PERFMATTERS for Android - DroidCon Turin 2016

The slogan '#PERFMATTERS' by Colt McAnlis is one of the most important statements, which an Android developer should follow. A well performing application makes the difference between a nice looking application and one, which gets successful. You, as an Android developer, have the responsibility to delight your users, so they don't feel how much work your application has to burden to fulfill the user's needs. This session will show you some easy optimizations and how to avoid glitches in the application. A lot of developers would shout, that this is premature optimization. In contrary it is preventive. With a small amount of work while developing you can avoid a lot of pitfalls. You will hopefully have a better understanding of the underlying system afterwards. During the session some tools and frameworks will be presented for helping you to fulfill your mission.

Transcript

1. www.immobilienscout24.de
   Hasan Hosgel – Droidcon Turin 2016
   #PERFMATTERS
   for Android
   

   View Slide

2. #PERFMATTERS for Android | Hasan Hosgel
   Immobilien Scout GmbH
   Location: Berlin
   Employees: ~ 520
   > 850.000 ads per month
   > 787 M visits in 2014*
   > 6 M Android downloads
   *source: comScore Digital Analytix,
   January 2015,
   complete IS24 (Portal & Mobile)
   

   View Slide

3. Why we are developers?
   

   View Slide

4. Why we are developers?
   ESPECIALLY ANDROID
   

   View Slide

5. Source: https://www.flickr.com/photos/theinfamousgdub/1765952198
   

   View Slide

6. We are smart!
   Source: https://www.flickr.com/photos/theinfamousgdub/1765952198
   

   View Slide

7. Source: http://giphy.com/gifs/apple-looney-tunes-android-4A41Wn7C32RiM
   

   View Slide

8. We Love Android!
   Source: http://giphy.com/gifs/apple-looney-tunes-android-4A41Wn7C32RiM
   

   View Slide

9. Source: Photo by Sujaki-F
   

   View Slide

10. We have Passion!
    Source: Photo by Sujaki-F
    

    View Slide

11. We have Passion!
    Source: Photo by Sujaki-F
    

    View Slide

12. We have Passion!
    Source: Photo by Sujaki-F
    We Love Challenges
    

    View Slide

13. Source: https://www.flickr.com/photos/68751915@N05/6793826885
    

    View Slide

14. We Love Money
    Source: https://www.flickr.com/photos/68751915@N05/6793826885
    

    View Slide

15. Source: http://agenciabrasil.ebc.com.br/geral/foto/2014-07/cerimonia-de-premiacao-da-copa-do-mundo-no-brasil
    

    View Slide

16. We love to delight our
    users
    Source: http://agenciabrasil.ebc.com.br/geral/foto/2014-07/cerimonia-de-premiacao-da-copa-do-mundo-no-brasil
    

    View Slide

17. User Expectations
    

    View Slide

18. User Expectations
    Mobile App Users are impatient
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

19. User Expectations
    Mobile App Users are impatient
    •  61% app start < 4 s
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

20. User Expectations
    Mobile App Users are impatient
    •  61% app start < 4 s
    •  49% app start < 2 s
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

21. User Expectations
    Mobile App Users are impatient
    •  61% app start < 4 s
    •  49% app start < 2 s
    Users are intolerant of issues and are quick to uninstall mobile
    apps
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

22. User Expectations
    Mobile App Users are impatient
    •  61% app start < 4 s
    •  49% app start < 2 s
    Users are intolerant of issues and are quick to uninstall mobile
    apps
    •  80% three attempts or less
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

23. User Expectations
    Mobile App Users are impatient
    •  61% app start < 4 s
    •  49% app start < 2 s
    Users are intolerant of issues and are quick to uninstall mobile
    apps
    •  80% three attempts or less
    •  53% apps with severe issues like crashes, freezes or errors
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

24. User Expectations
    Mobile App Users are impatient
    •  61% app start < 4 s
    •  49% app start < 2 s
    Users are intolerant of issues and are quick to uninstall mobile
    apps
    •  80% three attempts or less
    •  53% apps with severe issues like crashes, freezes or errors
    •  36% heavy battery usage
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

25. User Expectations
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

26. User Expectations
    Users blame the app and the company who made it
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

27. User Expectations
    Users blame the app and the company who made it
    •  55% app is responsible for performance issues
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

