hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "He Tianyi (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-8829) DataNode sets SO_RCVBUF explicitly is disabling tcp auto-tuning
Date Wed, 29 Jul 2015 00:56:05 GMT

    [ https://issues.apache.org/jira/browse/HDFS-8829?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14645305#comment-14645305
] 

He Tianyi commented on HDFS-8829:
---------------------------------

This affects pipeline throughput, and I've observed a 30% performance gain with tcp auto-tuning
enabled. 
(since 128KB window size is not always optimal)

Could be particularly useful with SSD drives (where disk throughput may be greater than net
throughput).


> DataNode sets SO_RCVBUF explicitly is disabling tcp auto-tuning
> ---------------------------------------------------------------
>
>                 Key: HDFS-8829
>                 URL: https://issues.apache.org/jira/browse/HDFS-8829
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: datanode
>    Affects Versions: 2.3.0, 2.6.0
>            Reporter: He Tianyi
>            Assignee: kanaka kumar avvaru
>
> {code:java}
>   private void initDataXceiver(Configuration conf) throws IOException {
>     // find free port or use privileged port provided
>     TcpPeerServer tcpPeerServer;
>     if (secureResources != null) {
>       tcpPeerServer = new TcpPeerServer(secureResources);
>     } else {
>       tcpPeerServer = new TcpPeerServer(dnConf.socketWriteTimeout,
>           DataNode.getStreamingAddr(conf));
>     }
>     tcpPeerServer.setReceiveBufferSize(HdfsConstants.DEFAULT_DATA_SOCKET_SIZE);
> {code}
> The last line sets SO_RCVBUF explicitly, thus disabling tcp auto-tuning on some system.
> Shall we make this behavior configurable?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message