This document summarizes a presentation on serialization frameworks and their performance. It discusses the purpose of serialization, common data formats like binary, XML, and JSON, and several Java serialization frameworks including Java serialization, Kryo, Protocol Buffers, Jackson, Google GSON, and others. Benchmark results show that Kryo and Protocol Buffers generally have the best performance for serialization and deserialization speed as well as output size. The document provides recommendations on when to use different frameworks based on usage cases and priorities for speed, size, or flexibility.

1. Serialization and
performance
Сергей Моренец
23 мая 2014 г.

2. About author
Works in IT since 2000
10 year of Java SE/EE experience
Occupied senior Java developer/Team
Lead positions
Winner of 2013 JBoss Community
Recognition Award.
https://www.jboss.org/jbcra

3. Agenda
• Purpose of serialization
• Frameworks overview
• Performance testing
• Q & A

4. Serialization
File storages
Database
Network communication
Web usage

5. Serialization
Simple
Flexible
Compact
Versioning
Fast Scalable

6. Data formats
Binary
XML
JSON
YAML

7. Performance
• Native memory copying using C operations
• “Unsafe” operations
• Ignore object introspection
• Direct object-object copying

8. Java serialization
• The easiest programming effort
• Out-of-the-box functionality

9. Java serialization
• Serializable interface
• Decreases the flexibility to change a class’s
implementation once it has been released
• Doesn’t allow to exchange data with
C++/Python applications
• Due to default constructors hole for invariant
corruption and illegal access
• No customization
• You should have access to the source code
No customization
You should have access to the source code

10. Java externalization
• Serialization but by implementing
Externalizable interface to persist and
restore the object
• Responsibility of the class to save and
restore the contents of its instances
• Requires modifications in
marshalling/unmarshalling code if the class
contents changed
No customization
You should have access to the source code

11. Java externalization

12. Avro
• Schema evolution
• Binary and JSON encoding
• Dynamic typing
• Support of Java, C, C++, C# and Python
• Apache Hadoop integration

13. Avro

14. Avro

15. XML
• Interchangeable format
• Supported schemas
• Space intensive and huge performance loss
• Complex navigating

16. Simple
• High performance XML serialization and
configuration framework for Java.
• Requires absolutely no configuration
• Can handle cycles in the object graph

17. Simple

18. Javolution
• Fast real-time library for safety-critical
applications
• Based on OSGi context
• Parallel computing support

19. Javolution

20. Json-io
• Doesn’t require custom interfaces/attributes
usage/source code
• Handles cyclic references
• Reader/writer customization
• Does not depend on any native or 3rd party
libraries.

21. Google gson
• Java library to convert JSON to Java objects and
vice-versa
• Doesn’t require source code of serialized objects
• Allow custom representatives

22. Google gson

23. Jackson
• High-performance, ergonomic JSON processor
Java library
• Extensive customization tools
• Mix-in annotations
• Materialized interfaces
• Multiple data formats

24. Jackson
• JSON
• CSV
• Smile(binary JSON)
• XML
• YAML(similar to JSON)

25. BSON for Jackson
• Binary encoded JSON
• Main data exchange format for MongoDB
• Allows writing custom extensions

26. Protocol buffers
• Way of encoding structured data in an efficient
yet extensible format.
• Google uses Protocol Buffers for almost all of
its internal RPC protocols and file formats.
• Supported in Java, C++, Python

27. Protocol buffers
message User {
required string login = 1;
repeated Order orders = 2;
}
message Order {
required int32 id = 1;
optional string date = 2;
}

28. Protocol buffers

29. FST
• Java-to-java library
• No support for versioning
• Use case is high performance message oriented
software
• Drop-in replacement
• Custom optimization using annotations, custom
serializers

30. FST

31. GridGain
• Part of distributed computing system
• Don’t require any custom interfaces or API
• Direct memory copying by invoking native
"unsafe" operations
• Predefined fields introspection

32. GridGain

33. Kryo
• Fast and efficient object graph serialization
framework for Java
• Open source project on Google code
• Automatic deep and shallow copying/cloning
• Doesn’t put requirements on the source classes
(in most cases)

