[ https://issues.apache.org/jira/browse/HDFS-8829?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14702365#comment-14702365 ]
He Tianyi commented on HDFS-8829: --------------------------------- I think auto-tuning does the magic when network latency is relatively high (>1ms), and has no effect when latency is low (< 0.1ms). But in standard cluster, latency won't be that high. So enabling auto-tuning by default would probably brings no performance gaining. While setting default value to 128 * 1024 will be compatible with current behavior. > 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)