effSense: Energy-Efficient and Cost-Effective Data Uploading in Mobile Crowdsensing

Leye Wang, Daqing Zhang and Haoyi Xiong TELECOM SudParis, France

PUCAA 2013 Institut Mines-Télécom

Outline ■ Introduction ● ● ● ● ● ●

■ ■ ■ ■

2

Research Motivation User Needs and Objectives Some Observations Proposed Ideas Research Issues Main Contributions

The Proposed effSense Framework Uploading Schemes in effSense System Evaluation Conclusions and Future Work

Institut Mines-Télécom

Research Motivation ■ Mobile Crowdsensing can enable many applications, to encourage users to participate in Mobile Crowdsensing tasks, reducing cost incurred helps: ● Energy Consumption ● 3G Mobile Data Communication Cost

■ 3 stages of a mobile crowdsensing task ● Task Assignment ● Sensing ● Data Uploading (our focus)

3

Institut Mines-Télécom

User Needs and Objectives ■ User Types and Needs ● Data-plan users (DP) – with unlimited data plan − concerned about energy consumption

● Non-data-plan users (NDP) – pay according to data size uploaded via 3G − concerned about data cost

● Example: MIT Reality Mining − ~30 users with data plan (data-plan users) − ~70 users with no data plan (non-data-plan users) − The data-plan/non-data-plan user ratio might change

■ Objectives ● Data-plan users (DP) – Reduce energy consumption ● Non-data-plan users (NDP) – Reduce data cost

4

Institut Mines-Télécom

Some Observations ■ Some Known techniques to Reduce Energy Consumption (for DP) ● Baseline: Initiate a 3G connection to upload data ● Uploading data via a Bluetooth gateway or WiFi consumes less energy ● Uploading data via 3G parallel with a phone call saves >75% energy − J. Nurminen, “Parallel connections and their effect on the battery consumption of a mobile phone,”

■ Some Known techniques to Eliminate Data Cost (for NDP) ● Uploading data via a Bluetooth gateway or WiFi ● Offloading data by Bluetooth to a DP user, who can later relay data to the server

5

Institut Mines-Télécom

Proposed Ideas for Energy/Data Cost Cut ■ Leveraging Critical Events ● ● ● ●

making a phone call encountering a Bluetooth gateway connecting to WiFi meeting another user

■ Use Time to Trade-off Events - Delay-tolerant data uploading ● Traditionally, data is uploaded as soon as it is sensed. ● We allow delay between data sensing and uploading, so that critical events can happen and be leveraged to reduce energy and/or data cost.

6

Institut Mines-Télécom

A Running Example (How to Reduce Cost)

S

Server

ri: sensing data of ui r’i: sensing data of u’i WiFi/LAN

3 uploading paths: 1. u’3  G  S 2. u’1  u2  G  S 3. u’2  u1  S

3G (parallel with voice call)

{r’2,r1}

Bluetooth Gateway

G

{r’3}

{r’1,r2}

{r’1}

data-plan users 7

u’1

u2

u1

{r’2}

Institut Mines-Télécom

u’2

u’3

non-data-plan users

Research Issues 1. Identify the critical events and predict the probability of future occurrence of these critical events.

2. Estimate data uploading energy consumption associated to each critical event.

3. Design a real-time algorithm distributed on each user’s phone to decide whether to upload/offload data or keep data when a sequence of critical events occur.

8

Institut Mines-Télécom

Research Contributions ■ Intend to achieve two objectives for two types of users for 1st Time ● energy consumption (data-plan users) ● data cost (non-data-plan users)

■ Propose a novel mechanism to address the problem ● delay-tolerant data uploading ● critical events

■ Develop 4 different data uploading schemes ● Two for data-plan users ● Two for non-data-plan users

■ Evaluate the proposed mechanism and schemes with two datasets ● MIT Reality Mining ● Nodobo

9

Institut Mines-Télécom

Outline ■ Introduction ● ● ● ● ● ●

■ ■ ■ ■

10

Research Motivation User Needs and Objectives Some Observations Proposed Ideas Research Issues Main Contributions

The Proposed effSense Framework Uploading Schemes in effSense System Evaluation Conclusions and Future Work

Institut Mines-Télécom

The Proposed effSense Framework Mobile Crowdsensing Framework Sensing Decision Making (offload or keep data) Data Uploading

11

Institut Mines-Télécom

The Proposed effSense Framework ■ Key Concepts in effSense Framework ● Two types of crowdsensing users – DP and NDP ● Critical Events Prediction for each user − mobility prediction • meet other user/ Bluetooth gateway/ WiFi AP

− call prediction

● Data Uploading Schemes − Cold-Start – works without user logs − Prediction-Based - leverage user logs

● Force uploading at deadline − At the end of the data uploading period (max delay), effSense checks data-plan users whether they have non-uploaded data: for those with data, effSense forces them to establish a 3G connection to upload data to the server.

