[
https://issues.apache.org/jira/browse/HDFS-9201?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
nijel updated HDFS-9201:
------------------------
Attachment: HDFS-9201_draft.patch
Had a try with NNThroughputBenchmark for read and write flow in a 40 core
machines with few changes in core flow.
config : -threads 200 -files 500000 -filesPerDir 100
Results (in ops per second)
||Read ~ 5% improvement observed||
|| trial || Without change || After the change |
| trail 1 | 187336 | 198886 |
| trail 2 | 181752 | 200642 |
| trail 3 | 195388 | 200964 |
| | | |
||Write - No change in write flow||
| trail 1 | 29585 | 29330 |
| trail 2 | 29670 | 29577 |
| trail 3 | 29584 | 29670 |
Attached the draft patch with the changes used for test
Please give your opinion
> Namenode Performance Improvement : Using for loop without iterator
> ------------------------------------------------------------------
>
> Key: HDFS-9201
> URL: https://issues.apache.org/jira/browse/HDFS-9201
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: nijel
> Assignee: nijel
> Labels: namenode, performance
> Attachments: HDFS-9201_draft.patch
>
>
> As discussed in HBASE-12023, the for each loop syntax will create few extra
> objects and garbage.
> For arrays and Lists can change to the traditional syntax.
> This can improve memory foot print and can result in performance gain.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
