Yunqi Zhang created HADOOP-12356:
------------------------------------

             Summary: CPU usage statistics on Windows
                 Key: HADOOP-12356
                 URL: https://issues.apache.org/jira/browse/HADOOP-12356
             Project: Hadoop Common
          Issue Type: Bug
          Components: util
         Environment: CPU: Intel Xeon
OS: Windows server
            Reporter: Yunqi Zhang


In {{hadoop-common/src/main/java/org/apache/hadoop/util/SysInfoWindows.java}} 
(in trunk), the reported CPU usage is incorrect.

{code:title=SysInfoWindows.java|borderStyle=solid}
// This is where we calculate all the usage information
void refreshIfNeeded() {
  ...
  if (lastCumCpuTimeMs != -1) {
    cpuUsage = (cumulativeCpuTimeMs - lastCumCpuTimeMs)
        / (refreshInterval * 1.0f);
  }
  ...
}
{code}

This {{cpuUsage}} is the cumulative CPU usage across all logical cores. For 
example, this number will range in [0.0, 12.0] if a server has 12 cores. 
Instead of reporting this raw number, we should do the following to be 
consistent with {{SysInfoLinux.java}}. In {{SysInfoLinux.java}}, {{cpuUsage}} 
ranges in [0.0, 1200.0], and the return value of {{getCpuUsage()}} ranges from 
[0.0, 100.0].

{code:title=SysInfoWindows.java|borderStyle=solid}
// This is where we calculate all the usage information
void refreshIfNeeded() {
  ...
  if (lastCumCpuTimeMs != -1) {
    cpuUsage = (cumulativeCpuTimeMs - lastCumCpuTimeMs)
        * 100F / refreshInterval;
  }
  ...
}

@Override
public float getCpuUsage() {
  refreshIfNeeded();
  return cpuUsage / numProcessors;
}
{code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to