Awesome
Fast Binary Encoding (FBE)
Fast Binary Encoding allows to describe any domain models, business objects, complex data structures, client/server requests & responses and generate native code for different programming languages and platforms.
Fast Binary Encoding documentation<br/> Fast Binary Encoding downloads<br/> Fast Binary Encoding specification
Performance comparison to other protocols can be found here:
Protocol | Message size | Serialization time | Deserialization time |
---|---|---|---|
Cap'n'Proto | 208 bytes | 558 ns | 359 ns |
FastBinaryEncoding | 234 bytes | 66 ns | 82 ns |
FlatBuffers | 280 bytes | 830 ns | 290 ns |
Protobuf | 120 bytes | 628 ns | 759 ns |
JSON | 301 bytes | 740 ns | 500 ns |
Typical usage workflow is the following:
- Create domain model using base types, enums, flags and structs
- Generate domain model for any supported programming languages (C++, C#, Go, Java, JavaScript, Kotlin, Python, Ruby, Swift)
- Build domain model library
- Serialize/Deserialize objects from the domain model in unified, fast and compact FastBinaryEncoding (FBE) format
- JSON convert objects from the domain model in order to use them in Web API
- Implement Sender/Receiver interfaces to create a communication protocol
Sample projects:
- C++ sample
- C# sample
- Go sample
- Java sample
- JavaScript sample
- Kotlin sample
- Python sample
- Ruby sample
- Swift sample
Contents
- Features
- Requirements
- How to build?
- Create domain model
- Generate domain model
- Build domain model
- FBE serialization
- FBE final serialization
- JSON serialization
- Packages and import
- Struct keys
- Struct numeration
- Struct inheritance
- Versioning
- Sender/Receiver protocol
- Performance benchmarks
Features
- Cross platform (Linux, MacOS, Windows)
- Generators for C++, C#, Go, Java, JavaScript, Kotlin, Python, Ruby, Swift
- Fast binary encoding format
- Supported base types (byte, bool, char, wchar, int8, int16, int32, int64, float, double)
- Supported complex types (bytes, decimal, string, timestamp, uuid)
- Supported collections (array, vector, list, map, hash)
- Supported nullable optional types, enums, flags and structs
- Serialization/Deserialization to/from binary format
- Serialization/Deserialization to/from JSON
- Sender/Receiver interfaces for communication protocols
- Versioning solution
- Excellent performance
Requirements
Optional:
How to build?
Linux: install required packages
sudo apt-get install -y binutils-dev uuid-dev flex bison
MacOS: install required packages
brew install flex bison
Windows: install required packages
choco install winflexbison3
Install gil (git links) tool
pip3 install gil
Setup repository
git clone https://github.com/chronoxor/FastBinaryEncoding.git
cd FastBinaryEncoding
gil update
Linux
cd build
./unix.sh
MacOS
cd build
./unix.sh
Windows (Cygwin)
cd build
unix.bat
Windows (MSYS2)
cd build
unix.bat
Windows (MinGW)
cd build
mingw.bat
Windows (Visual Studio)
cd build
vs.bat
Create domain model
To use Fast Binary Encoding you should provide a domain model (aka business objects). A domain model is a set of enums, flags and structures that relate to each other and might be aggregated in some hierarchy.
Fast Binary Encoding (FBE) format specification
There is a sample domain model which describes Account-Balance-Orders relation of some abstract trading platform:
// Package declaration
package proto
// Domain declaration
domain com.chronoxor
// Order side declaration
enum OrderSide : byte
{
buy;
sell;
}
// Order type declaration
enum OrderType : byte
{
market;
limit;
stop;
}
// Order declaration
struct Order
{
[key] int32 uid;
string symbol;
OrderSide side;
OrderType type;
double price = 0.0;
double volume = 0.0;
}
// Account balance declaration
struct Balance
{
[key] string currency;
double amount = 0.0;
}
// Account state declaration
flags State : byte
{
unknown = 0x00;
invalid = 0x01;
initialized = 0x02;
calculated = 0x04;
broken = 0x08;
good = initialized | calculated;
bad = unknown | invalid | broken;
}
// Account declaration
struct Account
{
[key] int32 uid;
string name;
State state = State.initialized | State.bad;
Balance wallet;
Balance? asset;
Order[] orders;
}
Generate domain model
The next step is a domain model compilation using 'fbec' compiler which will create a generated code for required programming language.
