Extreme Scaling … and friends Presented by Cory Sharp UC Berkeley.

Extreme ScalingExtreme Scaling… and friends… and friends

Presented by Cory SharpPresented by Cory Sharp

UC BerkeleyUC Berkeley

OutlineOutline

• PEG: Things we did well

• Extreme scaling goals

• Extreme scaling issues

• Final demo goals

PEG: Things we did wellPEG: Things we did well

NEST PEG Midterm DemoNEST PEG Midterm Demo

• Problem: detect unfriendly vehicle enter sensitive area, track using magnetics, pursue and capture by UGV

• 10x10 array of robust wireless, self-localizing sensors over 400 m2 area

• Low cost, robust ‘mote’ device: magnetometer, microcontroller, radio network, ultrasonic transceiver

• Evader: human controlled Rover• Pursuer: autonomous rover with

mote, embedded PC, GPSdot

mag ultrasound

acoustic

pursuer

evader

PEG VideoPEG Video

PEG: Things we did wellPEG: Things we did well• Modular software, SystemC

– Actually reusable, compose system services, remote invocation

• Routing, landmark-based– Robust mobile-to-mobile routing within the nodes

• Neighborhood– Node identification, membership, and data sharing abstraction

• Remote Config, Ident– Delayed parameter binding reduces reprogramming

• Remote command-line interface– When GUI’s just don’t cut it: scripting 100 real nodes, baby

• Heirarchical system structure– Primary sensors simple, defer processing to agents

Extreme Scaling GoalsExtreme Scaling Goals

Extreme Scaling GoalsExtreme Scaling Goals

• 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, by the end of FY ’04

• NEST Extreme Scaling Analysis– To discuss the NEST plan– Robust 10,000 node autonomous ad hoc sensor network– Operationally relevant environment– Demonstrating the capabilities– Monitoring and Protection – Long linear structures (Pipelines, borders, …)– End of FY ’04

300 kmborder

IRAQ

SYRIA

AL QAIM

IRAQ / SYRIA Border October 2003

25 km

dirt road

500m

500m

Sensor node with 50m range Exfiltration node (PDA)

400 nodes/km2

Both sides

1% of Total Laydown

Schedule

3 month6 month

9 month12 month

100030006000

Full systemdemonstration

Specs

• Dismounts – armed and unarmed• Vehicles• 2 second latency• < $150/node• Pfa < 1 alarm/day• Pd = 0.95 at walking speed

““Kansas Pipeline” – Canonical Kansas Pipeline” – Canonical 10,000 Node Problem10,000 Node Problem

Extreme Scaling ResponsibilitiesExtreme Scaling Responsibilities

• Ohio State is the lead

• Berkeley’s role– Reprogramming a 10,000 node network

• Software modules• Network reprogramming boot loader

– System monitoring / watchdog– System support

Scaling IssuesScaling Issues

1. Timeline is too aggressiveJune 04 100December 04 1000June 05 10,000

2. Sensing is a problema. Magnetics unlikely to detect weapons at ranges > 1mb. PIR unlikely to be a cure-all c. Consider McEwan radars

3. Comms range seen as problematic

4. Need 4 X more PDA’s – Pfa / lifetime / latency very aggressiveand triplet may be impossible to achieve simultaneously

Extreme Scaling IssuesExtreme Scaling Issues… raised in a NEST Sep’03 telecon… raised in a NEST Sep’03 telecon

5. Need for rock-solid over air loading capability needed to demonstrate scalability – and this is not currently in hand.Architecture work needed.

6. Power is a huge problem for a demonstration- Always on needed to achieve Pfa?- Debugging and software loads are battery intensive- Lifetimes are currently far too short

7. Solid packaging needed (“one touch only”)

8. Need a separate contractor to actually camp in desertand deploy sensors

9. Next tier PDA work needed , as well as highest tier (GUI)

Extreme Scaling IssuesExtreme Scaling Issues… raised in a NEST Sep’03 telecon… raised in a NEST Sep’03 telecon

10. Dynamic tagging seen as an alternate approach that demonstrates scalability without problems in static sensor laydown

11. Rush to production seen as a costly mistake. A disciplined,Step-by-step process with formalized acceptance criteria at eachStage was recommended to avoid a failure.

Extreme Scaling IssuesExtreme Scaling Issues… raised in a NEST Sep’03 telecon… raised in a NEST Sep’03 telecon

NEST Final Demo GoalsNEST Final Demo Goals

NEST Final Demo GoalsNEST Final Demo Goals• “Bigger and better” than the midterm demo• We have the opportunity to push the demo instead

of being pulled (kicking and screaming) by it– The midterm demo is money in the bank, let’s spend it– Push to get the experiments and data we need for

research we care about

• Extracting data from and manipulating large deployments– Establish a 100-node UCB testbed– NEST extreme scaling demo– NEST final demo

Steering the Final DemoSteering the Final DemoA.K.A. “If DARPA paid for and deployed an insane number A.K.A. “If DARPA paid for and deployed an insane number of nodes, and you wanted to explore your research ideas of nodes, and you wanted to explore your research ideas

on a real, large deployment, what research would that be?”on a real, large deployment, what research would that be?”

• Identifying failing or malicious nodes• Measuring application-level metrics• Testing multi-object tracking• Exploring the small-world connectivity model• 100 small robots as agents of the network• … your ideas?

Thanks!Thanks!

gg ppt, owntgg ppt, ownt