28. User Expectations
    Users blame the app and the company who made it
    •  55% app is responsible for performance issues
    •  37% Stated that app crashes or errors make them think less of a company’s brand
    Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
    

    View Slide

29. Performance Impacts
    

    View Slide

30. Performance Impacts
    500 ms delay
    Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs-
    Server-Side-Solutions-Infographic.jpg
    

    View Slide

31. Performance Impacts
    500 ms delay
    bounce rate -4.7% -- conversion rate -1.9%
    Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs-
    Server-Side-Solutions-Infographic.jpg
    

    View Slide

32. Performance Impacts
    500 ms delay
    bounce rate -4.7% -- conversion rate -1.9%
    1,000 ms delay
    Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs-
    Server-Side-Solutions-Infographic.jpg
    

    View Slide

33. Performance Impacts
    500 ms delay
    bounce rate -4.7% -- conversion rate -1.9%
    1,000 ms delay
    bounce rate -8.3% -- Conversion rate -3.5%
    Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs-
    Server-Side-Solutions-Infographic.jpg
    

    View Slide

34. Performance Impacts
    500 ms delay
    bounce rate -4.7% -- conversion rate -1.9%
    1,000 ms delay
    bounce rate -8.3% -- Conversion rate -3.5%
    100ms delay mean for Amazon -1%
    revenue
    Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs-
    Server-Side-Solutions-Infographic.jpg
    

    View Slide

35. Performance Impacts
    500 ms delay
    bounce rate -4.7% -- conversion rate -1.9%
    1,000 ms delay
    bounce rate -8.3% -- Conversion rate -3.5%
    100ms delay mean for Amazon -1%
    revenue
    11% scream at there device
    Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs-
    Server-Side-Solutions-Infographic.jpg
    

    View Slide

36. Performance Impacts
    500 ms delay
    bounce rate -4.7% -- conversion rate -1.9%
    1,000 ms delay
    bounce rate -8.3% -- Conversion rate -3.5%
    100ms delay mean for Amazon -1%
    revenue
    11% scream at there device
    4 % throw their phones
    Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs-
    Server-Side-Solutions-Infographic.jpg
    https://www.flickr.com/photos/
    9009139@N08/1263954439
    

    View Slide

37. View Slide

38. Source: http://giphy.com/gifs/sad-kawaii-pretty-10XOyCb09KyjLy
    

    View Slide

39. www.immobilienscout24.de
    Let‘s find some performance improvements
    

    View Slide

40. Possible improvements
    CPU
    Memory
    I/O
    

    View Slide

41. Possible improvements
    CPU
    Memory
    I/O
    

    View Slide

42. Possible improvements
    CPU
    Memory
    I/O
    •  Keep the work of the main
    thread ! ANR
    

    View Slide

43. Possible improvements
    CPU
    Memory
    I/O
    •  Keep the work of the main
    thread ! ANR
    •  Avoid GPU overdrawings
    • https://www.youtube.com/
    watch?v=T52v50r-JfE
    

    View Slide

44. Possible improvements
    CPU
    Memory
    I/O
    •  Keep the work of the main
    thread ! ANR
    •  Avoid GPU overdrawings
    • https://www.youtube.com/
    watch?v=T52v50r-JfE
    •  Avoid nested multi-pass
    layouts
    • http://goo.gl/Q5te24
    

    View Slide

45. Possible improvements
    CPU
    Memory
    I/O
    •  Keep the work of the main
    thread ! ANR
    •  Avoid GPU overdrawings
    • https://www.youtube.com/
    watch?v=T52v50r-JfE
    •  Avoid nested multi-pass
    layouts
    • http://goo.gl/Q5te24
    •  Consider RenderScript for
    performance critical code
    

    View Slide

46. Possible improvements
    CPU
    Memory
    I/O
    

    View Slide

47. Possible improvements
    CPU
    Memory
    I/O
    •  Do not allocate memory inside
    onDraw()
    

    View Slide

48. Possible improvements
    CPU
    Memory
    I/O
    •  Do not allocate memory inside
    onDraw()
    •  Avoid Leaks
    

    View Slide

49. LeakCanary to the Rescue
    https://github.com/square/
    leakcanary
    

    View Slide

50. Setup LeakCanary
    debugCompile 'com.squareup.leakcanary:leakcanary-android:1.3.1'
    releaseCompile 'com.squareup.leakcanary:leakcanary-android-no-op:1.3.1’
    build.gradle
    public class DcItApplication extends Application {
    @Override
    public void onCreate() {
    super.onCreate();
    LeakCanary.install(this);
    }
    }
    DcItApplication.java
    