The following command will create a C++ generated code:
fbec --c++ --input=proto.fbe --output=.
All possible options for the 'fbec' compiler are the following:
Usage: fbec [options]
Options:
--version show program's version number and exit
-h, --help show this help message and exit
-h HELP, --help=HELP Show help
-i INPUT, --input=INPUT
Input path
-o OUTPUT, --output=OUTPUT
Output path
-q, --quiet Launch in quiet mode. No progress will be shown!
-n INDENT, --indent=INDENT
Format indent. Default: 0
-t, --tabs Format with tabs. Default: off
--cpp Generate C++ code
--cpp-logging Generate C++ logging code
--csharp Generate C# code
--go Generate Go code
--java Generate Java code
--javascript Generate JavaScript code
--kotlin Generate Kotlin code
--python Generate Python code
--ruby Generate Ruby code
--swift Generate Swift code
--final Generate Final serialization code
--json Generate JSON serialization code
--proto Generate Sender/Receiver protocol code
Build domain model
Generated domain model is represented with source code for the particular language. Just add it to your project and build it. There are several issues and dependencies that should be mentioned:
C++
- C++ standard is limited to C++17 in order to have the implementation of std::optional;
- C++ has no native support for decimal type. Currently decimal type is emulated with a double type. FBE does not use GMPLib because of heavy dependency in generated source code;
- C++ formatting is supported with {fmt} library of version
started from 9.0.0. Required include headers are
<fmt/format.h>
and<fmt/ostream.h>
; - JSON serialization is implemented using RapidJSON library;
C#
- JSON serialization is implemented using Json.NET library. Therefore it should be imported using NuGet;
- Fast JSON serialization libraty is also available - Utf8Json . If you want to try it, you should import is with NuGet and build domain model with 'UTF8JSON' definition;
Go
- Assert testing is based on stretchr/testify package (
go get github.com/stretchr/testify
); - JSON serialization is based on jsoniter package (
go get github.com/json-iterator/go
); - Decimal type is based on shopspring/decimal package (
go get github.com/shopspring/decimal
); - UUID type is based on google/uuid package (
go get github.com/google/uuid
);
Java
- JSON serialization is implemented using Gson package. Therefore it should be imported using Maven;
JavaScript
- JavaScript domain model is implemented using ECMAScript 6 (classes, etc.);
- JSON serialization of set, map and hash types is limited to key with string type;
Kotlin
- Starting from the version 1.3 Kotlin supports unsigned integer numbers (UByte, UShort, UInt, ULong). This gives ability to represent FBE domain model more accurately than Java language does;
- JSON serialization is implemented using Gson package. Therefore it should be imported using Maven;
Python
- Python 3.7 is required because of time.time_ns();
Ruby
- Some Ruby dependencies should be installed from Gems:
gem install json
gem install uuidtools
Swift
- Swift domain model is implemented using uses SwiftPM as its build tool;
- JSON serialization is implemented using Codable protocol;
- JSON serialization of set, map and hash types is limited to key with string type.
FBE serialization
Fast Binary Encoding (FBE) is a fast and compact binary format of representing single domain models in different programming languages and platforms. Also FBE format solves protocol versioning problem.
