2023版：深度比较几种.NET Excel导出库的性能差异

引言

背景和目的

本文介绍了几个常用的电子表格处理库，包括EPPlus、NPOI、Aspose.Cells和DocumentFormat.OpenXml，我们将对这些库进行性能测评，以便为开发人员提供实际的性能指标和数据。

下表将功能/特点、开源/许可证这两列分开，以满足需求：

功能 / 特点	EPPlus	NPOI	Aspose.Cells	DocumentFormat.OpenXml
开源	是	是	否	是
许可证	MIT	Apache	商业	MIT
支持的 Excel 版本	Excel 2007 及更高版本	Excel 97-2003	Excel 2003 及更高版本	Excel 2007 及更高版本

测评电脑配置

组件	规格
CPU	11th Gen Intel(R) Core(TM) i5-11320H @ 3.20GHz,2496 Mhz，4 个内核，8 个逻辑处理器
内存	40 GB DDR4 3200MHz
操作系统	Microsoft Windows 10 专业版
电源选项	已设置为高性能
软件	LINQPad 7.8.5 Beta
运行时	.NET 6.0.21

准备工作

使用Bogus库生成6万条标准化的测试数据。

void Main()
{
	string path = Path.Combine(Environment.GetFolderPath(Environment.SpecialFolder.Desktop), "test-data.json");
	using var file = File.Create(path);
	using var writer = new Utf8JsonWriter(file, new JsonWriterOptions { Indented = true });
	var data = new Bogus.Faker<Data>()
		.RuleFor(x => x.Id, x => x.IndexFaker + 1)
		.RuleFor(x => x.Gender, x => x.Person.Gender)
		.RuleFor(x => x.FirstName, (x, u) => x.Name.FirstName(u.Gender))
		.RuleFor(x => x.LastName, (x, u) => x.Name.LastName(u.Gender))
		.RuleFor(x => x.Email, (x, u) => x.Internet.Email(u.FirstName, u.LastName))
		.RuleFor(x => x.BirthDate, x => x.Person.DateOfBirth)
		.RuleFor(x => x.Company, x => x.Person.Company.Name)
		.RuleFor(x => x.Phone, x => x.Person.Phone)
		.RuleFor(x => x.Website, x => x.Person.Website)
		.RuleFor(x => x.SSN, x => x.Person.Ssn())
		.GenerateForever().Take(6_0000)
		.Dump();
	JsonSerializer.Serialize(writer, data);
	Process.Start("explorer", @$"/select, ""{path}""".Dump());
}

Bogus输出结果

Id	Gender	FirstName	LastName	Email	BirthDate	Company	Phone	Website	SSN
1	Male	Antonio	Paucek	Antonio.Paucek@gmail.com	1987/10/31 5:46:50	Moen, Willms and Maggio	(898) 283-1583 x88626	pamela.name	850-06-4706
2	Male	Kurt	Gerhold	Kurt.Gerhold40@yahoo.com	1985/11/1 18:41:01	Wilkinson and Sons	(698) 637-0181 x49124	cordelia.net	014-86-1757
3	Male	Howard	Hegmann	Howard2@hotmail.com	1979/7/20 22:35:40	Kassulke, Murphy and Volkman	(544) 464-9818 x98381	kari.com	360-23-1669
4	Female	Rosemarie	Powlowski	Rosemarie.Powlowski48@hotmail.com	1964/5/18 1:35:45	Will Group	1-740-705-6482	laurence.net	236-10-9925
5	Female	Eunice	Rogahn	Eunice84@gmail.com	1979/11/25 11:53:14	Rippin - Rowe	(691) 491-2282 x3466	yvette.net	219-75-6886
……

创建公共类方便正式测评使用

void Main()
{
    string path = Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\test-data.json";
    LoadUsers(path).Dump();
}

List<User> LoadUsers(string jsonfile)
{
    string path = jsonfile;
    byte[] bytes = File.ReadAllBytes(path);
    return JsonSerializer.Deserialize<List<User>>(bytes);
}

