hadoop-hdfs-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Todd Lipcon (JIRA)" <j...@apache.org>
Subject [jira] Created: (HDFS-1037) DataXceiver client should use TCP_NODELAY
Date Thu, 11 Mar 2010 23:30:27 GMT
DataXceiver client should use TCP_NODELAY

                 Key: HDFS-1037
                 URL: https://issues.apache.org/jira/browse/HDFS-1037
             Project: Hadoop HDFS
          Issue Type: Improvement
          Components: hdfs client
    Affects Versions: 0.22.0
            Reporter: Todd Lipcon

When the BlockReader connects to the DataXceiver and sends OP_READ_BLOCK, it doesn't set TCP_NODELAY
to true first. This means that the OS will wait some number of ms to try to bundle further
writes into the same TCP packet. Since we are only writing a small amount, and we already
use a buffer, we can safely turn on NODELAY to avoid the nagling delay, potentially improving
random read performance.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message