[MAPREDUCE-6729] Accurately compute the test execute time in DFSIO

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.9.0
• Fix Version/s: 2.8.0, 3.0.0-alpha1
• Component/s: benchmarks, performance, test
• Labels: performance test
• Target Version/s: 2.9.0
• Hadoop Flags: Reviewed
• Assignee: zhangminglei
• Reporter: zhangminglei
• Created: 06/Jul/16 09:15
• Updated: 09/Dec/16 02:38
• Resolved: 30/Jul/16 13:18

Description

When doing DFSIO test as a distributed i/o benchmark tool. Then especially writes plenty of files to disk or read from, both can cause performance issue and imprecise value in a way. The question is that existing practices needs to delete files when before running a job and that will cause extra time consumption and furthermore cause performance issue, statistical time error and imprecise throughput as the files are lots of. So we need to replace or improve this hack to prevent this from happening in the future.

public static void testWrite() throws Exception {
    FileSystem fs = cluster.getFileSystem();
    long tStart = System.currentTimeMillis();
    bench.writeTest(fs); // this line of code will cause extra time consumption as fs.delete(*,*) by the writeTest method
    long execTime = System.currentTimeMillis() - tStart;
    bench.analyzeResult(fs, TestType.TEST_TYPE_WRITE, execTime);
  }

private void writeTest(FileSystem fs) throws IOException {
  Path writeDir = getWriteDir(config);
  fs.delete(getDataDir(config), true);
  fs.delete(writeDir, true);    
  runIOTest(WriteMapper.class, writeDir);
  }

　

https://github.com/apache/hadoop/blob/trunk/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-jobclient/src/test/java/org/apache/hadoop/fs/TestDFSIO.java

Issue Links

• links to: GitHub Pull Request #112