IObservable<object> Measure(Action action, int times = 5)
{
    return Enumerable.Range(1, times).Select(i =>
    {
        var sw = Stopwatch.StartNew();

        long memory1 = GC.GetTotalMemory(true);
        long allocate1 = GC.GetTotalAllocatedBytes(true);
        {
            action();
        }
        long allocate2 = GC.GetTotalAllocatedBytes(true);
        long memory2 = GC.GetTotalMemory(true);

        sw.Stop();
        return new
        {
            次数 = i,
            分配内存 = (allocate2 - allocate1).ToString("N0"),
            内存提高 = (memory2 - memory1).ToString("N0"),
            耗时 = sw.ElapsedMilliseconds,
        };
    }).ToObservable();
}

class User
{
    public int Id { get; set; }
    public int Gender { get; set; }
    public string FirstName { get; set; }
    public string LastName { get; set; }
    public string Email { get; set; }
    public DateTime BirthDate { get; set; }
    public string Company { get; set; }
    public string Phone { get; set; }
    public string Website { get; set; }
    public string SSN { get; set; }
}

代码解释

1、上面的代码单位是字节 (bytes)

2 、其中IObservable（System.IObservable）是用于处理事件流的接口，它实现了观察者模式。它表示一个可观察的序列，可以产生一系列的事件，并允许其他对象（观察者）来订阅和接收这些事件。IObservable 适用于动态的、实时的事件流处理，允许观察者以异步方式接收事件，可以用于响应式编程、事件驱动的编程模型等。

3、GC.GetTotalAllocatedBytes(true) 获取分配内存大小

GC.GetTotalMemory(true) 获取占用内存大小

性能测评

EPPlus

string path = Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\test-data.json";
List<User> users = LoadUsers(path);

Measure(() =>
{
    Export(users, Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\export.epplus.xlsx");
}).Dump("EPPlus");

void Export<T>(List<T> data, string path)
{
    using var stream = File.Create(path);
    using var excel = new ExcelPackage(stream);
    ExcelWorksheet sheet = excel.Workbook.Worksheets.Add("Sheet1");
    PropertyInfo[] props = typeof(User).GetProperties();
    for (var i = 0; i < props.Length; ++i)
    {
        sheet.Cells[1, i + 1].Value = props[i].Name;
    }
    for (var i = 0; i < data.Count; ++i)
    {
        for (var j = 0; j < props.Length; ++j)
        {
            sheet.Cells[i + 2, j + 1].Value = props[j].GetValue(data[i]);
        }
    }
    excel.Save();
}

输出结果

EPPlus (6.2.8) （2023/8/15）输出结果

次数ΞΞ	分配内存ΞΞ	内存提高ΞΞ	耗时ΞΞ
1	454,869,176	970,160	2447
2	440,353,488	176	1776
3	440,062,264	0	1716
4	440,283,584	0	1750
5	440,653,264	0	1813

EPPlus （4.5.3.2）（2019/6/16）输出结果

次数ΞΞ	分配内存ΞΞ	内存提高ΞΞ	耗时ΞΞ
1	963,850,944	192,048	2765
2	509,450,792	600	1897
3	509,872,160	424	1920
4	509,858,576	424	1989
5	509,651,512	424	2076

由此看出 相比2019,到了2023年EPPlus的性能得到了略微的提升

NPOI

示例代码一：XSSFWorkbook

List<User> users = LoadUsers(Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\test-data.json");

Measure(() =>
{
    Export(users, Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\export.npoi.xlsx");
}).Dump("NPOI");

void Export<T>(List<T> data, string path)
{
    IWorkbook workbook = new XSSFWorkbook();
    ISheet sheet = workbook.CreateSheet("Sheet1");

    var headRow = sheet.CreateRow(0);
    PropertyInfo[] props = typeof(User).GetProperties();
    for (var i = 0; i < props.Length; ++i)
    {
        headRow.CreateCell(i).SetCellValue(props[i].Name);
    }
    for (var i = 0; i < data.Count; ++i)
    {
        var row = sheet.CreateRow(i + 1);
        for (var j = 0; j < props.Length; ++j)
        {
            row.CreateCell(j).SetCellValue(props[j].GetValue(data[i]).ToString());
        }
    }

    using var file = File.Create(path);
    workbook.Write(file);
	workbook.Close();
}

