Golang Serialization Benchmark 2020 Edition

These benchmark results taken from electhomas' repo, have a quite interesting result (new serialization formats). Let's see how much serialization cost, the fastest:

Format	type	ns/op	bytes/op	allocs/op	ns/alloc
Mum	ser	97	48	0	0.00
GencodeUnsafe	ser	98	46	48	2.05
Colfer	ser	124	51	64	1.94
Bebop	ser	124	55	64	1.94
Gotiny	ser	130	48	0	0.00
GotinyNoTime	ser	136	48	0	0.00
Gogoprotobuf	ser	147	53	64	2.30
XDR2	ser	159	60	64	2.48
Msgp	ser	174	97	128	1.36
Gencode	ser	186	53	80	2.33
FlatBuffers	ser	298	95	0	0.00
Goprotobuf	ser	317	53	64	4.95
CapNProto	ser	386	96	56	6.89
CapNProto2	ser	586	96	244	2.40
Hprose2	ser	604	85	0	0.00
Ikea	ser	670	55	72	9.31
ShamatonArrayMsgpack	ser	758	50	176	4.31
Protobuf	ser	801	52	152	5.27
ShamatonMapMsgpack	ser	819	92	208	3.94
Gob	ser	834	63	48	17.38
Hprose	ser	915	85	402	2.28
GoAvro2Binary	ser	950	47	488	1.95
VmihailencoMsgpack	ser	1116	100	400	2.79
Bson	ser	1171	110	392	2.99
Binary	ser	1364	61	320	4.26
UgorjiCodecMsgpack	ser	1422	91	1312	1.08
UgorjiCodecBinc	ser	1468	95	1328	1.11
EasyJson	ser	1789	149	895	2.00
XDR	ser	1836	92	456	4.03
Json	ser	2212	150	208	10.63
JsonIter	ser	2392	139	248	9.65
GoAvro	ser	2828	47	1008	2.81
Sereal	ser	2936	132	904	3.25
GoAvro2Text	ser	2975	134	1320	2.25
SSZNoTimeNoStringNoFloatA	ser	4922	55	440	11.19

As we can see, Mum, Gencode, Colfer, Bebop, Gotiny, XDR2, MsgPack wins in terms of performace, in cost of serialization size. Let's check the deserialization performace, the fastest are:

Format	type	ns/op	bytes/op	allocs/op	ns/alloc
Bebop	des	104	55	32	3.25
XDR2	des	131	60	32	4.09
GencodeUnsafe	des	161	46	96	1.68
Colfer	des	197	50	112	1.76
Mum	des	216	48	80	2.70
Gencode	des	222	53	112	1.98
Gogoprotobuf	des	230	53	96	2.40
GotinyNoTime	des	241	48	96	2.51
FlatBuffers	des	265	95	112	2.37
Gotiny	des	267	48	112	2.38
Msgp	des	314	97	112	2.80
CapNProto	des	443	96	200	2.21
Goprotobuf	des	481	53	168	2.86
ShamatonArrayMsgpack	des	483	50	144	3.35
Hprose2	des	609	85	144	4.23
ShamatonMapMsgpack	des	738	92	144	5.12
CapNProto2	des	778	96	320	2.43
Protobuf	des	790	52	192	4.11
Ikea	des	871	55	160	5.44
Gob	des	900	63	112	8.04
GoAvro2Binary	des	1092	47	560	1.95
Hprose	des	1195	85	319	3.75
UgorjiCodecMsgpack	des	1398	91	496	2.82
Binary	des	1511	61	320	4.72
UgorjiCodecBinc	des	1587	95	656	2.42
Bson	des	1694	110	232	7.30
VmihailencoMsgpack	des	1722	100	416	4.14
EasyJson	des	1724	150	288	5.99
JsonIter	des	1874	139	264	7.10
XDR	des	2255	90	235	9.60
GoAvro2Text	des	2826	134	799	3.54
Sereal	des	3377	132	1008	3.35
Json	des	4574	149	391	11.70
GoAvro	des	6962	47	3328	2.09
SSZNoTimeNoStringNoFloatA	des	7694	55	1392	5.53

In this part, Bebop, XDR2, Gencode, Colfer, Mum, Gogoprotobuf, Gotiny, FlatBuffers, MsgPack wins this part. So if your bandwidth is unlimited, you can choose these format as your serialization format. You can access the reformatted spreadsheet here. Here's combined result of serialization and deserialization and addition of serialization size and allocation needed to deserialize.

Format	ns/op	bytes
Bebop	228	110
GencodeUnsafe	259	92
XDR2	290	120
Mum	313	96
Colfer	321	101
GotinyNoTime	377	96
Gogoprotobuf	377	106
Gotiny	397	96
Gencode	408	106
Msgp	488	194
FlatBuffers	563	190
Goprotobuf	798	106
CapNProto	829	192
Hprose2	1,213	170
ShamatonArrayMsgpack	1,241	100
CapNProto2	1,364	192
Ikea	1,541	110
ShamatonMapMsgpack	1,557	184
Protobuf	1,591	104
Gob	1,734	126
GoAvro2Binary	2,042	94
Hprose	2,110	170
UgorjiCodecMsgpack	2,820	182
VmihailencoMsgpack	2,838	200
Bson	2,865	220
Binary	2,875	122
UgorjiCodecBinc	3,055	190
EasyJson	3,513	299
XDR	4,091	182
JsonIter	4,266	278
GoAvro2Text	5,801	268
Sereal	6,313	264
Json	6,786	299
GoAvro	9,790	94
SSZNoTimeNoStringNoFloatA	12,616	110

Here are the links for high performing libraries used:

• Bebop codegen from .bop (need to create schema like protobuf)
• Gencode codegen from .schema (similar to go syntax)
• XDR2 codegen version (other libs are using reflection/automatic), unmaintained
• Mum manual (must create the method to serialize and deserialize yourself), new name is enkodo
• Colfer codegen from .colf (similar to go syntax)
• Gogoprotobuf codegen from .proto
• Gotiny codegen, not recommended for production
• MsgPack codegen version (other libs are using reflection/automatic), from .go source file using go:generate, supported in lots of language
• FlatBuffers codegen, can be used in gRPC

What are the difference between codegen, automatic, manual?

• codegen means there are step to generate golang function by writing a schema definition file then run a program to convert that file to specific programming language implementation, sometimes using other format (so you cannot add custom tag to generated .go file), sometimes using golang struct with tag (like codegen version of msgpack above you need to label each property with `msg:"foo"`, so you can add your own custom tag on the struct property, eg. `json:"bar,omitempty" form:"baz" bson:"blabla"`)
• manual means you must write the serialization and deserialization yourself for each property of the struct (the library only the helper), this allows highly flexible system, so for example you read 1 byte first, then if the value is 1 then you read a string, if 2 you read int32, and so on. This can also be useful to parse network packet if having the same endian.
• automatic means you don't need to write any schema, it uses reflection so should be slower than codegen version.