Awesome

go-serialization-benchmarks

In this benchmarks:

• All serializers use the same data, it is generated once and then used by everyone.
• Each serializer is described with a set of features.
• One serializer can have several benchmark results. For example, MUS can have mus+raw and mus+unsafe results. The first one indicates that the results were obtained with the raw feature enabled, the last one - that unsafe feature was used.
• Unmarshalled data are compared to the original data.

Benchmarks

Fastest Safe

Fastest Unsafe

All

, where iterations count, ns/op, B/op, allocs/op are standard go test -bench=. output and B/size - determines how many bytes were used on average by the serializer to encode Data.

Features

• benc: binary, manual, raw, reuse, unsafestr
• bebop200sc: binary, codegen, notunsafe, reuse
• gob: binary, int
• json: int, reflect, text
• mus: binary, int, manual, notunsafe, raw, reuse, unsafe, unsafestr, varint
• protobuf: binary, codegen, raw, varint

List of Features

Each feature describes a serializer:

• reflect - it uses reflection.
• codegen - it uses code generation.
• manual - there are only serialization primitives, so you have to use them manually.
• text - it has text serialization format.
• binary - it has binary serialization format.
• varint - it supports varint encoding.
• raw - it supports raw encoding.
• int - it supports int type.
• native - when a format is implemented with a set of primitives and native data is used.

Features that must be in the result name when used:

• reuse - it supports buffer reuse.
• unsafe - it supports unsafe code.
• unsafestr - it supports only unsafe string serialization.
• unsafeunm - it supports only unsafe unmarshalling.
• notunsafe - it uses the unsafe code for all types except string and copies data on unmarshal.
• fixbuf - if a fixed buffer is used.

This list can be expanded.

Data

Randomly generated data has the following form:

type Data struct {
  Str     string
  Bool    bool
  Int32   int32
  Float64 float64
  Time    time.Time
}

It does not have an int type because many serializers do not support it.

Run Benchmarks

From the benchmarks/ folder:

go test -bench=.

We can also filter the serializers by features, for example:

go test -bench=. -- -f binary,manual

In this case we will see the results of serializers that have both binary and manual features. To run benchmarks for one particular case just name it, for example:

go test -bench=BenchmarkSerializers/mus

or

go test -bench=/.+reuse

to see the results obtained using the reuse feature.

Generate README.md

From the benchmarks/ folder:

go generate ./...

Recomendation

When creating README.md on a laptop, please make sure that it is connected to a charger and the fan is running at full speed.

Contribution

First of all, you need to create a new package for your serializer if it doesn't already exist. Then:

1. Implement serializer.Serializer[serializer.Data] interface. Doing this you can use:
   • serializer.NewResultName(...) - which creates a correct result name.
   • serializer.BufSize - defines the recommended buffer size for reuse.
2. If you use own Data make shure it implements EqualTo(data Data) error method, also add func ToYourData(data serializer.Data) (d Data) function (an example can be found in bbebop200sc/serializers.go).
3. Define

   var Serializers = []serializer.Serializer[serializer.Data]{Serializer{}}
   

   variable. Note that it can contain several serializers that produce different results.
4. Create PR.

If you want to run benchmarks from your own project, there is the benchser.BenchmarkSerializer(...) function.