Computing for theComputing for theFuture of the PlanetFuture of the Planet

Andy Hopper

The Computer LaboratoryUniversity of Cambridge

A. Hopper and A. Rice, “Computing for the Future of the Planet”, Phil. Trans. R. Soc. A, Oct 2008.

http://www.cl.cam.ac.uk/research/dtg/research/wiki/CFTFP

Computing for the Future of the PlanetComputing for the Future of the Planet

1. Optimal Digital Infrastructure

2. Sense and Optimise

3. Predict and React

4. Digital Alternatives to Physical Activities

1 – Optimal Digital Infrastructure1 – Optimal Digital InfrastructureSource: Data Center Efficiency in the Scalable Enterprise, Dell Power Solutions, Feb 2007

8% of overall energy

28% 31%41%

63% 22% 15%

26% of overall energy

Energy in Data Centres

Utilisation of ServersUtilisation of ServersL.A. Barroso and U. Hö̈lzle, The Case for Energy-Proportional Computing, IEEE Computer 40, 33–37. (DOI 10.1109/MC.2007.443.)

Power and Load – Multiple ServersPower and Load – Multiple Servers

• Tasks are moved between machines• Machines not in use are switched off• Non-interactive jobs are delay tolerant

– data indexing, batch simulation, climate modeling

• XEN virtualisation approximates energy proportional computing– move on 10G Ethernet in 250msec down time, 10sec migration time

Siemens press pictureSun

• Keep moving computing tasks to where energy is available– Cheaper to transmit data rather than energy– Use energy that cannot be used for another purpose– At what granularity should jobs be shipped?– Do we ship program, data, or both?

Use of Remote Energy SourcesUse of Remote Energy Sources

7

S. Akoush, A. Rice, R. Sohan, A. Moore

Migration Performance (10G Ethernet)Migration Performance (10G Ethernet)

The Overall GoalThe Overall Goal

• Optimal Digital Infrastructure– Components switched off if not doing useful work– Energy proportional computing and communications at many levels– Use of energy that is not suitable for other purposes

• Components– Servers / Server Farms– Networks– Workstations– Terminals

• For the first time over-provisioning may not save the day!

2 - Sense and Optimise2 - Sense and Optimise

• A sensor-based digital model of the planet

• “Googling” Earth! • “Googling” Space-Time!• “The Power of Data”

• How do we do it?– coverage– fidelity– scalability– performance– usefulness

From Identification to Precise From Identification to Precise LocationLocation

Barcode PassiveRFID

UWBRTLS

Identification

Active RFID/ WiFi

Presence Detection

manual Auto ID 3m – 10m < 1m1m – 3m

Presence Detection

Location Sensing IndoorsLocation Sensing Indoors

• Ultra wideband radio location system• Measure pulse time-differences-of-arrival and angles-of-arrival

World ModelWorld Model

A 3D rendering of a “World Model”

constructed and updated in real

time using location and other systems

A real-world environment

where people are wearing location

tags

Programming with SpaceProgramming with Space

• Spatial index:– Define spaces– Monitor interactions– Generate events

• Containment

• Overlapping

– +ve and –ve events

BMW PlantBMW Plant

Installation facts: Installation facts:

1.7 km line | Sensors: 350 | Tags: 1,000 |1.7 km line | Sensors: 350 | Tags: 1,000 | Accuracy: 30 cm Accuracy: 30 cm in 3D | Events/day: 150,000 | Reliability: 99.99%in 3D | Events/day: 150,000 | Reliability: 99.99%

VBL Bus Depot LuzernVBL Bus Depot Luzern

Installation facts:Installation facts:

Area: 15,000 m² in 1 depot | Sensors: 48 | Area: 15,000 m² in 1 depot | Sensors: 48 | Accuracy: 100 cm in 2D | Reliability: 99.9%Accuracy: 100 cm in 2D | Reliability: 99.9%

Sensing Outdoors - Sensing Outdoors - Use of Vehicles, Aircraft, Mobiles, HumansUse of Vehicles, Aircraft, Mobiles, Humans

More GPSreadings

Fewer GPSreadings

Converting GPS data to Road MapsConverting GPS data to Road Maps

• GPS location data converted into a directed graph of the road network

J. Davies

Mapping the SpectrumMapping the Spectrum

• Measured 3G signal strength

• Red is poor reception

Blue is excellent

Orange circles are base stations

D. Cottingham

RFeye™ DC-6GHz Node

Converting RSS Data to Coverage MapsConverting RSS Data to Coverage Maps

• Need to convert thousands of points to ~20, per road

D. Cottingham, R. Harle

Personal Energy Meter - PEMPersonal Energy Meter - PEM

• Information about individual energy consumption– Measured, shared, embedded

• Use World Model– upload own energy use to help digital optimisation– download energy profile of devices and goods

• Lots of lovely computing problems!– measurement, indexing, caching, event-delivery, prediction, use of social

networking, security, privacy, correctness, …

S. Hay, A. Rice

Principles of PEM ApportionmentPrinciples of PEM Apportionment

• Complete• all energy accounted for

• Accurate / bounded / personalised• my actions relate to me only

• Sensible• incentives work correctly

• Trustworthy• rules are understood: for the Active Badge these were reciprocity, availability• fidelity / error bounds• security / privacy - “bad” things cannot happen

21

Power consumption varies with Power consumption varies with occupancyoccupancy

Policy can penalise some working patternsPolicy can penalise some working patterns

User A

User B

User C

Better policy provides correct Better policy provides correct incentivesincentives

OpenRoomMap: crowd-sourced maps OpenRoomMap: crowd-sourced maps of buildingsof buildingsA. Rice, O. Woodman

• Requirements– Accurate and correct model– Algorithm separated from implementation, verified code– Shared data, latest data– Deadline/parameter driven computation– Scalable computer power

3 - Predict and React3 - Predict and React

Fre

quen

cy o

f tre

e in

spec

tion

look

ing

for

dise

ase

Size of surrounding area to purgewhen an infected tree is found

Epidemiological model: Citrus CankerEpidemiological model: Citrus CankerA. Rice, C. Gilligan, Defeng Wu, M. Dingley

More is to come!More is to come!

4 - Shift to Cyberspace4 - Shift to Cyberspace

• Can we construct a digital world in which we can conduct our lives? – on a ultra-cheap open platform– using miniscule power– fed with sensor data from the real-world– accessible to every human

• Key to wealth creation in the Developing World?

Developing World: Digital PlatformDeveloping World: Digital Platform

• Capacity– Fibre links to the world

• Wireless networks– 3G/EDGE– WiMax– Femtocells

• Cheap, commodity hardware:– (Smart)phones ($5)– Netbooks ($70)

• Low-cost applications

Developing World: ThoughtsDeveloping World: Thoughts

• Local support for internet services (optimum digital infrastructure)– Majority of core internet services still hosted outside the continent– Urbanisation is advantageous– Opportunity to use renewable/haphazard energy sources

• Traffic congestion (predict and react)– Large (African) cities have underdeveloped road networks– Modelling enables efficient capacity planning

• Road conditions (sense and optimise)– Road conditions in Africa can change in a short period– Opportunity for agile re-planning

• Global trading (physical to digital)

How can-might-should-will this evolve?How can-might-should-will this evolve?

Computing for the Future of the PlanetComputing for the Future of the Planet

• Targets the world outside (and inside) computing

• A vision and an architecture

• Lists, dimensions, and quantifies computing problems

• Contemplates the unbounded upside of computing!