12

Institut Mines-Télécom

Outline ■ Introduction ● ● ● ● ● ●

■ ■ ■ ■

13

Research Motivation User Needs and Objectives Some Observations Proposed Ideas Research Issues Main Contributions

The Proposed effSense Framework Uploading Schemes in effSense System Evaluation Conclusions and Future Work

Institut Mines-Télécom

Uploading Schemes in effSense ■ When a critical event occurs, decide whether to offload/keep data ■ For NDP ● SimpleGreedy (cold-start) ● AdvancedGreedy (prediction-based)

■ For DP ● Greedy (cold-start) ● ExpectationBased (prediction-based)

■ Assumption - Offload and Dismiss: Once a user offloads data to another user, won’t be responsible for sending the data any more. ● Avoid redundant data offloading to reduce energy consumption

14

Institut Mines-Télécom

cold-start schemes for NDP/DP users

SimpleGreedyndp/Greedydp ■ Cold-start schemes make the decision directly according to the event type. ● without using user history logs NDP user: SimpleGreedyndp upload data without incurring NDP users’ data cost

Critical Event

15

Call Bluetooth Gateway Connect WiFi

DP user: Greedydp upload data with less energy consumption than 3G

Critical Event

Call Bluetooth Gateway Connect WiFi

meet DP user

meet DP user

meet NDP user

meet NDP user

Institut Mines-Télécom

prediction-based scheme for NDP users

AdvancedGreedyndp ■ Difference from SimpleGreedyndp ● a NDP user might offload data to another NDP user encountered Call Bluetooth Gateway

Critical Event

16

Institut Mines-Télécom

Connect WiFi

meet DP user

the encountered user has higher probability to upload data without data cost

meet NDP user

Otherwise

prediction-based scheme for DP users

ExpectationBaseddp ■ Calculate DP user’s expected energy consumption under different conditions when encountering a critical event ● Call / Bluetooth Gateway / WiFi − Upload data − Keep Data

● Meet another DP user (suppose two users: ui and uj) − ui offloads data to uj − uj offloads data to ui − ui and uj both keep own data

■

● Decide according to the condition with smallest value Calculation is based on ● predict the future occurrence of critical events ● estimate the data transmission energy of critical events

17

Institut Mines-Télécom

Outline ■ Introduction ● ● ● ● ● ●

■ ■ ■ ■

18

Research Motivation User Needs and Objectives Some Observations Proposed Ideas Research Issues Main Contributions

The Proposed effSense Framework Uploading Schemes in effSense System Evaluation Conclusions and Future Work

Institut Mines-Télécom

Experiment Setup ■ Two datasets ● MIT Reality Mining, Nodobo

■ Max delay: 24 hours

MIT

Nodobo

#DP user

30

11

#NDP user

41

16

#BT gateway

2

0

?WiFi traces

no

yes

● Start of the data uploading: 00:00:00 for each day

■ Estimation of the energy consumption under critical events ●

N. Balasubramanian, etc. , “Energy consumption in mobile phones: a measurement study and implications for network applications,” 2009

■ Prediction Method ● count week-level frequency ● e.g. 3 weeks out of 5-week logs, a user makes calls during Monday 8:00~12:00 − predict that this user’s call probability is 60% during next Monday 8:00~12:00

19

Institut Mines-Télécom

Questions ■ NDP user (Data Cost) ● How many NDP users could successfully upload their sensed data to the server without data cost?

■ DP user (Energy Consumption) ● How much energy do DP users save compared with traditional 3G data uploading?

20

Institut Mines-Télécom

NDP user (Data Cost) ■ NDP users uploading data without data cost ● more than 45% (MIT) ; 54% (Nodobo)

MIT (41 NDP users)

21

Nodobo (16 NDP users)

Working Days

Holidays

All Days

SimpleGreedy

23.3

10.6

18.8 (45%)

AdvancedGreedy

24.3

11.0

19.6 (48%)

Institut Mines-Télécom

Working Days

Holidays

All Days

SimpleGreedy

10.00

6.20

8.64 (54%)

AdvancedGreedy

10.33

6.40

8.93 (56%)

DP user (Energy Consumption) ■ Energy consumption of DP users ● 4 scheme pairs: 2 (NDP)*2 (DP) ● traditional: establish a new 3G connection to upload data

• Compared with traditional 3G • Greedydp reduces ~55% • ExpectationBaseddp reduces ~65%.

22

Institut Mines-Télécom

Conclusions and Future Work ■ effSense is a novel data uploading framework for mobile crowdsensing ● ● ● ●

reduce both energy consumption and data cost introduce delay-tolerant data uploading and critical events design cold-start/prediction-based uploading schemes for DP/NDP users evaluation − more than 45% NDP users uploading data without data cost − DP users reducing more than 55% energy compared with traditional 3G uploading

■ Future Work ● ● ● ● ●

23

Other critical events Precise energy consumption estimation mechanism Advanced mobility/call prediction method Incentives for DP users to relay data Real-life deployment

Institut Mines-Télécom

Institut Mines-Télécom