Mobile Cloud, Crowd & Fog Computing, … Cloud, Crowd & Fog Computing, Communications and Sensing Ivan Stojmenovic . Contents 1 Mobile Cloud Computing 2 Applications 3 Crowd & Fog

[email protected] www.site.uottawa.ca/~ivan

Mobile Cloud, Crowd & Fog Computing,

Communications and

Sensing

Ivan Stojmenovic

Contents

1 Mobile Cloud Computing

2 Applications

3 Crowd & Fog sensing & computing

4 Vehicular cloud/crowd

5 Green Computing

0 Wireless and Cloud

2/24/2010 3

Mobile phones replacing desktop

computers for cloud access

Screen? Wireless? Computing? Sensing?

1926 Nikola Tesla: Teleautomation ‘When wireless is perfectly applied,

the whole Earth will be converted into a huge brain,

4

and the instruments

through which we shall be

able to do this will be

amazingly simple

compared with our

present telephone. A men

will be able to carry one in

his vest pocket.’ =

smartphone

Mobile Terminal

Intelligent Network

Cloud Computing

Mobile Cloud

Computing

Early Adopters: MCC Services

billpetro.com

handle e-mail, notepad items, contact book, photos and

documents,

automatically synchronized to iMac, iPod, iPhone and other

Apple’s terminal devices.

7

iCloud: Cloud Storage and Cloud Computing.

Mobile cloud computing management

Offloading methods

Mobile Cloud Computing Technologies

• Model the dependencies between

application modules, and optimize the

partitions

• Automatically allocate applications from

different levels to mobile devices and

cloud servers

• Provide solution for situations where the

latency is too high for distant cloud

resource to kick in

• Construct a mobile cloud computing

platform using cell phones

• enables smartphone applications with

distributed data and computation 10

Program Partition

Cloudlet CMU: Mahadev Satyanarayanan, where the latency is

too high for distant cloud resource to kick in directly.

‘data center in a box’

Ex: language translation app on the local cloudlet

Fog computing

Hyrax CMU Eugene E. Marinelli

Example: Hyrax multimedia search and sharing application,

HyraxTube, allows users to browse videos and images stored on a

network of phones and search by time, location, and quality.

Crowd Computing

Murray, Yoneki, Crowcroft, Hand, MobiHeld 2010

Combining mobile devices and social interactions to achieve large-scale distributed computation

analyze two encounter traces to place upper bound on the amount of useful computation on other devices that is possible

Task Farming

Single master process manages a queue of tasks, and distributes to ensemble of workers

Computation at node 0 is helped by nodes 1-4

Arrow: encounter

Useful computation

Wasted computation

Results need to be returned by deadline to be useful

Social Aware Task Farming

Exploit the social network formed by human interaction

master should meet a large number of other devices

Community structure: devices partitioned into groups: highly connected within, but few connections between

assign one master in each community?

accept only tasks from master in own community?

Opportunistic forwarding of results in addition to direct

Task dependencies and scheduling

Power consumption, task replication issues

COMMUNITY STRUCTURE Li,Wang,Yang,Jiang,Stojmenovic,IEEE INFOCOM 2014

Physical Proximity Community (PP Community)

Access Point Community (AP Community)

Space Crossing Community (SC Community)

Improving data forwarding (application)

17

Fog computing: cloud close to ground

Bonomi @ Cisco, 2012

e.g. cloudlet

FedCSIS 2014

invited paper

Fog computing: traffic lights

Micro-grids as fog devices

Wei, Fadllulah, Kato

Stojmenovic IEEE JSAC

2014

VANET as SDN

Liu, Ng, Lee, Son,

Stojmenovic 2014

SDN: Software Defined Networks

Emergent computing and networking paradigm

Separate control and data communication layers

Control is done at ‘centralized server’

Nodes follow communication path decided by the server

‘Centralized server’ may need distributed implementation

Applications and devices for mobile cloud computing

Cloud Robotics

Peer

Proxy

Clone models

SDN?

Biometric applications: verification and identification

e.g., find name of person

Real time forensic applications by experts at the scene

Socialize spontaneously with mobile applications (Liu, Feng, Li INFOCOM 2012)

achieve spontaneous social interaction

with other users in the same mobile application,

be they in the same living room or around the world.

Composers collaborate in real time

eSmall talker

Champion, Yang, Zhang, Dai, Xuan, Li, TPDS 2012

Helps strangers in physical proximity to find potential small talk opportunities

each device creates a Bloom filter based on the small talk topics, e.g., hobbies

this filter will be advertised through Bluetooth’ service discovery protocol (SDP)

Multiple round Bloom filter advertising

Encoded common topic candidates

Each topic hashed into k bits of a common vector

Topic is candidate if vector from neighbors covers corresponding k bits, but some bits might be covered by union of other topics, eliminated for the next round

From Cloud to Crowd Computing

Remove cloud: computing in mobile phones

Spontaneous wireless ad hoc networks

Creation: Lacuesta, Lloret, Garcia, Penalver IEEE TPDS 2012

Authentication issues: AES symmetric encryption or Diffie-Hellman public keys

Trust issues: adding 0/1 trust value to connections

Applications: content delivery, games…

Cloud for language translation

Crowd for language translation

Mobile Crowdsensing

Ganti, Ye, Lei, 2011

ECG enabled mobile phone

Bluetooth to

mobile phone

iPhone 4: Camera,audio,

GPS, Accelerometer,

Gyroscope,

Compass,Proximity,

ambient light

Intel’s sensor

air quality

People-centric sensing Campbell et all 2008

Personal sensing

socialize

Public sensing

Smart city Social sensing

Best restaurant?

