T ruth in Advertising: The Hidden Cost of Mobile Ads for Software - PowerPoint PPT Presentation

T ruth in Advertising: The Hidden Cost of Mobile Ads for Software Developers Jiaping Gui (jgui@usc.edu), , Stuart t Mcil ilroy, , Meiy iyappa pan Nagapp ppan an, , Will lliam iam G.J. Half lfond ond This work was supported by NSF Grant No. CCF-1321141 to the University of Southern California

Motivation P Prevalence Revenue R Win-win W “Free” ? H Hidden Costs 2

H idden Costs of Mobile Ads Performance cost processing power Runtime responsiveness 3

H idden Costs of Mobile Ads Energy cost More recharging cycles Users may be aware of battery limitations, but they do not have any way to isolate and evaluate the energy cost of ads within apps 4

H idden Costs of Mobile Ads Network cost 5

H idden Costs of Mobile Ads Maintenance effort Maintenance effort Update on main functions of the application Update on ad functions and other functions 6

H idden Costs of Mobile Ads Impact on the ratings Good app ratings and reviews are essential for the success of an app in the marketplace. “Annoying full screen ads every time you run the app. Uninstall” 7

Research Questions to Quantify Hidden Costs RQ1: Performance cost RQ2: Energy cost Users RQ3: Network cost RQ4: Maintenance effort Developers RQ5: Ratings impact 8

Evaluation Overview - Subject Apps App Selector Ad Removal Workload Generator Distimo 10,750 Jan 2014 Aug 2014 apps ranking # downloads Top 21 subject apps user user rating reviews 9

Evaluation Overview – Subject Apps App Name Category # Versions # Reviews Avg. Rating SofaScore LiveScore sports 39 1158 4.72082 7Zipper communication 19 699 4.66026 Guess The Song trivia 40 3426 4.62825 Podcast Republic news & magazines 39 1723 4.58928 Smart Booster - Free Cleaner tools 22 1258 4.50653 The Best Life Quotes entertainment 31 784 4.42554 3D Sense Clock & Weather travel & local 20 399 4.42509 Polaris Navigation GPS travel & local 29 960 4.41557 Radaee PDF Reader productivity 21 587 4.41044 Player dreams music & audio 40 693 4.40827 Translator Speak & Translate travel & local 22 1024 4.38482 Pixer social 27 599 4.36689 Restaurant Finder travel & local 24 464 4.35366 Lomo Camera photography 20 942 4.33325 Arcus Weather weather 30 513 4.32317 Smileys for Chat (memes,emoji) communication 16 613 4.32011 VLC Direct Streaming Pro Free media & video 25 868 4.28025 Drudge Report news & magazines 20 1317 4.24225 Public Radio & Podcast news & magazines 21 671 4.2379 English for kids learning free education 21 90 4.13483 Followers For Instagram social 17 1337 3.75924 10

Evaluation Overview – Ad Removal App Selector Ad Removal Workload Generator bytecode rewriting With ad No ad 11

Evaluation Overview - Workload App Selector Ad Removal Workload Generator With ad No ad Criteria Complete Repeatable Long enough 12

RQ1: Performance Cost Approach top tool: 1s interval CPU utilization (%) Memory (K) Resident Set Size (RSS) 13

RQ1: Performance Cost Increase Evaluation with ads 80 Mean 56% Average Performance Cost of Ads (%) CPU: 6%~84% 60 Median 48% 40 20 Mean & Memory: Median 3%~27% 0 22% M1 M3 M5 M7 M9 M11 M13 M15 M17 M19 M21 Apps 14

RQ1: Performance Cost Increase Evaluation with ads Response time: 7% longer for event handler with ads With ads workload No ads workload 15