    View Slide

51. Use Your Debug Build Normally
    

    View Slide

52. Analyze Memory
    

    View Slide

53. Analyze Heap Dump
    

    View Slide

54. Analyze Heap Dump
    

    View Slide

55. Analyze Heap Dump since AS 1.5
    

    View Slide

56. Analyze Memory
    

    View Slide

57. Possible improvements
    CPU
    Memory
    I/O
    •  Do not allocate memory inside
    onDraw()
    •  Avoid Leaks
    •  Avoid listener ! EventBus
    

    View Slide

58. Possible improvements
    CPU
    Memory
    I/O
    •  Do not allocate memory inside
    onDraw()
    •  Avoid Leaks
    •  Avoid listener ! EventBus
    •  Use SparseArray* family
    instead of JDK Collections/
    Maps
    

    View Slide

59. Possible improvements
    CPU
    Memory
    I/O
    •  Do not allocate memory inside
    onDraw()
    •  Avoid Leaks
    •  Avoid listener ! EventBus
    •  Use SparseArray* family
    instead of JDK Collections/
    Maps
    •  Be aware that enums need
    more resource, but use them, if
    it will make sense for you
    

    View Slide

60. Possible improvements
    CPU
    Memory
    I/O
    

    View Slide

61. Possible improvements
    CPU
    Memory
    I/O
    •  Activate gzip compression on
    incoming and outgoing
    network traffic or FlatBuffer
    

    View Slide

62. Analyze First Start
    

    View Slide

63. Analyze Second Start
    

    View Slide

64. Analyze Second Start
    

    View Slide

65. Possible improvements
    CPU
    Memory
    I/O
    •  Activate gzip compression on
    incoming and outgoing
    network traffic or FlatBuffer
    •  Cache data on disk (image,
    http responses) with
    reasonable TTL
    

    View Slide

66. Finally Second Start
    

    View Slide

67. Possible improvements
    CPU
    Memory
    I/O
    •  Activate gzip compression on
    incoming and outgoing
    network traffic or FlatBuffer
    •  Cache data on disk (image,
    http responses) with
    reasonable TTL
    •  Server side cache headers like
    ETag & Last-Modified
    

    View Slide

68. Possible improvements
    CPU
    Memory
    I/O
    •  Activate gzip compression on
    incoming and outgoing
    network traffic or FlatBuffer
    •  Cache data on disk (image,
    http responses) with
    reasonable TTL
    •  Server side cache headers like
    ETag & Last-Modified
    •  Use JobScheduler API to batch
    across system or better use
    push notifications for update
    information
    

    View Slide

69. Source: http://www.flickr.com/photos/16210667@N02/9172895225
    

    View Slide

70. Optimizing ArrayList
    

    View Slide

71. Optimizing ArrayList
    public List getDcItNames(List attendees) {
    List names = new ArrayList<>();
    for (Attendee attendee : attendees) {
    names.add(attendee.name);
    }
    return names;
    }
    JAVA
    

    View Slide

72. Optimizing ArrayList
    public List getDcItNames(List attendees) {
    List names = new ArrayList<>(attendees.size());
    for (Attendee attendee : attendees) {
    names.add(attendee.name);
    }
    return names;
    }
    JAVA
    

    View Slide

73. Optimizing ArrayList
    @Override public boolean add(E object) {
    Object[] a = array;
    int s = size;
    if (s == a.length) {
    Object[] newArray = new Object[s +
    (s < (MIN_CAPACITY_INCREMENT / 2) ?
    MIN_CAPACITY_INCREMENT : s >> 1)];
    System.arraycopy(a, 0, newArray, 0, s);
    array = a = newArray;
    }
    a[s] = object;
    size = s + 1;
    modCount++;
    return true;
    }
    JAVA
    

    View Slide

74. Optimizing ArrayList
    void addDcItSpeaker(ArrayList attendees, List speakers) {
    for (String speaker : speakers) {
    attendees.add(speaker);
    }
    }
    JAVA
    