Typical Functioning in Applications

Architecture of social crowd

Sharing sensing presence

Crowd-Sourced Sensing and Collaboration using Twitter

Demirbas, Bayir, Akcora, Yilmaz WoWMoM 2010

Tweet: 20 char username + 140 char post field

News, alert systems (e.g. connect city residents)

Twitter can provide an ‘open’ publish-subscribe infrastructure for sensors and smart phones, allowing for data mining

Participatory sensing by volunteering smart phones

E.g. noise level mapping (with GPS) and querying

Crowd-sourcing (distributing a query to several Twitter users)

E.g. weather radar, polling for best restaurant

Social collaboration (back-and-forth interaction): e.g,. Arrange ride sharing, support group for addicts, social events…

Crowdsourcing Maps

Masli, 2011

User contributed change shortens a route

Research Challenges

Localized analytics

Data mediation (e.g. noise elimination),

context inference (in a bus? Walking? Watching TV?)

Resource limitations

Energy, bandwidth, computation

Privacy, security, data integrity

Data perturbation (adding random noise)

Aggregate analytics

Data mining

Architecture

Unify for different applications, cooperation in sensing…

Vehicular Ad Hoc Networks = VANET

Vehicular Clouds/Crowds

VC – Vehicular Cloud

A group of vehicles whose corporate

Computing, sensing, communication and physical resources can be coordinated and dynamically allocated to authorized users

How are VCs different from the classic clouds?

Mobility: close proximity to an event is often un-planned

pooling of the resources in support of mitigating the event must

occur spontaneously

Autonomy: for the decision of each vehicle to participate in the VC

Agility: ability of VCs to tailor the amount of shared resources to the actual needs of the situation in support of which the VC was constituted

Mobile Cloud service by a vehicle with RSU to RSU service connection

A cloud in your parking lot

44 9/17/2014

• parking lot of a typical enterprise on a typical workday

• hundreds/thousands cars go unused for hours on end

• Why rent computational/storage resources elsewhere?

• you have them in your own backyard; they are yours to waste!

Data center at the shopping mall

45 9/17/2014

If drivers just attach to the internet by cable then malls can

• provide real data center computing services • by using the resources of the parked cars

• The shoppers cars get free parking + other perks in return

Dynamically rescheduled traffic lights

46 9/17/2014

• Reschedule traffic lights to help mitigate

congestion

• The municipality has the authority and

the code but does not have the hardware

• The cars have the

computational power but lack the authority and the code

A Possible AVC Architecture

Dynamic HoV lane designation (contraflow)

48 9/17/2014

• schedule HoV lanes in real time as required by traffic flow vehicular clouds to the rescue!

Planned evacuations

49 9/17/2014

• several inter-operating VC of vehicles involved in evacuation coordinated the emergency management center

• the emergency managers learn and upload real-time information about open gas stations, shelters, open medical

facilities etc

Network as a Service – Naas

Sending adds to the traveling public

People can subscribe to email, Internet access or location specific services in a pay-as-you-go fashion

Sharing Network Resources between Cars

Vehicles with Internet access

can be used as a network

cloud to reach thousands of

customers on the move

Vehicular social networking architecture

53

Transmission

Network interface

Computation

CPU

Memory

Sensing

GPS

Camera

Energy Consumption of Mobiles: User Side of Green MCC

Green Mobile Cloud Computing -Transmission

Significant energy cost on mobile device WiFi radio

Cellular network

Challenges Unstable wireless quality

• Various energy consumption status

Heterogeneous interfaces • Various transmission modes (PSM/CAM of wifi)

Different traffic demands • Real-time/delay-tolerant applications

Solutions Sleep during idle time by using PSM mode

Predict signal strength & traffic pattern to avoid rush hour

Send in a burst by traffic shaping

54

Green Mobile Cloud Computing - Computation

Challenges

Limited resources

• computational capability

• memory

Rely on a finite energy source

Solutions

Task out-sourcing schedule & cloud-assisted

CPU optimization

55

Task Outsourcing to The Cloud

Which can be offloaded?

High computation cost & low transfer cost

How to profile applications based on energy?

Energy state prediction

Power modeling

56

Energy Consumption Pattern on Modern Smart Phone

Tail power states NICs, sdcard and

GPS Stay at high power

state after I/O activities

Non-utilization system calls slowly change power state

Several components do not have quantitative utilization

57

Non-Utilization based Power Modeling

Tracing system calls of the applications

Accurate fine-grained energy estimation

Per-subroutine & per-thread & per process

58

Green Mobile Cloud Computing -Sensing

Challenges

High energy consumption of specific sensors

• GPS used for location-based service

59

Energy Saving of Location-based Service

Shortcomings of existing smart phones

Static use of location sensing mechanisms

Absence of use of power-efficient sensors

Lack of cooperation among multiple LBAs

60

Energy Saving of Location-based Service

Solutions

Substitution

• To make use of alternative location-sensing mechanism (e.g.,

cell-based location tracking, interpolation according to history..)

Suppression

• Use less power-intensive sensors to suppress unnecessary

GPS sensing (accelerometer, wireless data)

Piggybacking

• Synchronizes the location sensing requests from multiple

running LBAs (location based applications) e.g.,

– New LBA may delay GPS registration until existing LBA does it

61

Conclusion

We expect mobile cloud computing to see a phenomenal adoption rate and penetration of the IT market

Cloud computing will be extended to

Vehicular assets from individual vehicles to

entire fleets

Cell phones and other commodity consumer

products

Questions?