为什么异步CTP表现不佳？

Question

我真的不明白为什么await,async也不要像我们应该那样提高我的代码的性能.

虽然持怀疑态度,但我认为编译器应该重写我的方法,以便下载并行完成...但似乎并没有真正发生.
(我确实意识到await并且async不创建单独的线程;但是,操作系统应该在parallal中进行下载,并在原始线程中回调我的代码 - 不应该吗？)

我使用async和await不正确吗？使用它们的正确方法是什么？

码:

using System;
using System.Net;
using System.Threading;
using System.Threading.Tasks;

static class Program
{
    static int SumPageSizesSync(string[] uris)
    {
        int total = 0;
        var wc = new WebClient();
        foreach (var uri in uris)
        {
            total += wc.DownloadData(uri).Length;
            Console.WriteLine("Received synchronized data...");
        }
        return total;
    }

    static async Task<int> SumPageSizesAsync(string[] uris)
    {
        int total = 0;
        var wc = new WebClient();
        foreach (var uri in uris)
        {
            var data = await wc.DownloadDataTaskAsync(uri);
            Console.WriteLine("Received async'd CTP data...");
            total += data.Length;
        }
        return total;
    }

    static int SumPageSizesManual(string[] uris)
    {
        int total = 0;
        int remaining = 0;
        foreach (var uri in uris)
        {
            Interlocked.Increment(ref remaining);
            var wc = new WebClient();
            wc.DownloadDataCompleted += (s, e) =>
            {
                Console.WriteLine("Received manually async data...");
                Interlocked.Add(ref total, e.Result.Length);
                Interlocked.Decrement(ref remaining);
            };
            wc.DownloadDataAsync(new Uri(uri));
        }
        while (remaining > 0) { Thread.Sleep(25); }
        return total;
    }

    static void Main(string[] args)
    {
        var uris = new string[]
        {
            // Just found a slow site, to demonstrate the problem :)
            "http://www.europeanchamber.com.cn/view/home",
            "http://www.europeanchamber.com.cn/view/home",
            "http://www.europeanchamber.com.cn/view/home",
            "http://www.europeanchamber.com.cn/view/home",
            "http://www.europeanchamber.com.cn/view/home",
        };
        {
            var start = Environment.TickCount;
            SumPageSizesSync(uris);
            Console.WriteLine("Synchronous: {0} milliseconds", Environment.TickCount - start);
        }
        {
            var start = Environment.TickCount;
            SumPageSizesManual(uris);
            Console.WriteLine("Manual: {0} milliseconds", Environment.TickCount - start);
        }
        {
            var start = Environment.TickCount;
            SumPageSizesAsync(uris).Wait();
            Console.WriteLine("Async CTP: {0} milliseconds", Environment.TickCount - start);
        }
    }
}

输出:

Received synchronized data...
Received synchronized data...
Received synchronized data...
Received synchronized data...
Received synchronized data...
Synchronous: 14336 milliseconds
Received manually async data...
Received manually async data...
Received manually async data...
Received manually async data...
Received manually async data...
Manual: 8627 milliseconds          // Almost twice as fast...
Received async'd CTP data...
Received async'd CTP data...
Received async'd CTP data...
Received async'd CTP data...
Received async'd CTP data...
Async CTP: 13073 milliseconds      // Why so slow??

Answer 1

克里斯的答案几乎是正确的,但引入了竞争条件并同步阻止所有任务.

作为一般规则,如果您有await/ async可用,最好不要使用任务延续.另外,不要使用WaitAny/ WaitAll- async等价物是WhenAny和WhenAll.

我会这样写:

static async Task<int> SumPageSizesAsync(IEnumerable<string> uris)
{
  // Start one Task<byte[]> for each download.
  var tasks = uris.Select(uri => new WebClient().DownloadDataTaskAsync(uri));

  // Asynchronously wait for them all to complete.
  var results = await TaskEx.WhenAll(tasks);

  // Calculate the sum.
  return results.Sum(result => result.Length);
}