Don’t Get Lost in Translation for Serializing Data Structures

The current solutions for mapping data to java objects come in many shapes and sizes, with JSON leading as the most familiar and “coolest” solution. Yes, JSON is very readable, convenient, and offers great support libraries like GSON and Jackson, but these options don’t always get the job done. Accessing and parsing serialized data is a common source of runtime inefficiency within applications, and this delay can crush your app’s overall user experience. How can you bypass this inefficiency? Google’s FlatBuffer library is the answer. FlatBuffers are similar to Google’s Protocol Buffers, but with one key differentiator: the ability to access serialized data without parsing or unpacking it first. Imagine a serialization process with no temporary objects, no additional allocation, and no copying. Join us for a deep dive into Google’s FlatBuffers library to learn more about the advantages of using FlatBuffers and what makes it different from other commonly used libraries.

Transcript

1. Don’t Get Lost in Translation for Serializing Data Structures Christopher Brown Principal Mobile Developer – Team Lead, priceline.com DroidCon NYC August 28th, 2015
3. {          “trips”:  [    “offerDetails”:  {...},    “customer”:  {...},    “requestId”:  “...”,    ...          ]   }   10 trips Light Booker
4. {          “trips”:  [    “offerDetails”:  {...},    “customer”:  {...},    “requestId”:  “...”,    ...    ...    ...    ...    ...    ...   ...          ]   }   100 trips More Frequent Booker
5. {          “trips”:  [    “offerDetails”:  {...},    “customer”:  {...},    “requestId”:  “...”,    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...   ...    ...   ...    ...    ...    ...    ...   ...    ...    ...          ]   }   500 trips Heavy Booker
6. JSONObject
7. new  JSONObject(                                );   {          “trips”:  [    “offerDetails”:  {...},    “customer”:  {...},    “requestId”:  “...”,    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...   ...    ...   ...          ]   }    
8. new  JSONObject(                                );     {          “trips”:  [    “offerDetails”:  {...},    “customer”:  {...},    “requestId”:  “...”,    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...   ...    ...   ...          ]   }    
9. new  JSONObject(                                );     {          “trips”:  [    “offerDetails”:  {...},    “customer”:  {...},    “requestId”:  “...”,    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...    ...   ...    ...   ...          ]   }    
10. How can we improve the performance? Let’s experiment with different serialization libraries...
11. Gson
12. Gson Provides simple toJson() and fromJson() methods
13. Gson Provides simple toJson() and fromJson() methods Provides streaming access
14. Gson Provides simple toJson() and fromJson() methods Provides streaming access Operates as sequence of tokens traversed in depth-first order
15. Gson Provides simple toJson() and fromJson() methods Provides streaming access Operates as sequence of tokens traversed in depth-first order Minimal memory overhead
16. Implementation example public  List<Trip>  getListOfTrips(  InputStream  inputStream  )  throws  IOException  {                  Gson  gson  =  new  Gson();                                    JsonReader  reader  =  new  JsonReader(  new  InputStreamReader(  inputStream  )  );                    List<Trip>  trips  =  new  ArrayList<>();                    reader.beginArray();                    while  (  reader.hasNext()  )  {                          Trip  trip  =  gson.fromJson(  reader,  Trip.class  );                            trips.add(  trip  );                  }                    reader.endArray();                    reader.close();                    return  trips;   }    
17. Implementation example public  List<Trip>  getListOfTrips(  InputStream  inputStream  )  throws  IOException  {                  Gson  gson  =  new  Gson();                                    JsonReader  reader  =  new  JsonReader(  new  InputStreamReader(  inputStream  )  );                    List<Trip>  trips  =  new  ArrayList<>();                    reader.beginArray();                    while  (  reader.hasNext()  )  {                          Trip  trip  =  gson.fromJson(  reader,  Trip.class  );                            trips.add(  trip  );                  }                    reader.endArray();                    reader.close();                    return  trips;   }      
18. Implementation example public  List<Trip>  getListOfTrips(  InputStream  inputStream  )  throws  IOException  {                  Gson  gson  =  new  Gson();                                    JsonReader  reader  =  new  JsonReader(  new  InputStreamReader(  inputStream  )  );                    List<Trip>  trips  =  new  ArrayList<>();                    reader.beginArray();                    while  (  reader.hasNext()  )  {                          Trip  trip  =  gson.fromJson(  reader,  Trip.class  );                            trips.add(  trip  );                  }                    reader.endArray();                    reader.close();                    return  trips;   }      
19. Implementation example public  List<Trip>  getListOfTrips(  InputStream  inputStream  )  throws  IOException  {                  Gson  gson  =  new  Gson();                                    JsonReader  reader  =  new  JsonReader(  new  InputStreamReader(  inputStream  )  );                    List<Trip>  trips  =  new  ArrayList<>();                    reader.beginArray();                    while  (  reader.hasNext()  )  {                          Trip  trip  =  gson.fromJson(  reader,  Trip.class  );                            trips.add(  trip  );                  }                    reader.endArray();                    reader.close();                    return  trips;   }      
