JSON serialization benchmarks

Methodology

Participants:

• Boon
• Circe
• DSL-JSON
• Fastjson
• Genson
• Gson
• Jackson
• Jackson-afterburner
• JSR 353 Glassfish (javax.json)
• json.org
• Json-iterator
• Json-simple
• Klaxon
• Kotlinx.serialization
• Mjson
• Minimal-json
• Moshi
• Tools-json

For testing were selected data with different structure:

• citys - A large array (29470 items) of simple objects. The compact json representation takes about 2.5 MB. Main challenge: floating point numbers
• repos.json - An array of four objects with complex structure. The compact json representation takes about 342.8 kB. Main challenge: a lot of String fields
• user.json - one object with a complex structure. The compact json representation takes about 4.2 kB. Main challenge: dates
• request.json - one object with a simple structure. The compact json representation takes about 425 B. Main challenge: Maps of Strings, small json

Serialization and deserialization were tested with POJO and Map-like structures

Build and Run

./gradlew clean && ./gradlew shadowJar && java -jar build/libs/json-benchmarks-all.jar ".*Benchmarks.*"
# or individual benchmarks
./gradlew clean && ./gradlew shadowJar && java -jar build/libs/json-benchmarks-all.jar ".(Des|S)erializationBenchmarks.(pojo|map)_(dslplatform|tools|kotlinx|jsonIterator|klaxon).*"  

Results

Checking performance of Boon

Very different results for boon deserialization into maps and pojos looked suspicious for me, so I've made another benchmark: 'deserialize into map' vs 'deserialize into map and get value of one field' and compared it with Tools and pojos.

As I expected, Boon uses lazy maps and creates actual values only when requested