34. Kryo
• Twitter
• Apache Hive
• Akka
• Storm
• S4

35. Kryo

36. Kryo

37. Benchmark
• JDK 1.8.0.5
• Apache Avro 1.7.6
• Simple 2.7.1
• Json-io 2.5.2
• Google GSON 2.2.4
• Jackson 2.3.2
• BSON for Jackson 2.3.1
• Protocol buffers 2.5
• Kryo 2.23
• FST 1.54
• GridGain 6.0.2

38. Benchmark
• Speed(serialization and deserialization)
• Size(complex and ordinary objects)
• Flexibility

39. Benchmark

40. Benchmark

41. Issues
Library Description
Gson, Jackson Crashed when serializing cyclic
dependency
Simple Crashed for very big XML file
Avro Bug during deserialization

42. Serialization (complex)
# Library Time(ms)
1 Kryo(optimized) 134
2 Protocol buffers 165
3 GridGain 196
4 FST 207
5 Kryo 209
6 Jackson(smile) 275
7 Kryo(unsafe) 306
8 Jackson 491
9 Java serialization 605
10 Javolution 1043

43. Serialization (simple)
# Library Time(ms)
1 Protocol buffers <1
2 Google GSON 5
3 Java serialization 10
4 BSON for Jackson 10
5 Jackson(smile) 11
6 Kryo(optimized) 17
7 Kryo 18
8 Jackson 18
9 Kryo(unsafe) 20
10 Javolution 21

44. Deserialization (complex)
# Library Time(ms)
1 Kryo(optimized) 113
2 Protocol buffers 165
3 GridGain 196
4 FST 207
5 Kryo 209
6 Jackson(smile) 275
7 Kryo(unsafe) 306
8 Jackson 491
9 Java serialization 605
10 BSON for Jackson 930

45. Deserialization (simple)
# Library Time(ms)
1 Protocol buffers 1
2 GridGain 3
3 Google GSON 6
4 Jackson(smile) 9
5 BSON for Jackson 9
6 Kryo(optimized) 18
7 Kryo 18
8 Jackson 18
9 Kryo(unsafe) 20
10 FST 42

46. Size (complex)
# Library Size(bytes)
1 Kryo(optimized) 33904
2 FST 34069
3 Kryo 35674
4 Protocol buffers 39517
5 Kryo(unsafe) 40554
6 Jackson(smile) 44840
7 Java serialization 49757
8 GridGain 58288
9 Jackson 67858
10 Google GSON 68338

47. Size (simple)
# Library Size(bytes)
1 Kryo(optimized) 18
2 Kryo 18
3 Protocol buffers 20
4 Kryo(unsafe) 21
5 GridGain 33
6 Jackson(smile) 40
7 Jackson 41
8 Google GSON 41
9 Jackson(YAML) 41
10 BSON for Jackson 46

48. Usability
# Library
1 Google GSON
2 Kryo
2 Kryo(unsafe)
3 Jackson
3 Jackson(XML, Smile, YAML)
3 BSON for Jackson
4 Json-io
5 FST
6 Java serialization
7 Kryo(optimized)

49. Overall rating (2014)
# Library Rating
1 Kryo(optimized) 67
2 Protocol buffers 65
3 Kryo 58
4 Jackson(smile) 55
5 Kryo(unsafe) 46
6 GridGain 44
7 Google GSON 43
8 FST 43
9 Jackson 40
10 BSON for Jackson 33
11 Java serialization 32

50. Overall rating (2013)
# Library Rating
1 Kryo(optimized) 67
2 Kryo(unsafe) 65
3 Protocol buffers 63
4 Kryo 59
5 Jackson(smile) 51
6 Google GSON 45
7 FST 42
8 GridGain 34
9 Jackson 32
10 Java serialization 30
11 BSON for Jackson 24

51. Advices
Library Usage
Kryo Fast and compact serializer for complex
objects over network
Protocol buffers Fast serializer for simple objects
Jackson(smile) Jackson-based serializer for Web usage
Google JSON Dirty solution to quickly serialize/deserialize
objects
Apache Avro Serialize objects into files with possible
schema changes
Java Out-of-the-box trusted solution without
additional libraries

52. Сергей Моренец
morenets@mail.ru
Q&A