输出结果

NPOI （2.6.1）（2023/7/12）输出结果

次数ΞΞ	分配内存	内存提高	耗时ΞΞ
1	1,589,285,792	567,272	5549
2	1,577,028,664	96	7043
3	1,577,398,488	48	8107
4	1,576,360,696	-90,512	9336
5	1,576,226,688	-3,120	8289

NPOI （2.4.1）（2018/12/18）输出结果

次数ΞΞ	分配内存	内存提高	耗时ΞΞ
1	1,648,548,696	526,824	6947
2	1,633,685,136	120	7921
3	1,634,033,296	24	8864
4	1,634,660,176	-90,200	8945
5	1,634,205,368	-2,584	8078

示例代码二：SXSSFWorkbook

List<User> users = LoadUsers(Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\test-data.json");

Measure(() =>
{
    Export(users, Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\export.npoi.xlsx");
}).Dump("NPOI");

void Export<T>(List<T> data, string path)
{
    IWorkbook workbook = new SXSSFWorkbook();
    ISheet sheet = workbook.CreateSheet("Sheet1");

    var headRow = sheet.CreateRow(0);
    PropertyInfo[] props = typeof(User).GetProperties();
    for (var i = 0; i < props.Length; ++i)
    {
        headRow.CreateCell(i).SetCellValue(props[i].Name);
    }
    for (var i = 0; i < data.Count; ++i)
    {
        var row = sheet.CreateRow(i + 1);
        for (var j = 0; j < props.Length; ++j)
        {
            row.CreateCell(j).SetCellValue(props[j].GetValue(data[i]).ToString());
        }
    }

    using var file = File.Create(path);
    workbook.Write(file);
	workbook.Close();
}

输出结果

NPOI （2.6.1）（2023/7/12）输出结果

次数	分配内存	内存提高	耗时
1	571,769,144	11,495,488	2542
2	482,573,584	96	5106
3	481,139,296	24	1463
4	481,524,384	48	1510
5	481,466,616	48	1493

NPOI （2.4.1）（2018/12/18）输出结果

次数	分配内存	内存提高	耗时
1	660,709,472	537,512	7808
2	650,060,376	8,128	8649
3	649,006,952	4,136	7064
4	649,267,920	-89,776	6973
5	649,955,024	48	6538

经过测试 发现SXSSFWorkbook 确实比XSSFWorkbook 性能好，有显著提升

由此看出 相比2018,到了2023年NPOI的性能得到了略微的提升

Aspose.Cells

List<User> users = LoadUsers(Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\test-data.json");

Measure(() =>
{
    Export(users, null);
}).Dump("Baseline");

string Export<T>(List<T> data, string path)
{
    PropertyInfo[] props = typeof(User).GetProperties();
	string noCache = null;
    for (var i = 0; i < props.Length; ++i)
    {
        noCache = props[i].Name;
    }
    for (var i = 0; i < data.Count; ++i)
    {
        for (var j = 0; j < props.Length; ++j)
        {
            noCache = props[j].GetValue(data[i]).ToString();
        }
    }
	return noCache;
}

输出结果

Aspose.Cells （23.8.0）（2023/8/9）输出结果

次数ΞΞ	分配内存	内存提高	耗时ΞΞ
1	9,878,784	19,248	82
2	9,854,688	96	63
3	9,854,688	24	59
4	9,854,688	24	62
5	10,106,352	24	72

Aspose.Cells （19.8.0）（2019/8/20）输出结果

次数ΞΞ	分配内存	内存提高	耗时ΞΞ
1	9,878,784	19,248	89
2	9,854,688	96	64
3	9,854,688	24	62
4	9,854,688	24	58
5	10,101,632	24	63

由此看出 相比2019,到了2023年Aspose.Cells的性能还是一样差不多

DocumentFormat.OpenXml

List<User> users = LoadUsers(Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\test-data.json");

