PERFMATTERS for Android - DroidCon Berlin 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. #DroidconDE www.immobilienscout24.de Hasan Hosgel – Droidcon Berlin 2016 #PERFMATTERS for Android
2. #DroidconDE #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)
3. #DroidconDE Why we are developers?
4. #DroidconDE Why we are developers? ESPECIALLY ANDROID
5. Source: https://www.flickr.com/photos/theinfamousgdub/1765952198
6. We are smart! Source: https://www.flickr.com/photos/theinfamousgdub/1765952198
7. Source: http://giphy.com/gifs/apple-looney-tunes-android-4A41Wn7C32RiM
8. We Love Android! Source: http://giphy.com/gifs/apple-looney-tunes-android-4A41Wn7C32RiM
9. Source: Photo by Sujaki-F
10. We have Passion! Source: Photo by Sujaki-F
11. We have Passion! Source: Photo by Sujaki-F
12. We have Passion! Source: Photo by Sujaki-F We Love Challenges
13. Source: https://www.flickr.com/photos/68751915@N05/6793826885
14. We Love Money Source: https://www.flickr.com/photos/68751915@N05/6793826885
15. Source: http://agenciabrasil.ebc.com.br/geral/foto/2014-07/cerimonia-de-premiacao-da-copa-do-mundo-no-brasil
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
17. #DroidconDE User Expectations
18. #DroidconDE User Expectations Mobile App Users are impatient Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
19. #DroidconDE 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
20. #DroidconDE 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
21. #DroidconDE 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
22. #DroidconDE 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
23. #DroidconDE 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
24. #DroidconDE 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
25. #DroidconDE User Expectations Source: https://ssl.www8.hp.com/ww/en/secure/pdf/4aa5-7696enw.pdf
26. #DroidconDE 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
27. #DroidconDE 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
28. #DroidconDE 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
29. #DroidconDE Performance Impacts
30. #DroidconDE Performance Impacts 500 ms delay Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs- Server-Side-Solutions-Infographic.jpg
31. #DroidconDE 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
32. #DroidconDE 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
33. #DroidconDE 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
34. #DroidconDE 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
35. #DroidconDE 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 their device Source: http://www.mobilejoomla.com/media/press/responsive-vs-serverside/Responsive-Design-vs- Server-Side-Solutions-Infographic.jpg
36. #DroidconDE 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 their 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
38. Source: http://giphy.com/gifs/sad-kawaii-pretty-10XOyCb09KyjLy
39. #DroidconDE www.immobilienscout24.de Let‘s find some performance improvements
40. #DroidconDE Possible improvements CPU Memory I/O
41. #DroidconDE Possible improvements CPU Memory I/O
42. #DroidconDE Possible improvements CPU Memory I/O •  Keep the work of the main thread ! ANR
43. #DroidconDE 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
44. #DroidconDE 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
45. #DroidconDE 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
46. #DroidconDE Possible improvements CPU Memory I/O
47. #DroidconDE Possible improvements CPU Memory I/O •  Do not allocate memory inside onDraw()
48. #DroidconDE Possible improvements CPU Memory I/O •  Do not allocate memory inside onDraw() •  Avoid Leaks
49. #DroidconDE LeakCanary to the Rescue https://github.com/square/ leakcanary
50. #DroidconDE 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
51. #DroidconDE Use Your Debug Build Normally
52. #DroidconDE Analyze Memory
53. #DroidconDE Analyze Heap Dump
54. #DroidconDE Analyze Heap Dump
55. #DroidconDE Analyze Heap Dump since AS 1.5
56. #DroidconDE Analyze Memory
57. #DroidconDE Possible improvements CPU Memory I/O •  Do not allocate memory inside onDraw() •  Avoid Leaks •  Avoid listener ! EventBus
58. #DroidconDE 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
59. #DroidconDE 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
60. #DroidconDE Possible improvements CPU Memory I/O
61. #DroidconDE Possible improvements CPU Memory I/O •  Activate gzip compression on incoming and outgoing network traffic or FlatBuffer
62. #DroidconDE Analyze First Start
63. #DroidconDE Analyze Second Start
64. #DroidconDE Analyze Second Start
