Fury Benchmarks

Fury benchmarks with:

• jackson databind
• Microstream

Benchmark Setup

JMH config: -f 3 -wi 5 -i 5 -t 1 -w 2s -r 2s -rf csv OS: macos monterey Hardware: 2.6 GHz 6-Core Intel Core i7

Benchmark Results

Fury vs Jackson

Fury is 3x smaller for serialized binary size at most:

furyMediaContentBytes size 336
furyStructBytes size 90
jacksonMediaContentBytes size 664
jacksonStructBytes size 278

The benchmark results here are for reference only. Jackson is a json format, fury is a binary format, which are suitable for different scenarios.

• Fury is 32.1x faster than jackson for Struct serialization
• Fury is 45x faster than jackson for Struct deserialization
• Fury is 8.8x faster than jackson for MediaContent serialization
• Fury is 11.8x faster than jackson for MediaContent deserialization

Benchmark                                         Mode  Cnt         Score         Error  Units
JacksonBenchmark.furyDeserializeMediaContent     thrpt   15   2783918.470 ±   68075.351  ops/s
JacksonBenchmark.furyDeserializeStruct           thrpt   15  13686679.435 ± 1440816.885  ops/s
JacksonBenchmark.furySerializeMediaContent       thrpt   15   3410893.820 ±  265913.345  ops/s
JacksonBenchmark.furySerializeStruct             thrpt   15  17004519.876 ±  463107.804  ops/s
JacksonBenchmark.jacksonDeserializeMediaContent  thrpt   15    235938.122 ±   23054.241  ops/s
JacksonBenchmark.jacksonDeserializeStruct        thrpt   15    302600.964 ±   19586.158  ops/s
JacksonBenchmark.jacksonSerializeMediaContent    thrpt   15    386182.017 ±   38637.729  ops/s
JacksonBenchmark.jacksonSerializeStruct          thrpt   15    529044.347 ±   31945.826  ops/s

Compute speedup:

import pandas as pd
df = pd.read_csv("jmh-result.csv")
s = df["Score"]
s[:4].to_numpy()/s[4:].to_numpy()

Fury vs Microstream

Fury is 5x smaller for serialized binary size at most:

furyMediaContentBytes size 336
furyStructBytes size 90
microstream mediaContentBytes size 1527
microstream structBytes size 120

Fury is :

Benchmark                                                 Mode  Cnt         Score          Error  Units
MicrostreamBenchmark.furyDeserializeMediaContent         thrpt    3   2082751.515 ±  2638613.695  ops/s
MicrostreamBenchmark.furyDeserializeStruct               thrpt    3  10334292.992 ± 10542478.626  ops/s
MicrostreamBenchmark.furySerializeMediaContent           thrpt    3   2898943.216 ±   946978.784  ops/s
MicrostreamBenchmark.furySerializeStruct                 thrpt    3  13980885.316 ± 13424528.901  ops/s
MicrostreamBenchmark.microstreamDeserializeMediaContent  thrpt    3     76354.367 ±   231329.651  ops/s
MicrostreamBenchmark.microstreamSerializeMediaContent    thrpt    3     66316.198 ±   729572.587  ops/s
MicrostreamBenchmark.microstreamSerializeStruct          thrpt    3    233374.717 ±   732378.317  ops/s

Fury vs Kryo in JDK17:

• Fury is 7.4x faster than kryo for MediaContent serialization
• Fury is 12.4x faster than kryo for Struct serialization
• Fury is 4.8x faster than kryo for Media deserialization
• Fury is 9.8x faster than kryo for Struct deserialization

Benchmark                                   Mode  Cnt         Score         Error  Units
KryoBenchmark.furyDeserializeMediaContent  thrpt    9   2468765.648 ±  314266.100  ops/s
KryoBenchmark.furyDeserializeStruct        thrpt    9  20031536.933 ± 1893069.982  ops/s
KryoBenchmark.furySerializeMediaContent    thrpt    9   4041446.580 ±  402177.778  ops/s
KryoBenchmark.furySerializeStruct          thrpt    9  25475238.500 ±  920132.134  ops/s
KryoBenchmark.kryoDeserializeMediaContent  thrpt    9    545891.524 ±   19581.938  ops/s
KryoBenchmark.kryoDeserializeStruct        thrpt    9   2496417.446 ±   76754.451  ops/s
KryoBenchmark.kryoSerializeMediaContent    thrpt    9    478720.501 ±   69891.951  ops/s
KryoBenchmark.kryoSerializeStruct          thrpt    9   1942266.459 ±   63863.902  ops/s

Fury vs Avaje

For serialized data size, fury is half of avaje:

furyMediaContentBytes size 336
furyStructBytes size 70
avaje mediaContentBytes size 586
avaje structBytes size 149

• Fury is 5.6x faster than Avaje for Media serialization
• Fury is 29x faster than Avaje for Struct serialization
• Fury is 6x faster than Avaje for MediaContent deserialization
• Fury is 24x faster than Avaje for Struct deserialization

Benchmark                                     Mode  Cnt         Score         Error  Units
AvajeBenchmark.avajeDeserializeMediaContent  thrpt    3    435874.584 ±   54036.329  ops/s
AvajeBenchmark.avajeDeserializeStruct        thrpt    3    906966.403 ± 1349025.028  ops/s
AvajeBenchmark.avajeSerializeMediaContent    thrpt    3    669107.596 ±   11450.125  ops/s
AvajeBenchmark.avajeSerializeStruct          thrpt    3    990601.474 ±   61731.364  ops/s
AvajeBenchmark.furyDeserializeMediaContent   thrpt    3   2607302.095 ± 2653734.646  ops/s
AvajeBenchmark.furyDeserializeStruct         thrpt    3  21689653.040 ± 2702955.367  ops/s
AvajeBenchmark.furySerializeMediaContent     thrpt    3   3775725.625 ±  273398.873  ops/s
AvajeBenchmark.furySerializeStruct           thrpt    3  28858705.061 ± 4330421.563  ops/s