MONTHLY DATA REPORT: GOOGLE IO May 2016
Introduction In celebration of Google IO 2016, this month’s data report focuses on the Android ecosystem. We’ve analyzed Android fragmentation (or lack thereof!), Google Fi device usage, crashes that are isolated to specific devices and operating systems, and Android TV vs Apple TV usage. We finish by telling you which device manufacturers are the quickest to push out Android updates to their users.
ANDROID IS NOT FRAGMENTED
One of the major criticisms of Android is the fragmentation of its operating system versions. Releasing an Android update involves lengthy approval processes and coordination between Google, carriers, and OEMs. Google in particular has been investing a lot of time and effort to reducing this approval time. They’ve also taken steps to work around the system by pushing updates through Google Play Services. Google releases a monthly breakdown of Android operating system distribution, and the picture they paint definitely appears to be one of fragmentation: © Apteligent 2016
Android Install Distribution (from Google) Version
Codename
2.2
Froyo
0.1
2.3.3 - 2.3.7
Gingerbread
2.2
4.0.3 - 4.0.4
Ice Cream Sandwich
2.0
41.x
Jelly Bean
7.2
4.2.x
Jelly Bean
10.0
4.3
Jelly Bean
2.9
4.4
KitKat
32.5
5.0
Lollipop
16.2
5.1
Lollipop
19.4
6.0
Marshmallow
Total
--
Distribution
7.5 100.0
Android Install Distribution by Codename (from Google)
35.6% Lollipop
7.5% Marshmallow
32.5% KitKat
2.2% Gingerbread
20.1% Jelly Bean
2.0% Ice Cream Sandwich 0.1% Froyo
Data collected during a 7-day period ending on May 2, 2016. Any versions with less than 0.1% distribution are not shown.
However, there are a few problems with this presentation of the data. First, it only takes into account phones running Google Play, so a significant set of data is missing from users in countries like China. Second, and most importantly, it does not take into account usage. Let’s see how the picture changes when we look at Android OS distribution by usage of the operating systems for that last full week of May 2016: Android Usage Distribution (5/22 - 05/28) Version
Codename
2.2x
Froyo
Distribution
Android Usage Distribution by Codename (5/22 - 05/28)
0.00
2.3.3 - 2.3.7 Gingerbread
0.16
3.1.0 - 3.2.x Honeycomb
0.01
4.0.3 - 4.0.4 Ice Cream Sandwich
0.48
4.1.x
Jelly Bean
2.89
4.2.x
Jelly Bean
3.91
4.3.x
Jelly Bean
0.01
4.4.x
KitKat
24.91
5.0.x
Lollipop
19.07
48.2% Lollipop
0.5% Ice Cream Sandwich
5.1.x
Lollipop
29.14
24.9% KitKat
0.2% Gingerbread
6.0.x
Marshmallow
19.42
19.4% Marshmallow
0.0% Honeycomb
Total
--
100.0
6.8% Jelly Bean
0.0% Froyo © Apteligent 2016
THERE ARE ONLY 3 ANDROID OPERATING SYSTEMS THAT MATTER: KITKAT, LOLLIPOP, AND MARSHMALLOW App owners typically focus on the operating systems that are most widely used by their customer base. For these decision makers, the official Google dataset and resulting analysis (which represents a snapshot of market share) falls short of being actionable. Apteligent’s actual usage data shows that 93% of all Android traffic comes from just three operating systems: KitKat, Lollipop, and Marshmallow. Contrast this to iOS where there are only two operating systems that matter: iOS 8 and iOS 9. These two OS versions comprise 97% of iOS traffic: iOS Distribution as of May 31, 2016
8%
3%
89% iOS 9 iOS 8 iOS 7
© Apteligent 2016
PROJECT FI DEVICE BREAKDOWN In April 2015, Google launched its mobile virtual network operator (MNVO) service, called Project Fi. Fi provides a very simple pricing plan with global data access, powered behind the scenes by Sprint and T-Mobile. Through its Google Voice capabilities, Project Fi phones automatically switch between WiFi and Cellular data for the strongest call signal. Fi was at first invite-only. This past March, Google opened up the service to anyone in the US. We took a look at the most popular devices running Project Fi. Google Project Fi Device Distribution May 2016
26% 38%
36%
Nexus 5x Nexus 6P Nexus 6
You can see a majority of the device usage is split between the Nexus 6P and the Nexus 5x. The lagging popularity of the Nexus 6 on the Fi network isn’t surprising. The device is a year behind the others, and both the Nexus device line and Project Fi are targeted at the early adopter crowd.
