Computing for the Future of the Planet (at Scale) Andy Hopper

Computing for theComputing for theFuture of the PlanetFuture of the Planet

(at Scale)(at Scale)

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.

Google Tech Talk Video, 14 May 2008

Other papers at: www.cl.cam.ac.uk/research/dtg/research/wiki/CFTFP

CFP FrameworkCFP Framework

1. Optimal Digital Infrastructure

2. Sense and Optimise

3. Predict and React

4. Digital Alternatives to Physical Activities

• Energy proportional computing

• Virtual machine migration enables energy proportional computing

1 – Optimal Digital Infrastructure1 – Optimal Digital Infrastructure

• Locate data centres directly next to power source

• Use network to move jobs to data centre

• Maintain service level agreements 6

Use Renewable EnergyUse Renewable Energy

Siemens press pictureSun

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

Chase Surplus Energy Around the GlobeChase Surplus Energy Around the Globe

AVG Algorithm: Average of number of memory pages changed per unit time

(MT = Migration Time, DT = Downtime)

Migration PredictionMigration PredictionS. Akoush, R. Sohan, A. Rice

The Overall Goal (at Scale)The Overall Goal (at Scale)

• 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 Google of Things”

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

Future Street View – Heat Sensing?Future Street View – Heat Sensing?

A Better ShoeA Better Shoe

+ =

Custom wireless sensor node

Insole Augmented with Force Sensitive

Resistors

Insole Augmented with Force Sensitive

Resistors

R. Harle

60 Meter Sprint60 Meter Sprint

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

0.08

0.09

0.1

0.11

0.12

0.13

0.14

0.15

0.16

Step Number

Co

nta

ct T

ime

s (s

)Acceleration Deceleration

Speed Maintenance

Speed of RunnerSpeed of Runner

Contact Time (s)

Spe

ed (

m/s

)

Each colour represents data from a different

person

AnomaliesAnomalies

12

34

56

78

910

1112

1314

1516

1718

1920

2122

23

0.08

0.09

0.1

0.11

0.12

0.13

0.14

0.15

0.16

Step Number

Con

tact

Tim

es (

s)

Sensor Fusion and Fast Data AnalysisSensor Fusion and Fast Data Analysis

205fps video

Synchronised foot pressure data

Location in Industry - BMW PlantLocation in Industry - BMW 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 in 3D, Accuracy: 30 cm in 3D,

Events/day: 150,000, Reliability: 99.99%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%

VideoVideo

Precise Location Sensing at ScalePrecise Location Sensing at Scale

•Diffusion of hardware into conventional machines and devices

•System is cellular and scaleable at all levels

• Complete

• all energy accounted for: sensed, embedded, shared, hypothecated

• Accurate / Bounded / Personalised

• my actions relate to me only

• Sensible

• incentives work correctly

• Trustworthy

• rules are understood: reciprocity, availability

• fidelity / error bounds

• security / privacy: “bad” things cannot happen21

Global Personal Energy Meter - PEMGlobal Personal Energy Meter - PEMS. Hay, A. Rice

PEM ImplementationPEM Implementation

• Information about an individuals energy consumption• measure, interpret, postulate, allocate

• Use World Model• crowdsource data• upload own energy use to help global optimisation• download energy profile of devices, goods, physical places

• Apportion energy• to individual, group, thing, place

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

networking, security, privacy, correctness, …

S. Hay, A. Rice

Power consumption and occupancyPower consumption and occupancy

What are the rules for apportionment?

• Allocate equal share of total load independent of use

• Allocate unequal share of total load, eg to current occupants only

• Allocate equal share of base load, but my incremental load

Apportionment for transportation systems?Apportionment for transportation systems?• Walking

• my food intake?

• Car

• equivalent to office?

• Bus

• equivalent to building?

• apportion the cost of a bus service over all the passengers each day?

• All public transport ?

• What methods / policies / principles will be acceptable?

24

OpenRoomMap: crowd-sourced OpenRoomMap: crowd-sourced maps of buildingsmaps of buildings

Association of objects to places (and individuals)

Useful and rich dataset

A. Rice, O. Woodman

Personal Energy Meter: Android AppPersonal Energy Meter: Android AppD. Piggot, A. Beresford, A. Rice

3 - Predict and React3 - Predict and React

• More issues to come!

4 - Developing World: Digital Platform4 - Developing World: Digital PlatformCapacity

• Fibre links to the world

Wireless networks

• 3G/EDGE

• WiMax

• Femtocells

Cheap, commodity hardware:

• (Smart)phones ($5)

• Netbooks ($70)

Low-cost applications

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

What has been achieved since 1990?What has been achieved since 1990?