RQ2: Energy Cost Approach Monsoon Power Monitor Sampling 5KHz 16

RQ2: Energy Cost Evaluation Six apps had an over 20% increase in energy consumption Increase with ads Max 35 battery 33% 30 2.5 hrs baseline Average Energy Cost of Ads (%) 25 Mean 20 16% 15 2.1 hrs average Median 10 15% 5 Min 3% 0 worst 1.7 hrs M1 M3 M5 M7 M9 M11 M13 M15 M17 M19 M21 Apps 17

RQ3: Network Cost Approach Metrics: • Total bytes • Number of packets HTTP, TCP 18

RQ3: Network Cost Increase Evaluation with ads 120 Median 97% 100 Bytes: 4%~100% Average Network Cost (%) 80 0.24MB 60 40 3 mins $0.017  2 hrs/day  $20.4/mon 20 # Packets: Median 0 5%~100% 90% M1 M3 M5 M7 M9 M11 M13 M15 M17 M19 M21 16/21 apps, over 80% network traffic due to ads 19

RQ4: Maintenance Effort Approach Release i Release i+1 Different multisets of ad related calls VS App versions 20

RQ4: Maintenance Effort Evaluation 50 40 Percentage of Releases (%) Median Mean 22% 23% 30 20 5 releases Jan 2014 Aug 2014 GMA 10 GMA: Google Mobile Ad Networks 0 M1 M3 M5 M7 M9 M11 M13 M15 M17 M19 M21 Apps 10 apps: higher number of ad related changes 21

RQ5: Impact of Ads on Ratings Approach Annoying full screen ads every time you run the app. Uninstall … Energy 20,125 Network Reviews 2,964 reviews Performance Ad complaints 22

RQ5: Impact of Ads on Ratings Evaluation 20 5.04% Over 50% apps 3.28% 15 Percentage of Complaints (%) 10 0.003 stars 5 0 M1 M5 M9 M13 M17 M21 Apps It is sufficient to change the ranking!!! 23

Evaluation Overview – Subject Apps App Name Category # Versions # Reviews Avg. Rating SofaScore LiveScore sports 39 1158 4.72082 7Zipper communication 19 699 4.66026 Guess The Song trivia 40 3426 4.62825 Podcast Republic news & magazines 39 1723 4.58928 Smart Booster - Free Cleaner tools 22 1258 4.50653 The Best Life Quotes entertainment 31 784 4.42554 3D Sense Clock & Weather travel & local 20 399 4.42509 Polaris Navigation GPS travel & local 29 960 4.41557 Radaee PDF Reader productivity 21 587 4.41044 Player dreams music & audio 40 693 4.40827 Translator Speak & Translate travel & local 22 1024 4.38482 Pixer social 27 599 4.36689 Restaurant Finder travel & local 24 464 4.35366 Lomo Camera photography 20 942 4.33325 Arcus Weather weather 30 513 4.32317 Smileys for Chat (memes,emoji) communication 16 613 4.32011 VLC Direct Streaming Pro Free media & video 25 868 4.28025 Drudge Report news & magazines 20 1317 4.24225 Public Radio & Podcast news & magazines 21 671 4.2379 English for kids learning free education 21 90 4.13483 Followers For Instagram social 17 1337 3.75924 24

Summary  Median relative costs of ads CPU: 56% Memory: 22% Energy: 15% Bandwidth: 97% Developers need to expend more maintenance efforts on ads! Ad complaints affect app ratings on the app store! jgui@usc.edu jgui@usc.edu 25

Thank you! Trut uth h in in Adv Adver ertisi tising ng The he Hid Hidde den Cost n Cost of of Mobile Mobile Ads Ads for or Softw Softwar are e De Develop eloper ers Jiaping Gui (jgui@usc.edu), , Stuar Stuart t Mcil Mcilroy, , Meiy Meiyapp ppan an Na Naga gapp ppan an, , Wil illia liam m G.J .J. . Half Halfon ond 26

Generality Evaluation Generalizing RQ 1-5 for four apps with two other mobile ad networks GMA: Google Mobile Ad Networks AMA: Amazon Mobile Ad Networks MMA: Mopub Mobile Ad Networks 27