20. Gson performance 87.3 ms 56 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 10 Trips 100 Trips 500 Trips
21. Gson performance 87.3 ms 56 KB 142.9 ms 399 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 10 Trips 100 Trips 500 Trips
22. Gson performance 10 Trips 100 Trips 500 Trips 87.3 ms 56 KB 142.9 ms 399 KB 459.2 ms 599 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1
23. Jackson
24. Jackson Offers 3 alternative methods for processing JSON (Streaming API, Tree Model, Data Binding) Default Jackson only handles strictly valid JSON but can be configured Provides simple readValue() and readTree() methods ObjectMapper caching  
25. Jackson implementation example public  Trips  getTrips(  String  json  )  {          ObjectMapper  objectMapper  =  new  ObjectMapper();            return  objectMapper.readValue(  json,  Trips.class  );   }    
26. Jackson performance 351 ms 146 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 10 Trips 100 Trips 500 Trips
27. Jackson performance 351 ms 146 KB 455 ms 386 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 10 Trips 100 Trips 500 Trips
28. Jackson performance 10 Trips 100 Trips 500 Trips 351 ms 146 KB 636 ms 837 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 455 ms 386 KB
29. Protocol Buffers
30. Protocol Buffers Language and platform neutral mechanism for serializing structured data
31. Protocol Buffers Language and platform neutral mechanism for serializing structured data You specify how you want the information you’re serializing to be structured by defining protocol buffer message types in .proto files
32. Protocol Buffers Language and platform neutral mechanism for serializing structured data You specify how you want the information you’re serializing to be structured by defining protocol buffer message types in .proto files Each protocol buffer message is a logical record of information containing name- value pairs
33. Protocol Buffers Language and platform neutral mechanism for serializing structured data You specify how you want the information you’re serializing to be structured by defining protocol buffer message types in .proto files Each protocol buffer message is a logical record of information containing name- value pairs .proto files are used to generate special source code which allows you to populate, serialize, and parse protocol buffer objects
34. Protocol Buffers Language and platform neutral mechanism for serializing structured data You specify how you want the information you’re serializing to be structured by defining protocol buffer message types in .proto files Each protocol buffer message is a logical record of information containing name- value pairs .proto files are used to generate special source code which allows you to populate, serialize, and parse protocol buffer objects Protocol buffer messages are serialized into a tiny binary format. Serializing and parsing protocol buffer messages is much faster than JSON.
35. Protocol Buffers .proto file package  TripsList;     option  java_package  =  "com.priceline.droidcon.model.proto";   option  java_outer_classname  =  "TripsProtos";     message  Trips  {    ...   }   message  Trip  {    ...   }   message  OfferDetails  {    ...   }   message  PrimaryOffer  {    ...   }   message  Customer  {    ...   }  
36. Protocol Buffers .proto file package  TripsList;     option  java_package  =  "com.priceline.droidcon.model.proto";   option  java_outer_classname  =  "TripsProtos";     message  Trips  {    ...   }   message  Trip  {    ...   }   message  OfferDetails  {    ...   }   message  PrimaryOffer  {    ...   }   message  Customer  {    ...   }  
37. Protocol Buffers .proto file package  TripsList;     option  java_package  =  "com.priceline.droidcon.model.proto";   option  java_outer_classname  =  "TripsProtos";     message  Trips  {    ...   }   message  Trip  {    ...   }   message  OfferDetails  {    ...   }   message  PrimaryOffer  {    ...   }   message  Customer  {    ...   }  
38. Protocol Buffers .proto file package  TripsList;     option  java_package  =  "com.priceline.droidcon.model.proto";   option  java_outer_classname  =  "TripsProtos";     message  Trips  {    ...   }   message  Trip  {    ...   }   message  OfferDetails  {    ...   }   message  PrimaryOffer  {    ...   }   message  Customer  {    ...   }  
39. Protocol Buffers implementation example public  List<Trip>  getListOfTrips(  byte[]  data  )  {   Trips  trips  =  Trips.parseFrom(  data  );              return  trips.getTripList();   }    
40. Protocol Buffers implementation example public  List<Trip>  getListOfTrips(  byte[]  data  )  {   Trips  trips  =  Trips.parseFrom(  data  );              return  trips.getTripList();   }    
41. Protocol Buffers implementation example public  List<Trip>  getListOfTrips(  byte[]  data  )  {   Trips  trips  =  Trips.parseFrom(  data  );              return  trips.getTripList();   }    
42. Protocol Buffers performance 3.816 ms 264 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 10 Trips 100 Trips 500 Trips
43. Protocol Buffers performance 3.816 ms 264 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 24.231 ms 352 KB 10 Trips 100 Trips 500 Trips
44. Protocol Buffers performance 10 Trips 100 Trips 500 Trips 3.816 ms 264 KB 24.231 ms 352 KB 102.355 ms 1,728 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1
45. ...but, there’s something faster!
