批量上传文件到Azure(C#)的最快方法

Question

批量上传文件Azure Blob存储的最快方法是什么？我试过两种方法,sync并async上传,async显然是最快的,但我不知道是否有更好的方法？是否内置了对批量上传的支持？我在文档中找不到任何内容但可能错过了它.

这是我跑的测试:

static void Main(string[] args)
{
    int totalFiles = 10; //10, 50, 100
    byte[] randomData = new byte[2097152]; //2mb
    for (int i = 0; i < randomData.Length; i++)
    {
        randomData[i] = 255;
    }

    CloudStorageAccount cloudStorageAccount = CloudStorageAccount.Parse(ConfigurationManager.AppSettings["StorageConnectionString"]);
    var blobClient = cloudStorageAccount.CreateCloudBlobClient();

    var container = blobClient.GetContainerReference("something");
    container.CreateIfNotExists();


    TimeSpan tsSync = Test1(totalFiles, randomData, container);
    TimeSpan tsAsync = Test2(totalFiles, randomData, container);

    Console.WriteLine($"Sync: {tsSync}");
    Console.WriteLine($"Async: {tsAsync}");

    Console.ReadLine();

}

public static TimeSpan Test2(int total, byte[] data, CloudBlobContainer container)
{
    Stopwatch sw = new Stopwatch();
    sw.Start();
    Task[] tasks = new Task[total];
    for (int i = 0; i < total; i++)
    {
        CloudBlockBlob blob = container.GetBlockBlobReference(Guid.NewGuid().ToString());
        tasks[i] = blob.UploadFromByteArrayAsync(data, 0, data.Length);

    }
    Task.WaitAll(tasks);


    sw.Stop();
    return sw.Elapsed;
}

public static TimeSpan Test1(int total, byte[] data, CloudBlobContainer container)
{
    Stopwatch sw = new Stopwatch();
    sw.Start();
    for (int i = 0; i < total; i++)
    {
        CloudBlockBlob blob = container.GetBlockBlobReference(Guid.NewGuid().ToString());
        blob.UploadFromByteArray(data, 0, data.Length);

    }
    sw.Stop();
    return sw.Elapsed;
}

这个输出是:

10个文件

Sync: 00:00:08.7251781
Async: 00:00:04.7553491
DMLib: 00:00:05.1961654

Sync: 00:00:08.1169861
Async: 00:00:05.2384105
DMLib: 00:00:05.4955403

Sync: 00:00:07.6122464
Async: 00:00:05.0495365
DMLib: 00:00:06.4714047

50个文件

Sync: 00:00:39.1595797
Async: 00:00:22.5757347
DMLib: 00:00:25.2897623

Sync: 00:00:40.4932800
Async: 00:00:22.3296490
DMLib: 00:00:26.0631829

Sync: 00:00:39.2879245
Async: 00:00:24.0746697
DMLib: 00:00:26.9243116

我希望这是一个有效的问题.

谢谢

编辑:

我已经用"DMLib"测试更新了结果,以回应到目前为止给出的答案.DMLib是一个没有配置更改的测试(见上文)没有性能提升

我ServicePointManager.DefaultConnectionLimit = Environment.ProcessorCount * 8;按照文档的推荐运行了一些测试,这增加了相当多的上传速度,但它也提高了我的异步方法的上传速度.到目前为止,DMlib还没有给我任何值得的性能提升.我在下面的配置更改中添加了第二组测试结果.

我也设置ServicePointManager.Expect100Continue = false;了这个速度没有区别.

测试结果用 ServicePointManager.DefaultConnectionLimit = Environment.ProcessorCount * 8;

10个文件

Sync: 00:00:07.6199307
Async: 00:00:02.9615565
DMLib: 00:00:02.6629716

Sync: 00:00:08.7721797
Async: 00:00:02.8246599
DMLib: 00:00:02.7281091

Sync: 00:00:07.8437682
Async: 00:00:03.0171246
DMLib: 00:00:03.0190045

50个文件

Sync: 00:00:40.2395863
Async: 00:00:10.3157544
DMLib: 00:00:10.5107740

Sync: 00:00:40.2473358
Async: 00:00:10.8190161
DMLib: 00:00:10.2585441

Sync: 00:00:41.2646137
Async: 00:00:13.7188085
DMLib: 00:00:10.8686173

我是否错误地使用了库,因为它似乎没有提供比我自己的方法更好的性能.

Answer 1

请使用Azure存储数据移动库,它是AzCopy的核心.该库正是解决问题的工具.:)

Answer 2

使用Azcopy来完成您的工作。不幸的是，它是一个独立的exe。

您还可以将文件分割成块（按起始位置和偏移量读取）并并行上传。这有点复杂，你必须根据机器调整上传线程。