65. #DroidconDE 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
66. #DroidconDE Finally Second Start
67. #DroidconDE 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
68. #DroidconDE 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
69. Source: http://www.flickr.com/photos/16210667@N02/9172895225
70. #DroidconDE Optimizing ArrayList
71. #DroidconDE Optimizing ArrayList public List<String> getDcItNames(List<Attendee> attendees) { List<String> names = new ArrayList<>(); for (Attendee attendee : attendees) { names.add(attendee.name); } return names; } JAVA
72. #DroidconDE Optimizing ArrayList public List<String> getDcItNames(List<Attendee> attendees) { List<String> names = new ArrayList<>(attendees.size()); for (Attendee attendee : attendees) { names.add(attendee.name); } return names; } JAVA
73. #DroidconDE 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
74. #DroidconDE Optimizing ArrayList void addDcItSpeaker(ArrayList<String> attendees, List<String> speakers) { for (String speaker : speakers) { attendees.add(speaker); } } JAVA
75. #DroidconDE Optimizing ArrayList void addDcItSpeaker(ArrayList<String> attendees, List<String> speakers) { attendees.ensureCapacity(attendees.size() + speakers.size()); for (String speaker : speakers) { attendees.add(speaker); } } JAVA
76. #DroidconDE Optimizing Bundle
77. #DroidconDE 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
78. #DroidconDE 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
79. #DroidconDE Optimizing StringBuilder
80. #DroidconDE Optimizing StringBuilder public StringBuilder getDcItNames(List<String> names) { StringBuilder builder = new StringBuilder(); for (String name : names) { if (builder.length() > 0) { builder.append(", "); } builder.append(name); } return builder; } JAVA
81. #DroidconDE Optimizing StringBuilder public StringBuilder getDcItNames(List<String> names) { StringBuilder builder = new StringBuilder(names.size() * 13); for (String name : names) { if (builder.length() > 0) { builder.append(", "); } builder.append(name); } return builder; } JAVA
82. #DroidconDE 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
83. #DroidconDE Optimizing StringBuilder public StringBuilder addDcItOrganizer(StringBuilder names, List<String> organizers) { names.ensureCapacity(names.length() + organizers.size() * 13); for (String organizer : organizers) { if (names.length() > 0) { names.append(", "); } names.append(organizer); } } JAVA
84. #DroidconDE Optimizing Loops
85. #DroidconDE Optimizing Moar – The Loop public StringBuilder addDcItOrganizer(StringBuilder names, List<String> organizers) { names.ensureCapacity(names.length() + organizers.size() * 13); for (String organizer : organizers) { if (names.length() > 0) { names.append(", "); } names.append(organizer); } } JAVA
86. #DroidconDE Optimizing Moar – The Loop public StringBuilder addDcItOrganizer(StringBuilder names, List<String> 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
87. #DroidconDE Optimizing Moar – The Loop public StringBuilder addDcItOrganizer(StringBuilder names, List<String> 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
88. #DroidconDE Optimizing Moar – The Loop public StringBuilder addDcItOrganizer(StringBuilder names, List<String> 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
89. #DroidconDE Optimizing Moar – The Loop public StringBuilder addDcDeOrganizer(StringBuilder names, List<String> organizers) { final int size = organizers.size(); names.ensureCapacity(names.length() + size * 13); for (int i = 0; i < size; i++) { String organizer = organizers.get(i); names.append(organizer); names.append(", "); } names.substring(0, names.length()-1); } JAVA
90. #DroidconDE Optimizing Method invocations
91. #DroidconDE 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
92. #DroidconDE 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
93. #DroidconDE 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
94. #DroidconDE Optimizing Access Methods
95. #DroidconDE 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
96. #DroidconDE 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
97. #DroidconDE 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
98. Source: https://www.flickr.com/photos/marcus_t_ward/10715983613
99. Source: http://www.bhmpics.com/success_kid-wallpapers.html
100. Source: http://www.flickr.com/photos/21496790@N06/5065834411/
101. #DroidconDE 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-berlin-2016
102. #DroidconDE
103. #DroidconDE
104. #DroidconDE
105. #DroidconDE
106. #DroidconDE 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-berlin-2016
107. #DroidconDE 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