使用 C# 和 WMI 计算出错误的 CPU 使用率

Question

这是我的函数，用于枚举 Windows 上的进程并计算每个进程的 CPU 使用百分比，但结果不正确。

CPU 使用率加起来并没有达到 100%，而是接近 120% 或 130%，我不知道我做错了什么。 它似乎可以计算各种应用程序（如 firefox、VS2010、office 等）的正确 CPU 使用率，但在系统空闲进程方面存在问题。

public List<ProcInfo> GetRunningProcesses()
{
    List<ProcInfo> allProcesses = new List<ProcInfo>();
    UInt64 currentProcessCpuTime = 0;
    UInt64 allProcessCpuTime = 0;

    SelectQuery wmiQuery = new SelectQuery("SELECT Name, Description, ProcessId, KernelModeTime, UserModeTime FROM Win32_Process");
    ManagementObjectSearcher oSearcher = new ManagementObjectSearcher(connectionScope, wmiQuery);
    ManagementObjectCollection moc = oSearcher.Get();

    foreach (ManagementObject mo in moc)
    {
        allProcessCpuTime += (UInt64)mo["KernelModeTime"] + (UInt64)mo["UserModeTime"];
    }

    foreach (ManagementObject mo in moc)
    {
        currentProcessCpuTime = (UInt64)mo["KernelModeTime"] + (UInt64)mo["UserModeTime"];
        allProcesses.Add(new ProcInfo((string)mo["Name"], (string)mo["Description"], (UInt32)mo["ProcessId"], (currentProcessCpuTime / (double)allProcessCpuTime * 100));
    }

    return allProcesses;
}

编辑：

我发现我的功能全部错误。

我正在开始悬赏最佳可行的解决方案。解决方案需要适用于本地和远程系统，并且应该很快。

Answer 1

下面是带有性能计数器的 C# 代码：

public static void DumpProcessesCpu(string machineName)
{
    List<PerformanceCounter> counters = new List<PerformanceCounter>();

    foreach (Process process in Process.GetProcesses(machineName))
    {
        PerformanceCounter processorTimeCounter = new PerformanceCounter(
            "Process",
            "% Processor Time",
            process.ProcessName,
            machineName);

        processorTimeCounter.NextValue();
        counters.Add(processorTimeCounter);
    }

    Thread.Sleep(1000); // 1 second wait, needed to get a sample

    foreach (PerformanceCounter processorTimeCounter in counters)
    {
        Console.WriteLine("Process:{0} CPU% {1}",
            processorTimeCounter.InstanceName,
            processorTimeCounter.NextValue());
    }
}

它的灵感来自于：https://learn.microsoft.com/en-us/archive/blogs/bclteam/how-to-read-performance-counters-ryan-byington
你真的不可能比这更快，文章中解释了原因。基本上，您必须读取该值两次才能正确，因此您需要在样本之间等待。

但是，根据您想要执行的操作，例如，假设您想编写一个“远程任务管理器”，您可以在后台任务（线程）中对所有这些进行编码，并定期更新这些值，这样最终用户就不会真正查看样本之间的延迟。

Answer 2

这是经过测试和验证的 C# 代码块，感谢 fejesjoco，我使用了他的代码并进行了测试以使其正常工作。

public class CPUUtilizationTests
{

    [Test]
    public void TestPercentProcessorTime()
    {
        Assert.That(PercentProcessorTime("Idle"), Is.Not.GreaterThan(100.0));
    }

    public float PercentProcessorTime(string processName)
    {
        var mos = new ManagementObjectSearcher("SELECT * FROM Win32_PerfRawData_PerfProc_Process");
        var run1 = mos.Get().Cast<ManagementObject>().ToDictionary(mo => mo.Properties["Name"].Value, mo => mo);
        System.Threading.Thread.Sleep(1000); // can be an arbitrary number
        var run2 = mos.Get().Cast<ManagementObject>().ToDictionary(mo => mo.Properties["Name"].Value, mo => mo);

        if (!run2.ContainsKey(processName)) throw new Exception(string.Format("Process not found: {0}", processName));

        string percentageProcessorTime = "PercentProcessorTime";
        string total = "_Total";

        ulong percentageDiff = (ulong)run2[processName][percentageProcessorTime] - (ulong)run1[processName][percentageProcessorTime];
        ulong totalDiff = (ulong)run2[total][percentageProcessorTime] - (ulong)run1[total][percentageProcessorTime];

        return ((float)percentageDiff / (float)totalDiff)*100.0f;
    }

}

Answer 3

        var mos = new ManagementObjectSearcher("SELECT * FROM Win32_PerfRawData_PerfProc_Process");
        var run1 = mos.Get().Cast<ManagementObject>().ToDictionary(mo => mo.Properties["Name"].Value, mo => (ulong)mo.Properties["PercentProcessorTime"].Value);
        Thread.Sleep(570); // can be an arbitrary number
        var run2 = mos.Get().Cast<ManagementObject>().ToDictionary(mo => mo.Properties["Name"].Value, mo => (ulong)mo.Properties["PercentProcessorTime"].Value);

        var total = run2["_Total"] - run1["_Total"];

        foreach (var kvp in run1)
        {
            var proc = kvp.Key;
            var p1 = kvp.Value;
            if (run2.ContainsKey(proc))
            {
                var p2 = run2[proc];
                Console.WriteLine("{0}: {1:P}", proc, (double)(p2 - p1) / total);
            }
        }