Follow the steps below in order to serialize any domain object:
- Create a new domain object and fill its fields and collections (proto::Account account);
- Create a domain model with a write buffer (FBE::proto::AccountModelFBE::WriteBuffer writer)
- Serialize the domain object into the domain model buffer (writer.serialize(account))
- (Optional) Verify the domain object in the domain model buffer (assert(writer.verify()))
- Access the domain model buffer to store or send data (writer.buffer())
Follow the steps below in order to deserialize any domain object:
- Create a domain model with a read buffer (FBE::proto::AccountModelFBE::ReadBuffer reader)
- Attach a source buffer to the domain model (reader.attach(writer.buffer()))
- (Optional) Verify the domain object in the domain model buffer (assert(reader.verify()))
- Deserialize the domain object from the domain model buffer (reader.deserialize(account))
Here is an exmple of FBE serialization in C++ language:
#include "../proto/proto_models.h"
#include <iostream>
int main(int argc, char** argv)
{
// Create a new account with some orders
proto::Account account = { 1, "Test", proto::State::good, { "USD", 1000.0 }, std::make_optional<proto::Balance>({ "EUR", 100.0 }), {} };
account.orders.emplace_back(1, "EURUSD", proto::OrderSide::buy, proto::OrderType::market, 1.23456, 1000.0);
account.orders.emplace_back(2, "EURUSD", proto::OrderSide::sell, proto::OrderType::limit, 1.0, 100.0);
account.orders.emplace_back(3, "EURUSD", proto::OrderSide::buy, proto::OrderType::stop, 1.5, 10.0);
// Serialize the account to the FBE stream
FBE::proto::AccountModel<FBE::WriteBuffer> writer;
writer.serialize(account);
assert(writer.verify());
// Show the serialized FBE size
std::cout << "FBE size: " << writer.buffer().size() << std::endl;
// Deserialize the account from the FBE stream
FBE::proto::AccountModel<FBE::ReadBuffer> reader;
reader.attach(writer.buffer());
assert(reader.verify());
reader.deserialize(account);
// Show account content
std::cout << std::endl;
std::cout << account;
return 0;
}
Output is the following:
FBE size: 252
Account(
uid=1,
name="Test",
state=initialized|calculated|good,
wallet=Balance(currency="USD",amount=1000),
asset=Balance(currency="EUR",amount=100),
orders=[3][
Order(uid=1,symbol="EURUSD",side=buy,type=market,price=1.23456,volume=1000),
Order(uid=2,symbol="EURUSD",side=sell,type=limit,price=1,volume=100),
Order(uid=3,symbol="EURUSD",side=buy,type=stop,price=1.5,volume=10)
]
)
FBE final serialization
It is possible to achieve more serialization speed if your protocol is mature enough so you can fix its final version and disable versioning which requires extra size and time to process.
Protocol | Message size | Serialization time | Deserialization time | Verify time |
---|---|---|---|---|
FBE | 252 bytes | 88 ns | 98 ns | 33 ns |
FBE final | 152 bytes | 57 ns | 81 ns | 28 ns |
Final domain model can be compiled with --final flag. As the result additional final models will be available for serialization.
Follow the steps below in order to serialize any domain object in final format:
- Create a new domain object and fill its fields and collections (proto::Account account);
- Create a domain final model with a write buffer (FBE::proto::AccountFinalModelFBE::WriteBuffer writer)
- Serialize the domain object into the domain model buffer (writer.serialize(account))
- (Optional) Verify the domain object in the domain model buffer (assert(writer.verify()))
- Access the domain model buffer to store or send data (writer.buffer())
Follow the steps below in order to deserialize any domain object:
- Create a domain final model with a read buffer (FBE::proto::AccountFinalModelFBE::ReadBuffer reader)
- Attach a source buffer to the domain final model (reader.attach(writer.buffer()))
- (Optional) Verify the domain object in the domain model buffer (assert(reader.verify()))
- Deserialize the domain object from the domain model buffer (reader.deserialize(account))
Here is an exmple of FBE final serialization in C++ language:
#include "../proto/proto_models.h"
#include <iostream>
int main(int argc, char** argv)
{
// Create a new account with some orders
proto::Account account = { 1, "Test", proto::State::good, { "USD", 1000.0 }, std::make_optional<proto::Balance>({ "EUR", 100.0 }), {} };
account.orders.emplace_back(1, "EURUSD", proto::OrderSide::buy, proto::OrderType::market, 1.23456, 1000.0);
account.orders.emplace_back(2, "EURUSD", proto::OrderSide::sell, proto::OrderType::limit, 1.0, 100.0);
account.orders.emplace_back(3, "EURUSD", proto::OrderSide::buy, proto::OrderType::stop, 1.5, 10.0);
// Serialize the account to the FBE stream
FBE::proto::AccountFinalModel<FBE::WriteBuffer> writer;
writer.serialize(account);
assert(writer.verify());
// Show the serialized FBE size
std::cout << "FBE final size: " << writer.buffer().size() << std::endl;
// Deserialize the account from the FBE stream
FBE::proto::AccountFinalModel<FBE::ReadBuffer> reader;
reader.attach(writer.buffer());
assert(reader.verify());
reader.deserialize(account);
// Show account content
std::cout << std::endl;
std::cout << account;
return 0;
}
Output is the following:
FBE final size: 152
Account(
uid=1,
name="Test",
state=initialized|calculated|good,
wallet=Balance(currency="USD",amount=1000),
asset=Balance(currency="EUR",amount=100),
orders=[3][
Order(uid=1,symbol="EURUSD",side=buy,type=market,price=1.23456,volume=1000),
Order(uid=2,symbol="EURUSD",side=sell,type=limit,price=1,volume=100),
Order(uid=3,symbol="EURUSD",side=buy,type=stop,price=1.5,volume=10)
]
)
JSON serialization
If the domain model compiled with --json flag, then JSON serialization code will be generated in all domain objects. As the result each domain object can be serialized/deserialized into/from JSON format.
