我很惊讶我没有读到之前的答案，我认为这是最关键的方面:

当我想要一个没有标识的类型时，我使用结构体。例如一个3D点:

public struct ThreeDimensionalPoint
{
    public readonly int X, Y, Z;
    public ThreeDimensionalPoint(int x, int y, int z)
    {
        this.X = x;
        this.Y = y;
        this.Z = z;
    }

    public override string ToString()
    {
        return "(X=" + this.X + ", Y=" + this.Y + ", Z=" + this.Z + ")";
    }

    public override int GetHashCode()
    {
        return (this.X + 2) ^ (this.Y + 2) ^ (this.Z + 2);
    }

    public override bool Equals(object obj)
    {
        if (!(obj is ThreeDimensionalPoint))
            return false;
        ThreeDimensionalPoint other = (ThreeDimensionalPoint)obj;
        return this == other;
    }

    public static bool operator ==(ThreeDimensionalPoint p1, ThreeDimensionalPoint p2)
    {
        return p1.X == p2.X && p1.Y == p2.Y && p1.Z == p2.Z;
    }

    public static bool operator !=(ThreeDimensionalPoint p1, ThreeDimensionalPoint p2)
    {
        return !(p1 == p2);
    }
}

如果你有这个结构体的两个实例，你不关心它们是内存中的单个数据还是两个。你只关心他们持有的价值。

这是一个老话题，但希望提供一个简单的基准测试。

我已经创建了两个。cs文件:

public class TestClass
{
    public long ID { get; set; }
    public string FirstName { get; set; }
    public string LastName { get; set; }
}

and

public struct TestStruct
{
    public long ID { get; set; }
    public string FirstName { get; set; }
    public string LastName { get; set; }
}

运行基准:

创建一个TestClass 创建一个TestStruct 创建100个TestClass 创建100个TestStruct 创建10000个TestClass 创建10000 TestStruct

结果:

BenchmarkDotNet=v0.12.0, OS=Windows 10.0.18362
Intel Core i5-8250U CPU 1.60GHz (Kaby Lake R), 1 CPU, 8 logical and 4 physical cores
.NET Core SDK=3.1.101
[Host]     : .NET Core 3.1.1 (CoreCLR 4.700.19.60701, CoreFX 4.700.19.60801), X64 RyuJIT  [AttachedDebugger]
DefaultJob : .NET Core 3.1.1 (CoreCLR 4.700.19.60701, CoreFX 4.700.19.60801), X64 RyuJIT


|         Method |           Mean |         Error |        StdDev |     Ratio | RatioSD | Rank |    Gen 0 | Gen 1 | Gen 2 | Allocated |
|--------------- |---------------:|--------------:|--------------:|----------:|--------:|-----:|---------:|------:|------:|----------:|

|      UseStruct |      0.0000 ns |     0.0000 ns |     0.0000 ns |     0.000 |    0.00 |    1 |        - |     - |     - |         - |
|       UseClass |      8.1425 ns |     0.1873 ns |     0.1839 ns |     1.000 |    0.00 |    2 |   0.0127 |     - |     - |      40 B |
|   Use100Struct |     36.9359 ns |     0.4026 ns |     0.3569 ns |     4.548 |    0.12 |    3 |        - |     - |     - |         - |
|    Use100Class |    759.3495 ns |    14.8029 ns |    17.0471 ns |    93.144 |    3.24 |    4 |   1.2751 |     - |     - |    4000 B |
| Use10000Struct |  3,002.1976 ns |    25.4853 ns |    22.5920 ns |   369.664 |    8.91 |    5 |        - |     - |     - |         - |
|  Use10000Class | 76,529.2751 ns | 1,570.9425 ns | 2,667.5795 ns | 9,440.182 |  346.76 |    6 | 127.4414 |     - |     - |  400000 B |

MSDN给出了答案: 在类和结构之间选择。

基本上，这个页面给你一个4项清单，并告诉你使用一个类，除非你的类型满足所有的标准。

不定义结构，除非 类型包含以下所有内容 特点: 它在逻辑上表示一个值，类似于基本类型 (整数，双精度，等等)。 实例大小小于16字节。 它是不可变的。 它不需要经常装箱。

在以下情况下使用类:

它的身份很重要。结构在按值传递给方法时被隐式复制。 它将占用大量内存。 它的字段需要初始化式。 你需要从基类继承。 你需要多态行为;

在以下情况下使用结构:

It will act like a primitive type (int, long, byte, etc.). It must have a small memory footprint. You are calling a P/Invoke method that requires a structure to be passed in by value. You need to reduce the impact of garbage collection on application performance. Its fields need to be initialized only to their default values. This value would be zero for numeric types, false for Boolean types, and null for reference types. Note that in C# 6.0 structs can have a default constructor that can be used to initialize the struct’s fields to nondefault values. You do not need to inherit from a base class (other than ValueType, from which all structs inherit). You do not need polymorphic behavior.

✔️如果该类型的实例很小且通常存在时间很短，或者通常嵌入在其他对象中，请考虑定义一个结构而不是类。

我会在以下情况下使用结构体:

an object is supposed to be read only(every time you pass/assign a struct it gets copied). Read only objects are great when it comes to multithreaded processing as they don't requite locking in most cases. an object is small and short-living. In such a case there is a good chance that the object will be allocated on the stack which is much more efficient than putting it on the managed heap. What is more the memory allocated by the object will be freed as soon as it goes outside its scope. In other words it's less work for Garbage Collector and the memory is used more efficient.