© Apteligent 2016
Is More Data Generated by Apple Devices on Project Fi than from Android devices? It’s a little known fact that an additional Google Fi data-only SIM can be acquired that works on certain Apple device, including the iPad Air 2, Mini 4, and iPad Pro. Those three devices are about seven times more popular than all of the eligible Android devices combined. So, we’ll go out on a limb and state that it is very likely that more data is generated from Apple devices on Project Fi than on Android devices! For more (nonspeculative) information, Google publishes an eligible devices list. 400 Million Android Crashes Analyzed for Device and Operating System Bugs We analyzed 400 million crashes and looked for specific devices and operating system bugs. These crashes are isolated just to a specific configuration, and can be very tricky for developers to mitigate. Let’s first take a look at crashes specific to devices, and then we’ll dive into issues specific to certain operating systems. Top Android Device-Specific Crashes, May 2016 100% 64.43% 28.06% 26.09% 15.02% 14.23% 12.65% 11.07% 11.07% 10.67% 9.49% 9.49% 9.09% 7.51% 7.11% 6.32% 5.93% 5.93% 5.53% 5.53%
Samsung Galaxy Tab 3 10.1 Samsung Galaxy S2 Huawei p8 Lite Samsung Galaxy A3 Samsung Galaxy Tab 2 7.0 Samsung Galaxy Tab 3 8.0 One Samsung Galaxy J1 Samsung Galaxy Tab 4 Nook 7.0 Samsung Galaxy Tab 2 10.1 Samsung Galaxy Grand Duos Samsung Galaxy Grand 2 Samsung Galaxy Tab 4 10.1 G Pro2 Samsung Galaxy s3 Mini IM-A910S Samsung Galaxy S5 Mini Samsung Galaxy S6 Edge Samsung Galaxy S7 Edge Samsung Galaxy Grand Neo
© Apteligent 2016
The top 3 devices with the most device-specific bugs are the Samsung Galaxy Tab 3 10.1, the Samsung Galaxy S2, and the Huawei P8 Lite. The overall numbers are still small relative to the total crash population. We normalized our dataset to account for device usage by dividing the device-specific crashes by the total number of crashes for a given device. In our analysis, the Samsung Galaxy Tab 3 10.1 device-specific crashes represents 0.025% of all crashes we saw for the Galaxy Tab 3 10.1. This is the highest percentage we saw in the sample. Here’s the same view by operating system: Top Android Operating Systems with OS-Specific Crashes, May 2016 100% 71.43% 59.74% 37.66% 14.2% 7.79%
4.1 4.4 4.2 4.3 6.0 5.1 5.0
The operating system view makes sense -- as the operating systems age, they face bugs unaddressed by new patches. Android 5.0 is the most mature and most recently patched OS, and therefore has the least amount of OS-specific bugs. Android 6.0, while fairly stable, is still actively being updated and patched by Google.
© Apteligent 2016
The following table shows common operating system specific bugs: OS
Exception Name
Exception Reason
4.1
java.lang.ClassCastException
android.text.SpannableString cannot be cast to android. text.Editable
4.4
java.lang.NoSuchMethodError
com.facebook.AuthorizationClient$Result. createTokenResult
4.4
java.lang.IllegalStateException
This API not supported on Android 4.3 and earlier
4.4
java.lang.NoClassDefFoundError
android/webkit/WebViewFactory$Preloader
5
java.lang.IncompatibleClassChangeError
Couldn’t find com.google.gson.annotations. SerializedName.value
5
java.lang.NoSuchMethodException
java.lang.NoSuchMethodException: super$close$void []
5
java.lang.NoSuchMethodException
java.lang.NoSuchMethodException: super$setOption$void [int, class java.lang
5
java.lang.NoSuchMethodException
java.lang.NoSuchMethodException: super$bind$void [class java.net.InetAddress
5
java.lang.NoSuchMethodException
java.lang.NoSuchMethodException: super$connect$void [class java.net.Socket
5
java.lang.NoSuchMethodException
java.lang.NoSuchMethodException: super$getInputStream$java_io_InputStream
5
java.lang.NoSuchFieldException
Not a valid proxy instance
5
java.lang.IncompatibleClassChangeError
Couldn’t find com.fasterxml.jackson.annotation. JsonProperty.value
5
java.lang.NoSuchMethodException
java.lang.NoSuchMethodException: super$getOutputStream$java_io_OutputStream
6
android.os.DeadObjectException
Failure from system
6
Java.Lang.RuntimeException
Failure from system
6
android.os.TransactionTooLargeException
Failure from system
6
java.lang.NullPointerException
Failure from system
6
android.content. pm.PackageManager$NameNotFoundException
android.webkit. WebViewFactory$MissingWebViewPackageException: android.content
6
android.os.DeadObjectException
android.os.DeadObjectException: Transaction failed on small parcel; remote
6
java.lang.SecurityException
my location requires permission ACCESS_FINE_ LOCATION or ACCESS_COARSE_LOCATION
© Apteligent 2016
For example, there is a bug impacting WebViews on Android 5 (Lollipop), a java.lang. NoClassDefFoundError on the android/webkit/WebViewFactory$Preloader. This is caused by the WebView package updating via Google Play as someone tries to use an app that leverages a WebView. The last entry in the table, java.lang.SecurityException, is related to the new runtime permission system in Android 6 Marshmallow, causing issues for developers. We’ll dive more into this crash data in the coming months with best practices to mitigate these types of issues, and provide further analysis of operating system and device-specific bugs.