Please note that some programming languages have native JSON support (JavaScript, Python). Other languages requires third-party library to get work with JSON:
- C++ requires RapidJSON library
- C# requires Json.NET package or more faster Utf8Json package
- Go requires jsoniter package
- Java and Kotlin requires Gson package
- Ruby requires json gem
Here is an exmple of JSON serialization in C++ language:
#include "../proto/proto.h"
#include <iostream>
int main(int argc, char** argv)
{
// Create a new account with some orders
proto::Account account = { 1, "Test", proto::State::good, { "USD", 1000.0 }, std::make_optional<proto::Balance>({ "EUR", 100.0 }), {} };
account.orders.emplace_back(1, "EURUSD", proto::OrderSide::buy, proto::OrderType::market, 1.23456, 1000.0);
account.orders.emplace_back(2, "EURUSD", proto::OrderSide::sell, proto::OrderType::limit, 1.0, 100.0);
account.orders.emplace_back(3, "EURUSD", proto::OrderSide::buy, proto::OrderType::stop, 1.5, 10.0);
// Serialize the account to the JSON stream
rapidjson::StringBuffer buffer;
rapidjson::Writer<rapidjson::StringBuffer> writer(buffer);
FBE::JSON::to_json(writer, account);
// Show the serialized JSON and its size
std::cout << "JSON: " << buffer.GetString() << std::endl;
std::cout << "JSON size: " << buffer.GetSize() << std::endl;
// Parse the JSON document
rapidjson::Document json;
json.Parse(buffer.GetString());
// Deserialize the account from the JSON stream
FBE::JSON::from_json(json, account);
// Show account content
std::cout << std::endl;
std::cout << account;
return 0;
}
Output is the following:
JSON: {
"uid":1,
"name":
"Test",
"state":6,
"wallet":{"currency":"USD","amount":1000.0},
"asset":{"currency":"EUR","amount":100.0},
"orders":[
{"uid":1,"symbol":"EURUSD","side":0,"type":0,"price":1.23456,"volume":1000.0},
{"uid":2,"symbol":"EURUSD","side":1,"type":1,"price":1.0,"volume":100.0},
{"uid":3,"symbol":"EURUSD","side":0,"type":2,"price":1.5,"volume":10.0}
]
}
JSON size: 353
Account(
uid=1,
name="Test",
state=initialized|calculated|good,
wallet=Balance(currency="USD",amount=1000),
asset=Balance(currency="EUR",amount=100),
orders=[3][
Order(uid=1,symbol="EURUSD",side=buy,type=market,price=1.23456,volume=1000),
Order(uid=2,symbol="EURUSD",side=sell,type=limit,price=1,volume=100),
Order(uid=3,symbol="EURUSD",side=buy,type=stop,price=1.5,volume=10)
]
)
Packages and import
Packages are declared with package name and structs offset (optional). Offset will be add to incremented structure type if is was not provided explicit.
Here is an example of the simple package declaration:
// Package declaration. Offset is 0.
package proto
// Struct type number is 1 (proto offset 0 + 1)
struct Struct1
{
...
}
// Struct type number is 2 (proto offset 0 + 2)
struct Struct2
{
...
}
One package can be imported into another and all enums, flags and structs can be reused in the current package. Package offset is used here to avoid structs types intersection:
// Package declaration. Offset is 10.
package protoex offset 10
// Package import
import proto
// Struct type number is 11 (protoex offset 10 + 1)
struct Struct11
{
// Struct1 is reused form the imported package
proto.Struct1 s1;
...
}
// Struct type number is 12 (protoex offset 10 + 2)
struct Struct12
{
...
}
Multiple package import is possible as well:
// Package declaration. Offset is 100.
package test offset 100
// Package import
import proto
import protoex
...
