Adam Fuchs created HDFS-7380: -------------------------------- Summary: unsteady and slow performance when writing to file with block size >2GB Key: HDFS-7380 URL: https://issues.apache.org/jira/browse/HDFS-7380 Project: Hadoop HDFS Issue Type: Bug Affects Versions: 2.4.0 Reporter: Adam Fuchs Attachments: BenchmarkWrites.java
Appending to a large file with block size > 2GB can lead to periods of really poor performance (4X slower than optimal). I found this issue when looking at Accmulo write performance in ACCUMULO-3303. I wrote a small test application to isolate this performance down to some basic API calls (to be attached). A description of the execution can be found here: https://issues.apache.org/jira/browse/ACCUMULO-3303?focusedCommentId=14202830&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14202830 The specific hadoop version was as follows: {code} [root@n1 ~]# hadoop version Hadoop 2.4.0.2.1.2.0-402 Subversion g...@github.com:hortonworks/hadoop.git -r 9e5db004df1a751e93aa89b42956c5325f3a4482 Compiled by jenkins on 2014-04-27T22:28Z Compiled with protoc 2.5.0 >From source with checksum 9e788148daa5dd7934eb468e57e037b5 This command was run using /usr/lib/hadoop/hadoop-common-2.4.0.2.1.2.0-402.jar {code} -- This message was sent by Atlassian JIRA (v6.3.4#6332)