    View Slide

75. Optimizing ArrayList
    void addDcItSpeaker(ArrayList attendees, List speakers) {
    attendees.ensureCapacity(attendees.size() + speakers.size());
    for (String speaker : speakers) {
    attendees.add(speaker);
    }
    }
    JAVA
    

    View Slide

76. Optimizing Bundle
    

    View Slide

77. Optimizing Bundle
    public static DcItScheduleFragment instance(int startMode) {
    final DcItScheduleFragment fragment = new DcItScheduleFragment();
    final Bundle bundle = new Bundle();
    bundle.putInt(START_MODE, startMode);
    fragment.setArguments(bundle);
    return fragment;
    }
    JAVA
    

    View Slide

78. Optimizing Bundle
    public static DcItScheduleFragment instance(int startMode) {
    final DcItScheduleFragment fragment = new DcItScheduleFragment();
    final Bundle bundle = new Bundle(1);
    bundle.putInt(START_MODE, startMode);
    fragment.setArguments(bundle);
    return fragment;
    }
    JAVA
    

    View Slide

79. Optimizing StringBuilder
    

    View Slide

80. Optimizing StringBuilder
    public StringBuilder getDcItNames(List names) {
    StringBuilder builder = new StringBuilder();
    for (String name : names) {
    if (builder.length() > 0) {
    builder.append(", ");
    }
    builder.append(name);
    }
    return builder;
    }
    JAVA
    

    View Slide

81. Optimizing StringBuilder
    public StringBuilder getDcItNames(List names) {
    StringBuilder builder = new StringBuilder(names.size() * 13);
    for (String name : names) {
    if (builder.length() > 0) {
    builder.append(", ");
    }
    builder.append(name);
    }
    return builder;
    }
    JAVA
    

    View Slide

82. Optimizing StringBuilder
    final void append0(char[] chars) {
    int newCount = count + chars.length;
    if (newCount > value.length) {
    enlargeBuffer(newCount);
    }
    System.arraycopy(chars, 0, value, count, chars.length);
    count = newCount;
    }
    JAVA
    

    View Slide

83. Optimizing StringBuilder
    public StringBuilder addDcItOrganizer(StringBuilder names, List organizers)
    {
    names.ensureCapacity(names.length() + organizers.size() * 13);
    for (String organizer : organizers) {
    if (names.length() > 0) {
    names.append(", ");
    }
    names.append(organizer);
    }
    }
    JAVA
    

    View Slide

84. Optimizing Loops
    

    View Slide

85. Optimizing Moar – The Loop
    public StringBuilder addDcItOrganizer(StringBuilder names, List organizers)
    {
    names.ensureCapacity(names.length() + organizers.size() * 13);
    for (String organizer : organizers) {
    if (names.length() > 0) {
    names.append(", ");
    }
    names.append(organizer);
    }
    }
    JAVA
    

    View Slide

86. Optimizing Moar – The Loop
    public StringBuilder addDcItOrganizer(StringBuilder names, List organizers)
    {
    names.ensureCapacity(names.length() + organizers.size() * 13);
    for (int i = 0; i < organizers.size(); i++) {
    String organizer = organizers.get(i);
    if (names.length() > 0) {
    names.append(", ");
    }
    names.append(organizer);
    }
    }
    JAVA
    

    View Slide

87. Optimizing Moar – The Loop
    public StringBuilder addDcItOrganizer(StringBuilder names, List organizers)
    {
    names.ensureCapacity(names.length() + organizers.size() * 13);
    for (int i = 0, organizersSize = organizers.size(); i < organizersSize; i++) {
    String organizer = organizers.get(i);
    if (names.length() > 0) {
    names.append(", ");
    }
    names.append(organizer);
    }
    }
    JAVA
    

    View Slide

88. Optimizing Moar – The Loop
    public StringBuilder addDcItOrganizer(StringBuilder names, List organizers)
    {
    final int size = organizers.size();
    names.ensureCapacity(names.length() + size * 13);
    for (int i = 0; i < size; i++) {
    String organizer = organizers.get(i);
    if (names.length() > 0) {
    names.append(", ");
    }
    names.append(organizer);
    }
    }
    JAVA
    

    View Slide

89. Optimizing Method invocations
    

    View Slide

90. Optimizing Much Moar – Method Invocation
    public void onViewCreated(View view, Bundle savedInstanceState) {
    super.onViewCreated(view, savedInstanceState);
    text.setTextColor(attendee.isOrganizer() ?
    getResources().getColor(R.color.accent) :
    getResources().getColor(R.color.black));
    avatar.setImageResource(attendee.getAvatar());
    welcomeText.setText(getString(
    R.string.hello_world, attendee.name));
    }
    JAVA
    