Package import is implemented using:
- #include "..." directive in C++
- Namespaces in C#
- Packages in Go
- Packages in Java and Kotlin
- Modules in JavaScript
- Modules in Python
- Ruby require statement
Struct keys
Some of struct fileds (one or many) can be marked with '[key]' attribute. As the result corresponding compare operators will be generated which allow to compare two instances of the struct (equality, ordering, hashing) by marked fields. This ability allows to use the struct as a key in associative map and hash containers.
Example below demonstrates the usage of '[key]' attribute:
struct MyKeyStruct
{
[key] int32 uid;
[key] stirng login;
string name;
string address;
}
After code generation for C++ language the following comparable class will be generated:
struct MyKeyStruct
{
int32_t uid;
::sample::stirng login;
std::string name;
std::string address;
...
bool operator==(const MyKeyStruct& other) const noexcept
{
return (
(uid == other.uid)
&& (login == other.login)
);
}
bool operator!=(const MyKeyStruct& other) const noexcept { return !operator==(other); }
bool operator<(const MyKeyStruct& other) const noexcept
{
if (uid < other.uid)
return true;
if (other.uid < uid)
return false;
if (login < other.login)
return true;
if (other.login < login)
return false;
return false;
}
bool operator<=(const MyKeyStruct& other) const noexcept { return operator<(other) || operator==(other); }
bool operator>(const MyKeyStruct& other) const noexcept { return !operator<=(other); }
bool operator>=(const MyKeyStruct& other) const noexcept { return !operator<(other); }
...
};
Struct numeration
Struct type numbers are automatically increased until you provide it manually. There are two possibilities:
- Shift the current struct type number using '(+X)' suffix. As the result all new structs will have incremented type.
- Force set struct type number using '(X)' of '(base)' suffix. It will affect only one struct.
Example below demonstrates the idea:
// Package declaration. Offset is 0.
package proto
// Struct type number is 1 (implicit declared)
struct Struct1
{
...
}
// Struct type number is 2 (implicit declared)
struct Struct2
{
...
}
// Struct type number is 10 (explicit declared, shifted to 10)
struct Struct10(+10)
{
...
}
// Struct type number is 11 (implicit declared)
struct Struct11
{
...
}
// Struct type number is 100 (explicit declared, forced to 100)
struct Struct100(100)
{
...
}
// Struct type number is 12 (implicit declared)
struct Struct12
{
...
}
Struct inheritance
Structs can be inherited from another struct. In this case all fields from the base struct will be present in a child one.
package proto
// Struct type number is 1
struct StructBase
{
bool f1;
int8 f2;
}
// Struct type number is 2
struct StructChild : StructBase
{
// bool f1 - will be inherited from StructBase
// int8 f2 - will be inherited from StructBase
int16 f3;
int32 f4;
}
Also it is possible to reuse the base struct type number in a child one using '= base' operator. It is useful when you extend the struct from third-party imported package:
// Package declaration. Offset is 10.
package protoex offset 10
// Package import
import proto
// Struct type number is 1
struct StructChild(base) : proto.StructBase
{
// bool f1 - will be inherited from proto.StructBase
// int8 f2 - will be inherited from proto.StructBase
int16 f3;
int32 f4;
}
Versioning
Versioning is simple with Fast Binary Encoding.
Assume you have an original protocol:
package proto
enum MyEnum
{
value1;
value2;
}
flags MyFlags
{
none = 0x00;
flag1 = 0x01;
flag2 = 0x02;
flag3 = 0x04;
}
struct MyStruct
{
bool field1;
byte field2;
char field3;
}
You need to extend it with new enum, flag and struct values. Just add required values to the end of the corresponding declarations:
package proto
enum MyEnum
{
value1;
value2;
value3; // New value
value4; // New value
}
flags MyFlags
{
none = 0x00;
flag1 = 0x01;
flag2 = 0x02;
flag3 = 0x04;
flag4 = 0x08; // New value
flag5 = 0x10; // New value
}
struct MyStruct
{
bool field1;
byte field2;
char field3;
int32 field4; // New field (default value is 0)
int64 field5 = 123456; // New field (default value is 123456)
}
Now you can serialize and deserialize structs in different combinations:
- Serialize old, deserialize old - nothing will be lost (best case);
- Serialize old, deserialize new - all old fields will be deserialized, all new fields will be initialized with 0 or default values according to definition;
- Serialize new, deserialize old - all old fields will be deserialized, all new fields will be discarded;
- Serialize new, deserialize new - nothing will be lost (best case);
Versioning of the third-party protocol
If you are not able to modify some third-party protocol, you can still have a solution of extending it. Just create a new protocol and import third-party one into it. Then extend structs with inheritance:
package protoex
import proto
struct MyStructEx(base) : proto.MyStruct
{
int32 field4; // New field (default value is 0)
int64 field5 = 123456; // New field (default value is 123456)
}
Sender/Receiver protocol
If the domain model compiled with --sender flag, then Sender/Receiver protocol code will be generated.
