Download presentation
Presentation is loading. Please wait.
Published byWayne Steel Modified over 9 years ago
1
Diversity in Smartphone Usage MobiSys ‘10 June 17, 2010 UCLA, Microsoft, USC Hossein Falaki, Ratul Mahajan, Srikanth Kandula Dimitrios Lymberopoulos, Ramesh Govindan, Deborah Estrin
2
Smartphone Penetration Is on the Rise 2
3
Basic Facts about Smartphone Usage Are Unknown 3
4
Why Do We Need to Know These Facts? 4 How can we improve smartphone performance and usability? Identical users Everyone is different Can we improve resource management on smartphones through personalization?
5
Main Findings 5 1. Users are quantitatively very diverse in their usage 2. But invariants exist and can be harnessed
6
PlatformDemographics Android 16 high school students 17 knowledge workers WinMobile 16 Social Communicators 56 Life Power Users 59 Business Power Users 37 Organizer Practicals PlatformInformation Logged Android Screen state App usage Battery level Net traffic per app Call starts and ends WinMobile Screen state Applications used Data Sets 6 Platform# UsersDuration Android337-21 Weeks/user WinMobile2228-28 Weeks/user
7
Diversity in interaction Interaction model Diversity in application usage Application usage model Diversity in battery usage Energy drain model Outline 7 Comprehensive system view Interaction ApplicationEnergy
8
Users have disparate interaction levels 8 Two orders
9
Sources of Interaction Diversity 1.User demographics 2.Session count 3.Session length 4.Application use 5.Number of applications per session 9
10
User Demographics Do Not Explain Diversity 10
11
Session Lengths Contribute to Diversity 11
12
Number of Sessions Contribute to Diversity 12
13
Session Length and Count Are Uncorrelated 13
14
Close Look at Interaction Sessions 14 Most sessions are short Sessions terminated by screen timeout Few very long sessions Exponential distribution Shifted Pareto distribution
15
Modeling Interaction Sessions 15 Extremely long sessions are being modeled well
16
Implications of Interaction Diversity 16 System parameters such as timeouts can be tuned based on model parameters System can be designed with insights from the distributions Diversity Interaction Models System Design Implications
17
Diversity in application usage Application usage model Outline 17 Interaction Application Energy Diversity in interaction Interaction model
18
Users Run Disparate Number of Applications 18 50% of users run more than 40 apps
19
Application Breakdown 19
20
Close Look at Application Popularity 20 Straight line in semi-log plot appears for all users Different list for each user
21
Exponential Distribution Models App Popularity Well 21
22
Implications of Application Diversity 22 Most of a user’s attention is focused on a few applications Optimize the system for the top applications for each user Diversity Application Models System Design Implications
23
Diversity in application usage Application usage model Outline 23 Interaction Application Energy Diversity in interaction Interaction model Diversity in energy drain Predicting energy drain
24
Users Are Diverse in Energy Drain 24 Two orders
25
Close Look at Energy Drain 25 Significant variation across time High variation within each hour
26
“Trend Table” Based Framework to Model Energy Drain 26
27
Modeling Energy Drain 27
28
Conclusions Users are quantitatively diverse in their usage 28 Invariants exist and can be harnessed Building effective systems for all users is challenging Static policies cannot work well for all users Users have similar distributions with different parameters. This significantly facilitates the adaptation task
29
Diversity in Smartphone Usage MobiSys ‘10 June 17, 2010 Hossein Falaki falaki@cs.ucla.edu
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.