A Scala JSON serialization/deserialization benchmark

Introduction

When decision about JSON library should be made because of performance, this benchmark can be helpful. Benchmarked libraries:

• Argonaut
• circe
• JSON4S
• Lift
• spray-json
• play-json
• jsoniter-scala

This project also demonstrates how to use these different JSON libraries to serialize/deserialize a Scala case class.

Benchmark

Dataset and types (Bird and Place) are from this repo: scala-json-benchmark. For benchmark used jmh via sbt-jmh plugin. File birds.data contains 25000 lines. Each of line is serialized json of Bird.

• Serialization benchmark is in SerializationBenchmark.scala. To run it use jmh:run -i 5 -wi 5 -f1 -t1 .*SerializationBenchmark* in SBT
• Deserialization benchmark is in DeserializationBenchmark.scala. To run it use jmh:run -i 5 -wi 5 -f1 -t1 .*DeserializationBenchmark* in SBT

Results

MacBook Pro (Retina, 13-inch, Early 2015):

• OS: macOS Sierra 10.12.6
• CPU: Intel Core i7 3.1 GHz
• Memory: 16 GB
• JVM: Java HotSpot(TM) 64-Bit Server VM, 25.181-b13

Serialization

Library	Time, ms
jsoniter	50.113
circe	140.868
spray	165.045
lift	206.377
play-json	261.774
Argonaut	306.668
json4s	385.766

Deserialization

Library	Time, ms
jsoniter	78.070
circe	130.938
spray	180.159
lift	220.421
Argonaut	319.450
play-json	357.515
json4s	467.872

Desktop machine:

• OS: Microsoft Windows 10 Pro N x64 [Version 10.0.17134.590]
• CPU: AMD Ryzen 7 2700X Eight-Core Processor i7 3.7 GHz
• Memory: DDR4-3200 GHz 16 GB
• JVM: Java HotSpot(TM) 64-Bit Server VM, 25.201-b09

Serialization

Library	Time, ms
jsoniter	41.140
circe	104.434
spray	132.525
lift	171.015
play-json	215.128
Argonaut	255.371
json4s	310.488

Deserialization

Library	Time, ms
jsoniter	68.833
circe	117.592
spray	155.961
lift	220.304
Argonaut	264.283
play-json	342.450
json4s	403.946