10,000 nodes  concomitant increase in component depth and interaction complexity  comcomitant observations of failure of components fail to meet the demands of higher scale  progressive increase in deployment, reliability, & robustness, issues:  complex failure modes are emerging  limited node and network capacity imply nontrivial event loss  byzantine nodes  simulations have thus far been inadequate to validate solutions, experiments have been necessary and time consiuming

27. 27 Kansei testbed module

28. 28 Testbed addresses several needs Scalability  Modular model railroad construction materials Fidelity  Static and mobile sensor nodes, pursuers, evaders, environmental stimuli Usability  MoteLab with enhancements  Open to larger community Repeatability  Simply rerun previously uploaded programs, network topology, and agent actions/trajectories Out-of-Band Reporting  StarGate with 802.11 Card for out of band communications  Overhead camera and microphone for videotaping experiments

29. 29 Outline Site and logistics Deployment Timeline Xsm schedule Workplan Metrics Monitoring & Management Visualization Simulation & Testbed Beyond extreme scaling: Other proposed ExScal experiments for December ? What to measure ?

30. 30 Proposed experiments following demo MIT : co-dependent localization and tracking Mitre : ARL ? Others ? Q: Support