APPLE TV IS USED TWICE AS MUCH AS ANDROID TV Both Google and Apple have rebooted their TV platforms in recent years. At Google IO 2014, Google rebooted its lagging Google TV platform; they renamed the platform Android TV and launched it with a modified version of Android 5.0 Lollipop. At Apple’s September 2015 event, they launched a new Apple TV with the new tvOS. Since then, we’ve tracked the adoption and performance of both. TV OS’s by Apploads
Android
Apple TV
60 40 20 0
May 3
May 8
May 13
May 18
May 23
May 28 © Apteligent 2016
We’ve found that on average, Apple TV is used over twice as much as Android TV. However, Android TV is used more consistently throughout the weekdays and weekends, while Apple TV usage increases about 30% on weekends. This is mostly likely a result of Google’s effort to embed the OS directly into smart TVs, which the company believes to be an approach that provides a more consistent, ongoing experience for its users. Toward the end of May, Google quietly removed the Nexus Player from the Google Store. We take this to be further evidence of Google’s desire to prioritize TV OEMs as a competitive response to Apple’s focus on a standalone set-top box. As Google continues to work with TV manufacturers, we expect the gap to further close against Apple TV.
ANDROID TV APPS TWICE AS STABLE AS REGULAR ANDROID APPS One other interesting tidbit on Android TV performance. We found that apps running on Android TV are about twice as stable as apps running on Android phones/tablets. The average crash rate for May on Android TV was 1.5%, while the average for Lollipop and Marshmallow was close to 3%. We believe that a primary driver of this difference is the stability of a network connection for the smart TV. When you connect to a cloud service for content directly from your TV, you’re not taking your TV through a tunnel or dealing with connectivity changes between WiFi and cellular. Apteligent data shows that network issues often lead to crashes and errors, and network issues tend to be mitigated by the direct ethernet or home WiFi connections that are relatively more stable.
© Apteligent 2016
WHICH MANUFACTURER RELEASES THE FASTEST ANDROID UPDATES? Earlier, we showed that Android operating system fragmentation has largely been exaggerated. However, there is definitely still room for improvement. The biggest barrier to Android updates being rolled out quickly? Device manufacturers themselves. It’s been recently reported that Google has an internal “shame list” of manufacturers that fail to roll out updates in a timely manner. We have a significant amount of data on this subject that we’ll report shortly. Here is a preview of where eight large manufacturers rank:
The graph above displays how quickly various OEMs responded to the Android 5.0 update. The first source code was released November 3, 2014. It took most manufacturers seven months to release the update. Asus attempted the update first and halted the roll-out due to issues. Motorola ran into bugs as well in their initial release, and on July 20, 2015, it halted the update due to bugs. Samsung, LG, Sony, and HTC successfully pushed out the updates in that seven month timeframe, while Motorola, ZTE, and Amazon were the three slowest taking almost 1.5 years to get around to the update(!) The Fastest Manufacturers: Samsung, LG, Sony, and HTC The Slowest Manufactures: Amazon, Motorola, and ZTE Consolation Award for Trying 1st: Asus © Apteligent 2016
We plan to do a future report on device manufacturers. For immediate information, contact us at
[email protected], or check back in the future on data.apteligent.com/ research for the report. A note on our data Our data is benchmarked across tens of thousands of mobile apps representing hundreds of millions of application launches. Adoption rate is based on app loads, which means it is based on actual usage of the operating system. You may see slightly different numbers reported in the future by Apple, which are solely based on activations vs actual device usage. You can find mobile industry benchmarks updated daily at data.apteligent.com. In addition, every month we publish a recap of the trends and movements in the industry.
ABOUT APTELIGENT If you develop an app, or are responsible for the success of a mobile app, Apteligent’s lightweight SDK helps you optimize user experience by identifying performance issues, such as crashes and network failures, that impact user behavior. The company’s solution provides a real-time global view of apps across iOS, Android, Windows Phone 10, Hybrid and HTML5. Trusted by three of the top five credit card issuers, three of the top five media companies, three of the top five retailers, and two of the top three hotel chains with the success of their strategic mobile app initiatives. Apteligent is leading the drive to the App Economy.
www.apteligent.com | @apteligent |
[email protected] © Apteligent 2016