    View Slide

91. Optimizing Much Moar – Method Invocation
    public void onViewCreated(View view, Bundle savedInstanceState) {
    super.onViewCreated(view, savedInstanceState);
    final Resources res = getResources();
    text.setTextColor(attendee.isOrganizer() ?
    res.getColor(R.color.accent) :
    res.getColor(R.color.black));
    avatar.setImageResource(attendee.getAvatar());
    welcomeText.setText(getString(
    R.string.hello_world, attendee.name));
    }
    JAVA
    

    View Slide

92. Optimizing Much Moar – Method Invocation
    public void onViewCreated(View view, Bundle savedInstanceState) {
    super.onViewCreated(view, savedInstanceState);
    final Resources res = getResources();
    text.setTextColor(attendee.isOrganizer() ?
    res.getColor(R.color.accent) :
    res.getColor(R.color.black));
    avatar.setImageResource(attendee.getAvatar());
    welcomeText.setText(res.getString(
    R.string.hello_world, attendee.name));
    }
    JAVA
    

    View Slide

93. Optimizing Access Methods
    

    View Slide

94. Optimizing Access Methods
    private Attendee attendee;
    public void onViewCreated(View view, Bundle savedInstanceState) {
    button.setOnClickListener(new OnClickListener() {
    @Override
    public void onClick(View v) {
    makeText(getContext(), getWelcomeTextFor(attendee.getName()),
    LENGTH_SHORT).show();
    }
    });
    }
    private String getWelcomeTextFor(String name) {
    return getResources().getString(R.string.welcome_text, name);
    }
    JAVA
    

    View Slide

95. Finding Access Methods
    https://github.com/JakeWharton/dex-method-list
    $ ./build/dex-method-list src/test/resources/one.apk
    …
    package.DcItScheduleFragment access$000(DcItScheduleFragment )
    package.DcItScheduleFragment access$100(DcItScheduleFragment , String)
    …
    CL
    

    View Slide

96. Optimizing Access Methods
    Attendee attendee;
    public void onViewCreated(View view, Bundle savedInstanceState) {
    button.setOnClickListener(new OnClickListener() {
    @Override
    public void onClick(View v) {
    makeText(getContext(), getWelcomeTextFor(attendee.getName()),
    LENGTH_SHORT).show();
    }
    });
    }
    String getWelcomeTextFor(String name) {
    return getResources().getString(R.string.welcome_text, name);
    }
    JAVA
    

    View Slide

97. Source: https://www.flickr.com/photos/marcus_t_ward/10715983613
    

    View Slide

98. Source: http://www.bhmpics.com/success_kid-wallpapers.html
    

    View Slide

99. Source: http://www.flickr.com/photos/21496790@N06/5065834411/
    

    View Slide

100. www.immobilienscout24.de
     Contact:
     +HasanHosgel
     alosdev
     Thanks for attending!
     We are Hiring! http://www.immobilienscout24.de/jobs
     https://speakerdeck.com/alosdev/perfmatters-for-android-
     droidcon-turin-2016
     

     View Slide

101. View Slide

102. View Slide

103. View Slide

104. View Slide

105. www.immobilienscout24.de
     Contact:
     +HasanHosgel
     alosdev
     Thanks for attending!
     We are Hiring! http://www.immobilienscout24.de/jobs
     https://speakerdeck.com/alosdev/perfmatters-for-android-
     droidcon-turin-2016
     

     View Slide

106. Sources
     •  https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
     •  http://www.mobilejoomla.com/media/press/responsive-vs-
     serverside/Responsive-Design-vs-Server-Side-Solutions-
     Infographic.jpg
     •  https://speakerdeck.com/jakewharton/eliminating-code-overhead-
     square-hq-2015
     •  http://periplanisi.com/android/2013/11/multi-pass-viewgroup-
     and-performance/
     •  https://slideshare.net/dougsillars/android-app-performance-
     europe-2015
     •  https://www.youtube.com/playlist?
     list=PLWz5rJ2EKKc9CBxr3BVjPTPoDPLdPIFCE
     

     View Slide