Empirical Study of Android Alarm Usage for Application Scheduling

An Empirical Study of Android Alarm
Usage for Application Scheduling
Mario Almeida1
, Muhammad Bilal1
Jeremy Blackburn2
, Konstantina Papagiannaki2
1
Polytechnic University of Catalonia
2
Telefonica Research
1
(4knahs@gmail.com)

Energy: Major Concern for Mobiles
● Main Culprits?
○ Screen
○ Radio
○ CPU
○ Graphics
○ Sensors
● Interactive Use
● Background Use
(ACM ID 1855861 USENIXATC'10)
2

How Much Background Use Is There?
Online presence apps tend to run in
background 70-80% of the time!
(Andrius et al CONEXT’13)
3

How Much Background Use Is There?
Up to 9 times faster energy drain with 4 popular apps!
(Andrius et al CONEXT’13)
4

How To Schedule Background Tasks?
Android alarms to the rescue!
Time critical (exact) or not (inexact/deferrable)
Apps can determine:
● Deferrability
● Trigger time
● Repetition interval of alarms!
5

From Android Kitkat (19) alarms can be batched!
● Reduced wakeups
● Saves energy (data)
6
Major target of energy saving innovation in last 2 Google IO conferences!
Traffic shaping (e.g., batching) cannot ignore app-
protocol interaction & OS integration
*Vergara et al IWCMC’13

Android 6 (SDK23 ,Oct 15) -> 3 new APIsJobInfo APIs (21) not adopted by apps
7

8
Android
Studio
Eclipse

Android 6 (SDK23 ,Oct 15) -> 3 new APIsJobInfo APIs (21) not adopted by apps
9
API Fragmentation leads to unpredictable
QoS/performance

Questions
● How outdated are target SDKs?
● How widespread is alarm usage?
● How do apps use alarms?
● How do alarms impact a running device?
● How to improve alarm usage?
?
10

Dataset & Methodology
● Google Play Store Crawler
○ Up to 564 of the most popular free apps for each
app category
○ >22K unique apps
● Popular apps
○ + representative apps on user devices
○ + likely -> optimized for energy consumption
> 81% of total
downloads!
11
May 2015
*Viennot et al SIGMETRICS’14

Dataset & Methodology
Static analysis:
○ Alarm API usage
○ Annotations (e.g., TargetApi)
○ Package names
○ Target SDKs
12
Skype

How Outdated Are Target SDKs?
Majority (76%)
18 months.
(SDK<=19).
~Half lagging by >
21 months!
13

How Widespread Is Alarm Usage?
Most Alarms Exact Alarms
Widgets (80%) Casino games (37.9%)
Wallpapers (63%) Weather (35%)
Personalization (60%) Family games (32.5%)
Communication(59%) Communication (32.1%)
47% of apps use
alarms
Apps define up to 70
alarm!
14

Why Are Alarms So Widespread?
10% of ALL alarms
are defined in
these libraries
22% are
exact!
15

How Do Apps Use Alarms?
31.5% of all alarms are repeating!
*Note: apps can have both exact and inexact alarms!
16
SDK < 19
Alarm Type
Percentage of
Apps
Inexact Alarms 8.49%
Exact Alarm 44.05%
Any Alarm 46.06%
SDK >= 19
Alarm Type Percentage of
Apps
Inexact Alarms 52.91%
Exact Alarm 2.31%
Any Alarm 53.49%

Is It That Many Apps Are Outdated?
Even regularly
updated apps often
do not update their
SDK!
17

How Do Alarms Impact A Running
Device?
Experiment:
● Install dataset apps on freshly flashed device
● Run them once and let them on for 30m
● Reboot, let the phone go idle, wait 30m
● Record alarm definitions/wake ups for 3h
We used a device with Android 4.4.4 (SDK 19)
with alarm batching capabilities!
18

How Do Alarms Impact A Running
Device?
Dataset 1:
● 30 most popular apps
○ >1K to >500M downloads
Dataset 2:
● 30 apps with most setRepeating alarms and
SDK <19
○ deferred if SDK>= 19
19

How Frequent Are Device Wakeups?
Not deferred due to SDK target!
High correlation (r=0.86, p<0.01)
alarm registered <-> wakeups
20
Dataset 1:
● 261 alarms registered, 20% woke the
device
Dataset 2:
● 1041 alarms registered, 61% woke
device!

Examples Of Careless Alarm Usage
● Norton Security & Antivirus
○ Both SDK 17, 10M-50M downloads, together
woke the device 141 times!
● Spoora (instant messaging)
○ SDK 9, 10-50K downloads, woke the device 372
times!
● Facebook app & messenger
○ Do not share communication, both do keep-alives
and woke the device 47 times!
21

How To Improve Alarm Usage?
● Improve dev knowledge?
○ What about API fragmentation?
● Make all alarms inexact?
○ Does it break?
● Simply target a newer SDK?
○ Might work for more recent devices
● iOS?
○ Are alarms really needed?
22

Summary
● ~ 50% apps define alarms, from which ~40%
have exact alarms.
● Alarms & API fragmentation can greatly affect
energy efficiency.
● Most alarms use APIs outdated by >18months.
● Alarms:
○ Present in all categories.
○ Up to 70 alarm definitions per app.
○ At least 10% due to 3rd party libs.
○ Can wake the device very often -> wrong SDK 23

CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE
ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME CITE ME
Thank You For Your Time!
Questions?
24

News prior to study!
Doze:
● Android 6 -> SDK 23 (study is <= 22).
● Long period of inactivity -> defer alarms
● setExact() and setWindow() -> deferred
● BUT:
○ 3 new apis (setAndAllowWhileIdle,
setExactAndAllowWhileIdle and setAlarmClock)
○ Better name semantics, devs can still misuse!
25

Image rights
https://commons.wikimedia.org/wiki/File:
Android_sick.svg
https://c2.staticflickr.
com/6/5306/5642989630_bd05b2544c_b.jpg
26

Empirical Study of Android Alarm Usage for Application Scheduling

Recommended

More Related Content

Similar to Empirical Study of Android Alarm Usage for Application Scheduling (20)

More from Mário Almeida (14)

Recently uploaded (20)

Empirical Study of Android Alarm Usage for Application Scheduling