46. FlatBuffers
47. FlatBuffers Efficient open-source, cross platform serialization library
48. Efficient open-source, cross platform serialization library Similar to Protocol Buffers but allows you to access information without parsing the data which results in much better memory performance FlatBuffers
49. FlatBuffers Efficient open-source, cross platform serialization library Similar to Protocol Buffers but allows you to access information without parsing the data which results in much better memory performance A schema let’s you define how you want to structure your data
50. FlatBuffers Efficient open-source, cross platform serialization library Similar to Protocol Buffers but allows you to access information without parsing the data which results in much better memory performance A schema let’s you define how you want to structure your data FlatBuffer is a binary buffer containing data organized by offsets
51. FlatBuffers Efficient open-source, cross platform serialization library Similar to Protocol Buffers but allows you to access information without parsing the data which results in much better memory performance A schema let’s you define how you want to structure your data FlatBuffer is a binary buffer containing data organized by offsets Faster than JSON and smaller memory footprint than Protocol Buffers
52. FlatBuffers schema namespace  TripsList;     table  Trips  {    trips  :  [Trip];   }     table  Trip  {    ...   }     table  OfferDetails  {    ...   }     ...     root_type  Trips;      
53. FlatBuffers schema namespace  TripsList;     table  Trips  {    trips  :  [Trip];   }     table  Trip  {    ...   }     table  OfferDetails  {    ...   }     ...     root_type  Trips;      
54. FlatBuffers schema namespace  TripsList;     table  Trips  {    trips  :  [Trip];   }     table  Trip  {    ...   }     table  OfferDetails  {    ...   }     ...     root_type  Trips;      
55. FlatBuffers schema namespace  TripsList;     table  Trips  {    trips  :  [Trip];   }     table  Trip  {    ...   }     table  OfferDetails  {    ...   }     ...     root_type  Trips;      
56. FlatBuffers implementation example public  List<Trip>  getListOfTrips(  byte[]  data  )  {   Trips  trips  =  Trips.getRootAsTrips(  ByteBuffer.wrap(  data  )  );            List<Trip>  list  =  new  ArrayList<>();              for  (  int  i  =  0;  i  <  trips.tripsLength();  i++  )  {              list.add(  trips.trips(  i  )  );            }              return  list;   }      
57. FlatBuffers implementation example public  List<Trip>  getListOfTrips(  byte[]  data  )  {   Trips  trips  =  Trips.getRootAsTrips(  ByteBuffer.wrap(  data  )  );            List<Trip>  list  =  new  ArrayList<>();              for  (  int  i  =  0;  i  <  trips.tripsLength();  i++  )  {              list.add(  trips.trips(  i  )  );            }              return  list;   }      
58. FlatBuffers implementation example public  List<Trip>  getListOfTrips(  byte[]  data  )  {   Trips  trips  =  Trips.getRootAsTrips(  ByteBuffer.wrap(  data  )  );            List<Trip>  list  =  new  ArrayList<>();              for  (  int  i  =  0;  i  <  trips.tripsLength();  i++  )  {              list.add(  trips.trips(  i  )  );            }              return  list;   }      
59. FlatBuffers performance 10 Trips 1.304 ms 3.8 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 100 Trips 500 Trips
60. FlatBuffers performance 10 Trips 1.304 ms 3.8 KB 6.153 ms 6.944 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 100 Trips 500 Trips
61. FlatBuffers performance 10 Trips 100 Trips 500 Trips 1.304 ms 3.8 KB 6.153 ms 6.944 KB 13.704 ms 13 KB Incl. CPU Time Memory Allocation Tests were performed on a Motorola Nexus 6 running Android 5.1.1 Protocol Buffers On average 350% faster than Protocol Buffers
62. Overall Time Results Gson Jackson Protocol Buffers Flat Buffers 459 ms 14 ms Note: Time results shown for 500 trips 102 ms 636 ms
63. Our Selection FlatBuffers was faster and more efficient, but… Image credits: The Noun Project – Dalpat Prajapati
64. Our Selection FlatBuffers was faster and more efficient, but… Our final decision was Protocol Buffers
65. Our Selection FlatBuffers was faster and more efficient, but… Our final decision was Protocol Buffers ...it’s not the fastest, but it was the most optimal for us
66. Our Selection FlatBuffers was faster and more efficient, but… Our final decision was Protocol Buffers ...it’s not the fastest, but it was the most optimal for us FlatBuffers is still a great way to serialize and deserialize data
67. Wrap Up Don’t get lost in translation for serializing your data structures Image credits: The Noun Project – Hello Many
68. Wrap Up Don’t get lost in translation for serializing your data structures Different options available (Gson → Jackson → Protocol Buffers → FlatBuffers)
69. Wrap Up Don’t get lost in translation for serializing your data structures Different options available (Gson → Jackson → Protocol Buffers → FlatBuffers) FlatBuffers are the best balance of speed and memory efficiency (FlatBuffers won the battle, it did not win the war) Image credits: The Noun Project - Rediffusion
70. Questions? christopher.brown@priceline.com