hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bharath Mundlapudi (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-1796) Switch NameNode to use non-fair locks
Date Tue, 26 Jul 2011 06:53:09 GMT

    [ https://issues.apache.org/jira/browse/HDFS-1796?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13070976#comment-13070976
] 

Bharath Mundlapudi commented on HDFS-1796:
------------------------------------------

Fair lock guarantees the order of execution at the cost of performance. If delete and read
arrives at the same time but read operation takes the lock first, without fair lock, system
might schedule delete operation first and then read. Shouldn't we care about the correctness
and order of execution in file systems?

> Switch NameNode to use non-fair locks
> -------------------------------------
>
>                 Key: HDFS-1796
>                 URL: https://issues.apache.org/jira/browse/HDFS-1796
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>            Reporter: Hairong Kuang
>         Attachments: non-fair-lock.patch
>
>
> According to JavaDoc, a non-fair lock will normally have higher throughput than a fair
lock. Our experiment also shows an improved performance when using a non-fair lock. We should
switch namenode to use non-fair locks.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message