[jira] [Created] (HADOOP-10554) Performance: Scan metrics for 2.4 are notably down compared to 0.23.9

["patrick white (JIRA)" <ji...@apache.org>]

patrick white created HADOOP-10554:
--------------------------------------

             Summary: Performance: Scan metrics for 2.4 are notably down compared to 0.23.9
                 Key: HADOOP-10554
                 URL: https://issues.apache.org/jira/browse/HADOOP-10554
             Project: Hadoop Common
          Issue Type: Bug
    Affects Versions: 2.4.0
            Reporter: patrick white


Performance comparison benchmarks for Scan test's runtime and throughput metrics are slightly out of 5% tolerance in 2.x compared against 0.23. The trend is consistent across later releases in both lines, latest release numbers are;

Runtime:
2.4.0.0     ->    73.6 seconds (avg 5 passes)
0.23.9.12 ->    69.4 seconds (avg 5 passes)
Diff: -5.7%

Throughput:
2.4.0.0     ->   28.67 GB/s (avg 5 passes)
0.23.9.12 ->  30.41 GB/s (avg 5 passes)
Diff: -6.1%

Scan test is specifically measuring the average map's input read performance. The diff is consistent when run on a larger (350 node) perf environment, we are in process of seeing if this reproduces in a smaller cluster, using appropriately scaled inputs.




--
This message was sent by Atlassian JIRA
(v6.2#6252)