Sender interface contains 'send(struct)' methods for all domain model structs. Also it has abstract 'onSend(data, size)' method which should be implemented to send serialized data to a socket, pipe, etc.
Receiver interface contains 'onReceive(struct)' handlers for all domain model structs. Also it has public 'onReceive(type, data, size)' method which should be used to feed the Receiver with received data from a socket, pipe, etc.
Here is an exmple of using Sender/Receiver communication protocol in C++ language:
#include "../proto/proto_protocol.h"
#include <iostream>
class MySender : public FBE::proto::Sender<FBE::WriteBuffer>
{
protected:
size_t onSend(const void* data, size_t size) override
{
// Send nothing...
return 0;
}
void onSendLog(const std::string& message) const override
{
std::cout << "onSend: " << message << std::endl;
}
};
class MyReceiver : public FBE::proto::Receiver<FBE::WriteBuffer>
{
protected:
void onReceive(const proto::Order& value) override {}
void onReceive(const proto::Balance& value) override {}
void onReceive(const proto::Account& value) override {}
void onReceiveLog(const std::string& message) const override
{
std::cout << "onReceive: " << message << std::endl;
}
};
int main(int argc, char** argv)
{
MySender sender;
// Enable logging
sender.logging(true);
// Create and send a new order
proto::Order order = { 1, "EURUSD", proto::OrderSide::buy, proto::OrderType::market, 1.23456, 1000.0 };
sender.send(order);
// Create and send a new balance wallet
proto::Balance balance = { "USD", 1000.0 };
sender.send(balance);
// Create and send a new account with some orders
proto::Account account = { 1, "Test", proto::State::good, { "USD", 1000.0 }, std::make_optional<proto::Balance>({ "EUR", 100.0 }), {} };
account.orders.emplace_back(1, "EURUSD", proto::OrderSide::buy, proto::OrderType::market, 1.23456, 1000.0);
account.orders.emplace_back(2, "EURUSD", proto::OrderSide::sell, proto::OrderType::limit, 1.0, 100.0);
account.orders.emplace_back(3, "EURUSD", proto::OrderSide::buy, proto::OrderType::stop, 1.5, 10.0);
sender.send(account);
MyReceiver receiver;
// Enable logging
receiver.logging(true);
// Receive all data from the sender
receiver.receive(sender.buffer().data(), sender.buffer().size());
return 0;
}
Output is the following:
onSend: Order(uid=1,symbol="EURUSD",side=buy,type=market,price=1.23456,volume=1000)
onSend: Balance(currency="USD",amount=1000)
onSend: Account(uid=1,name="Test",state=initialized|calculated|good,wallet=Balance(currency="USD",amount=1000),asset=Balance(currency="EUR",amount=100),orders=[3][Order(uid=1,symbol="EURUSD",side=buy,type=market,price=1.23456,volume=1000),Order(uid=2,symbol="EURUSD",side=sell,type=limit,price=1,volume=100),Order(uid=3,symbol="EURUSD",side=buy,type=stop,price=1.5,volume=10)])
onReceive: Order(uid=1,symbol="EURUSD",side=buy,type=market,price=1.23456,volume=1000)
onReceive: Balance(currency="USD",amount=1000)
onReceive: Account(uid=1,name="Test",state=initialized|calculated|good,wallet=Balance(currency="USD",amount=1000),asset=Balance(currency="EUR",amount=100),orders=[3][Order(uid=1,symbol="EURUSD",side=buy,type=market,price=1.23456,volume=1000),Order(uid=2,symbol="EURUSD",side=sell,type=limit,price=1,volume=100),Order(uid=3,symbol="EURUSD",side=buy,type=stop,price=1.5,volume=10)])