Measure(() =>
{
    Export(users, Environment.GetFolderPath(Environment.SpecialFolder.Desktop) + @"\export.openXml.xlsx");
}).Dump("OpenXML");

void Export<T>(List<T> data, string path)
{
    using SpreadsheetDocument excel = SpreadsheetDocument.Create(path, SpreadsheetDocumentType.Workbook);

    WorkbookPart workbookPart = excel.AddWorkbookPart();
    workbookPart.Workbook = new Workbook();

    WorksheetPart worksheetPart = workbookPart.AddNewPart<WorksheetPart>();
    worksheetPart.Worksheet = new Worksheet(new SheetData());

    Sheets sheets = excel.WorkbookPart.Workbook.AppendChild<Sheets>(new Sheets());
    Sheet sheet = new Sheet
    {
        Id = excel.WorkbookPart.GetIdOfPart(worksheetPart),
        SheetId = 1,
        Name = "Sheet1"
    };
    sheets.Append(sheet);

    SheetData sheetData = worksheetPart.Worksheet.GetFirstChild<SheetData>();

    PropertyInfo[] props = typeof(User).GetProperties();
    {    // header
        var row = new Row() { RowIndex = 1 };
        sheetData.Append(row);
        row.Append(props.Select((prop, i) => new Cell
        {
            CellReference = ('A' + i - 1) + row.RowIndex.Value.ToString(),
            CellValue = new CellValue(props[i].Name),
            DataType = new EnumValue<CellValues>(CellValues.String),
        }));
    }
    sheetData.Append(data.Select((item, i) =>
    {
        var row = new Row { RowIndex = (uint)(i + 2) };
        row.Append(props.Select((prop, j) => new Cell
        {
            CellReference = ('A' + j - 1) + row.RowIndex.Value.ToString(),
            CellValue = new CellValue(props[j].GetValue(data[i]).ToString()),
            DataType = new EnumValue<CellValues>(CellValues.String),
        }));
        return row;
    }));
    excel.Save();
}

输出结果

DocumentFormat.OpenXml （2.20.0）（2023/4/7）输出结果

次数ΞΞ	分配内存	内存提高	耗时ΞΞ
1	614,013,080	421,552	3909
2	613,007,112	96	3487
3	613,831,672	104	3465
4	613,058,344	24	3650
5	613,161,096	24	3521

DocumentFormat.OpenXml （2.9.1）（2019/3/14）输出结果

次数ΞΞ	分配内存	内存提高	耗时ΞΞ
1	542,724,752	139,080	3504
2	542,478,208	96	2897
3	543,030,904	24	2826
4	542,247,544	24	2957
5	542,763,312	24	2941

由此看出 相比2019,到了2023年DocumentFormat.OpenXml的性能反而越差啦

结论和总结

结论一：如果你想找开源，(旧版本免费)，（最新版收费）EPPlus 依旧是最佳选择

次数ΞΞ	分配内存ΞΞ	内存提高ΞΞ	耗时ΞΞ
1	454,869,176	970,160	2447
2	440,353,488	176	1776
3	440,062,264	0	1716
4	440,283,584	0	1750
5	440,653,264	0	1813

结论二：如果你想找速度快，很稳定，但收费的，Aspose.Cells 依旧是最佳选择

次数ΞΞ	分配内存	内存提高	耗时ΞΞ
1	9,878,784	19,248	82
2	9,854,688	96	63
3	9,854,688	24	59
4	9,854,688	24	62
5	10,106,352	24	72

总结:

1、EPPlus表现不错，内存和耗时在开源组中表现最佳

2、收费的Aspose.Cells表现最佳，内存占用最低，用时也最短

作者 => 百宝门瞿佑明

此文章是对此前《.NET骚操作》2019年写的文章的更新和扩展

https://www.cnblogs.com/sdflysha/p/20190824-dotnet-excel-compare.html

原文地址：https://blog.baibaomen.com/2023版：深度比较几种-net-excel导出库的性能差异/