Swift JSON Benchmarks

Benchmark results for different Swift frameworks which work on JSON (encoding/decoding). For reference there is also a benchmark of .NET Core framework (Newtonsoft.JSON), however it's not included in ranks.

All benchmark was performed on my computer:

• Device: MacBook Pro (13-inch, 2016, Two Thunderbolt 3 ports)
• Processor: 2 GHz Intel Core i5
• Memory: 16 GB 1867 MHz LPDDR3
• Graphics: Intel Iris Graphics 540 1536 MB

Swift version:

Apple Swift version 4.2.1 (swiftlang-1000.11.42 clang-1000.11.45.1)
Target: x86_64-apple-darwin18.5.0

Here you can find other benchmark: https://github.com/bwhiteley/JSONShootout

Build & Run benchmarks

Prerequisites

Before you start to run benchmarks on your device you have to install:

• Swift
• .NET Core

When you want to change Swift Protobuf model you need to have additional applications:

• Google Protobuf compiler
• Swift Protobuf code generator

Build & Run

To build and run you have to execute following commands:

$ git clone https://github.com/mczachurski/SwiftBenchmarkJSON.git
$ cd SwiftBenchmarkJSON
$ ./build.sh
$ ./run.sh

List of frameworks

Below there are links to all benchmarked frameworks.

• HandyJSON
• Marshall
• ObjectMapper
• PMJSON
• SwiftProtobuf
• Codable - is built in Swift encoding & decoding based on Codable protocol
• NetCore - is .NET Core implementation of JSON decoding/encoding (Newtonsoft.JSON framework)

Encoding (serialization)

Benchmarks of algorithms which transform object(s) to the corresponding JSON string. Object has structure like on following snippet.

TaskClassDto(
    id: UUID().uuidString,
    createDate: Date(),
    name: "Task 1",
    isFinished: false,
    subtasks: 212,
    weight: 3.14,
    children: ["taska", "taskb", "taskc"],
    description: "Lorem ipsum dolor sit amet, consectetur adipiscing elit. 
        Aenean eget sem erat. Quisque dictum tellus in feugiat facilisis. 
        Vivamus porttitor vel arcu id cursus. Cras interdum massa ac rhoncus 
        ornare. Sed quis massa felis. Curabitur blandit tempor enim, vitae 
        euismod nibh tincidunt a. Duis faucibus dapibus purus nec dictum. 
        Suspendisse dignissim sapien et consequat lobortis.",
    deadline: Date(),
    tags: ["tag1", "tag2", "tag3", "tag4", "tag5"]
)

Benchmark #1 - Encoding single object

In this benchmark we are encoding 10,000 times single object.

Code

Simple implementation with Swift Codable protocol looks like below.

let entity = TaskDto(...)

let encoder = JSONEncoder()
let formatter = DateFormatter()
formatter.dateFormat = "yyyy-MM-dd'T'HH:mm:ssZ"
encoder.dateEncodingStrategy = .formatted(formatter)

do {
    for _ in 1...10_000 {
        let jsonData = try encoder.encode(entity)
        _ = String(data: jsonData, encoding: .utf8)
    }
} catch {
    print("Error during serialization object to JSON string: \(error)")
}

Results

Benchmark #2 - Encoding list of objects

In this benchmark we are encoding 10,000 times list of 100 objects.

Code

Implementation with Swift Codable protocol looks like on below snippet.

let list = getListOfTasks() // Returns array of 100 tasks
let encoder = JSONEncoder()

let formatter = DateFormatter()
formatter.dateFormat = "yyyy-MM-dd'T'HH:mm:ssZ"
encoder.dateEncodingStrategy = .formatted(formatter)

do {
    for _ in 1...10_000 {
        let jsonData = try encoder.encode(list)
        _ = String(data: jsonData, encoding: .utf8)
    }
} catch {
    print("Error during serialization object to JSON string: \(error)")
}

Results

Decoding (deserialization)

Benchmarks of algorithms which transform JSON to the corresponding object(s). JSON which is parsed looks like on below snippet.

{
    "id": "e24e39c2-7b96-4a16-8cb6-bb96239171e5",
    "createDate": "2009-02-15T00:00:00Z",
    "name": "Task 1",
    "isFinished": false,
    "subtasks": 212,
    "weight": 3.14,
    "children": ["taska", "taskb", "taskc"],
    "description": "Lorem ipsum dolor sit amet, consectetur adipiscing elit.
        Aenean eget sem erat. Quisque dictum tellus in feugiat facilisis. 
        Vivamus porttitor vel arcu id cursus. Cras interdum massa ac rhoncus 
        ornare. Sed quis massa felis. Curabitur blandit tempor enim, vitae 
        euismod nibh tincidunt a. Duis faucibus dapibus purus nec dictum. 
        Suspendisse dignissim sapien et consequat lobortis.",
    "deadline": "2019-02-15T01:10:00Z",
    "tags": ["tag1", "tag2", "tag3", "tag4", "tag5"]
}

Benchmark #3 - Decoding single object

In this benchmark we are decoding 10,000 times JSON string which represent single object.

Code

Simple implementation with Swift Codable protocol.

let jsonData = getJsonData() // Returns Data object (with single JSON object)
let decoder = JSONDecoder()
let formatter = DateFormatter()

formatter.dateFormat = "yyyy-MM-dd'T'HH:mm:ssZ"
decoder.dateDecodingStrategy = .formatted(formatter)

do {
    for _ in 1...10_000 {
        let entityJsonData = jsonData.data(using: .utf8)!
        _ = try decoder.decode(TaskClassDto.self, from: entityJsonData)
    }
} catch {
    print("Error during deserialization object from JSON string: \(error)")
}

Results

Benchmark #4 - Decoding list of objects

In this benchmark we are decoding 10,000 times JSON string which represent 100 objects.

Code

Simple implementation with Swift Codable protocol.

let listJson = getJsonListData() // Returns Data object (with list of JSON objects)
let decoder = JSONDecoder()
let formatter = DateFormatter()
formatter.dateFormat = "yyyy-MM-dd'T'HH:mm:ssZ"
decoder.dateDecodingStrategy = .formatted(formatter)

do {
    for _ in 1...10_000 {
        let listJsonData = listJson.data(using: .utf8)!
        _ = try decoder.decode([TaskClassDto].self, from: listJsonData)
    }
} catch {
    print("Error during deserialization object from JSON: \(error)")
}

Results

Bonus (Binary SwiftProtobuf)

SwiftProtobuf provides additionally binary decoding (instead of JSON string). That kind of encoding/decoding is very fast. Below there is table with result for each of above test.

It's fastest decoding/encoding. Unfortunately client have to support that kind of data. However, if your API have to be very fast you should take into account that kind of encoding/decoding.

Contributing

If you would like to add another framework for comparison, submit a pull request. Add folder with new benchmark implementation (we have separate application for each framework). Add new framework to build.sh and run.sh scripts.