Benchmarks against Serialization Systems
June 5, 2018 ยท View on GitHub
Characteristics
| Protocol Buffers (proto3) | Thrift (compact protocol) | Avro | CSV | JSON | MessagePack | |
|---|---|---|---|---|---|---|
| Schema-less? | No | No | No | No | Yes | Yes |
| Require compiling schema in advance? | Yes | Yes | No | No | No | No |
Run benchmarks
$ sbt $project/jmh:run
where $project is one of the name of sbt sub project (e.g. avroBench)
Sample data file generation
$ sbt $project/run
Schemas
The schemas used in this benchmark are under schema
Results
NOTE: This benchmark is taken place under the specific condition, results may be different under the other conditions.
Benchmark setup
- OS: Ubuntu 16.04 TLS
- CPU: Intel(R) Xeon(R) CPU E5-2680 v3 2.50GHz, 4 cores
- Memory: 14GB
- JDK 1.8.0_171, Java HotSpot(TM) 64-Bit Server VM, 25.171-b11
Average time to encode 100,000 records in milli seconds.
- 20 warming up iteration
- Average of 20 iteration
| Protocol Buffers (proto3) | Thrift (compact protocol) | Avro | CSV | JSON (with jsoniter-scala) | JSON (with circe) | MessagePack (jackson-module-msgpack) | MessagePack (msgpack4z) |
|---|---|---|---|---|---|---|---|
| 43.0 | 235.8 | 232.6 | 116.8 | 74.6 | 488.7 | 354.8 | 358.0 |
Average time to decode 100,000 binary records in milli seconds.
- 20 warming up iteration
- Average of 20 iteration
| Protocol Buffers (proto3) | Thrift (compact protocol) | Avro | CSV | JSON (with jsoniter-scala) | JSON (with circe) | MessagePack (jackson-module-msgpack) | MessagePack (msgpack4z) |
|---|---|---|---|---|---|---|---|
| 139.5 | 162.9 | 586.0 | 160.8 | 151.3 | 503.5 | 414.9 | 609.5 |
Data size of 100,000 encoded records in MB.
| Protocol Buffers (proto3) | Thrift (compact protocol) | Avro | CSV | JSON | MessagePack |
|---|---|---|---|---|---|
| 18.5 | 18.3 | 16.7 | 17.4 | 32.2 | 25.7 |