Performance benchmarks
All benchmarks use the same domain model to create a single account with three orders:
Account account = { 1, "Test", State::good, { "USD", 1000.0 }, std::make_optional<Balance>({ "EUR", 100.0 }), {} };
account.orders.emplace_back(1, "EURUSD", OrderSide::buy, OrderType::market, 1.23456, 1000.0);
account.orders.emplace_back(2, "EURUSD", OrderSide::sell, OrderType::limit, 1.0, 100.0);
account.orders.emplace_back(3, "EURUSD", OrderSide::buy, OrderType::stop, 1.5, 10.0);
- C++ benchmarks results were taken using CppBenchmark library
- C# benchmarks results were taken using BenchmarkDotNet library
- Go benchmarks results were taken using testing package
- Java benchmarks results were taken using JMH library
- JavaScript benchmarks results were taken using Benchmark.js library
- Kotlin benchmarks results were taken using JMH library
- Python benchmarks results were taken using timeit module
- Ruby benchmarks results were taken using Benchmark module
- Swift benchmarks results were taken using XCTest framework
Benchmark 1: Serialization
Serialization benchmark C++ code:
BENCHMARK_FIXTURE(SerializationFixture, "Serialize")
{
// Reset FBE stream
writer.reset();
// Serialize the account to the FBE stream
writer.serialize(account);
}
Serialization benchmark results:
Language & Platform | Message size | Serialization rate | Serialization time |
---|---|---|---|
C++ Win64 | 252 bytes | 10 416 667 ops/s | 96 ns |
C++ Win64 (Final) | 152 bytes | 16 129 032 ops/s | 62 ns |
C++ Win64 (JSON) | 353 bytes | 926 784 ops/s | 1 079 ns |
C# Win64 | 252 bytes | 1 432 665 ops/s | 698 ns |
C# Win64 (Final) | 152 bytes | 1 597 444 ops/s | 626 ns |
C# Win64 (JSON) | 341 bytes | 434 783 ops/s | 2 300 ns |
Go Win64 | 252 bytes | 2 739 726 ops/s | 365 ns |
Go Win64 (Final) | 152 bytes | 2 949 852 ops/s | 339 ns |
Go Win64 (JSON) | 341 bytes | 258 732 ops/s | 3 865 ns |
Java Win64 | 252 bytes | 4 247 162 ops/s | 236 ns |
Java Win64 (Final) | 152 bytes | 4 883 205 ops/s | 205 ns |
Java Win64 (JSON) | 353 bytes | 213 983 ops/s | 4 673 ns |
JavaScript Win64 | 252 bytes | 93 416 ops/s | 10 705 ns |
JavaScript Win64 (Final) | 152 bytes | 112 665 ops/s | 8 876 ns |
JavaScript Win64 (JSON) | 341 bytes | 217 637 ops/s | 4 595 ns |
Kotlin Win64 | 252 bytes | 3 546 694 ops/s | 282 ns |
Kotlin Win64 (Final) | 152 bytes | 4 096 406 ops/s | 244 ns |
Kotlin Win64 (JSON) | 353 bytes | 185 788 ops/s | 5 382 ns |
Python Win64 | 252 bytes | 9 434 ops/s | 105 999 ns |
Python Win64 (Final) | 152 bytes | 11 635 ops/s | 85 945 ns |
Python Win64 (JSON) | 324 bytes | 61 737 ops/s | 16 198 ns |
Ruby Win64 | 252 bytes | 23 013 ops/s | 43 453 ns |
Ruby Win64 (Final) | 152 bytes | 33 361 ops/s | 29 975 ns |
Ruby Win64 (JSON) | 353 bytes | 50 842 ops/s | 19 669 ns |
Swift macOS | 252 bytes | 74 002 ops/s | 13 513 ns |
Swift macOS (Final) | 152 bytes | 100 755 ops/s | 9 925 ns |
Swift macOS (JSON) | 353 bytes | 18 534 ops/s | 53 953 ns |
Benchmark 2: Deserialization
Deserialization benchmark C++ code:
BENCHMARK_FIXTURE(DeserializationFixture, "Deserialize")
{
// Deserialize the account from the FBE stream
reader.deserialize(deserialized);
}
Deserialization benchmark results:
Language & Platform | Message size | Deserialization rate | Deserialization time |
---|---|---|---|
C++ Win64 | 252 bytes | 9 523 810 ops/s | 105 ns |
C++ Win64 (Final) | 152 bytes | 10 989 011 ops/s | 91 ns |
C++ Win64 (JSON) | 353 bytes | 1 375 516 ops/s | 727 ns |
C# Win64 | 252 bytes | 1 014 199 ops/s | 986 ns |
C# Win64 (Final) | 152 bytes | 1 607 717 ops/s | 622 ns |
C# Win64 (JSON) | 341 bytes | 258 532 ops/s | 3 868 ns |
Go Win64 | 252 bytes | 1 510 574 ops/s | 662 ns |
Go Win64 (Final) | 152 bytes | 1 540 832 ops/s | 649 ns |
Go Win64 (JSON) | 341 bytes | 251 825 ops/s | 3 971 ns |
Java Win64 | 252 bytes | 2 688 084 ops/s | 372 ns |
Java Win64 (Final) | 152 bytes | 3 036 020 ops/s | 329 ns |
Java Win64 (JSON) | 353 bytes | 308 675 ops/s | 3 240 ns |
JavaScript Win64 | 252 bytes | 133 892 ops/s | 7 469 ns |
JavaScript Win64 (Final) | 152 bytes | 292 273 ops/s | 3 422 ns |
JavaScript Win64 (JSON) | 341 bytes | 289 417 ops/s | 3 455 ns |
Kotlin Win64 | 252 bytes | 2 280 923 ops/s | 438 ns |
Kotlin Win64 (Final) | 152 bytes | 2 652 728 ops/s | 277 ns |
Kotlin Win64 (JSON) | 353 bytes | 250 524 ops/s | 3 992 ns |
Python Win64 | 252 bytes | 8 305 ops/s | 120 411 ns |
Python Win64 (Final) | 152 bytes | 11 661 ops/s | 85 758 ns |
Python Win64 (JSON) | 324 bytes | 48 859 ops/s | 20 467 ns |
Ruby Win64 | 252 bytes | 24 351 ops/s | 41 066 ns |
Ruby Win64 (Final) | 152 bytes | 33 555 ops/s | 29 802 ns |
Ruby Win64 (JSON) | 353 bytes | 42 860 ops/s | 23 331 ns |
Swift macOS | 252 bytes | 86 288 ops/s | 11 589 ns |
Swift macOS (Final) | 152 bytes | 10 3519 ops/s | 9 660 ns |
Swift macOS (JSON) | 353 bytes | 17 077 ops/s | 58 558 ns |
Benchmark 3: Verify
Verify benchmark C++ code:
BENCHMARK_FIXTURE(VerifyFixture, "Verify")
{
// Verify the account
model.verify();
}
Verify benchmark results:
Language & Platform | Message size | Verify rate | Verify time |
---|---|---|---|
C++ Win64 | 252 bytes | 31 250 000 ops/s | 32 ns |
C++ Win64 (Final) | 152 bytes | 35 714 286 ops/s | 28 ns |
C# Win64 | 252 bytes | 4 504 505 ops/s | 222 ns |
C# Win64 (Final) | 152 bytes | 8 064 516 ops/s | 124 ns |
Go Win64 | 252 bytes | 8 474 576 ops/s | 118 ns |
Go Win64 (Final) | 152 bytes | 9 090 909 ops/s | 110 ns |
Java Win64 | 252 bytes | 11 790 374 ops/s | 85 ns |
Java Win64 (Final) | 152 bytes | 16 205 533 ops/s | 62 ns |
JavaScript Win64 | 252 bytes | 1 105 627 ops/s | 905 ns |
JavaScript Win64 (Final) | 152 bytes | 5 700 408 ops/s | 175 ns |
Kotlin Win64 | 252 bytes | 8 625 935 ops/s | 116 ns |
Kotlin Win64 (Final) | 152 bytes | 13 373 757 ops/s | 75 ns |
Python Win64 | 252 bytes | 20 825 ops/s | 48 019 ns |
Python Win64 (Final) | 152 bytes | 23 590 ops/s | 42 391 ns |
Ruby Win64 | 252 bytes | 57 201 ops/s | 17 482 ns |
Ruby Win64 (Final) | 152 bytes | 74 262 ops/s | 13 466 ns |
Swift macOS | 252 bytes | 164 446 ops/s | 6 081 ns |
Swift macOS (Final) | 152 bytes | 228 154 ops/s | 4 383 ns |