[HDFS-8829] Make SO_RCVBUF and SO_SNDBUF size configurable for DataTransferProtocol sockets and allow configuring auto-tuning

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.3.0, 2.6.0
• Fix Version/s: 2.8.0, 3.0.0-alpha1, 2.7.5
• Component/s: datanode
• Target Version/s: 2.8.0

Release Note

HDFS-8829 introduces two new configuration settings: dfs.datanode.transfer.socket.send.buffer.size and dfs.datanode.transfer.socket.recv.buffer.size. These settings can be used to control the socket send buffer and receive buffer sizes respectively on the DataNode for client-DataNode and DataNode-DataNode connections. The default values of both settings are 128KB for backwards compatibility. For optimum performance it is recommended to set these values to zero to enable the OS networking stack to auto-tune buffer sizes.

• Assignee: He Tianyi
• Reporter: He Tianyi
• Created: 28/Jul/15 13:51
• Updated: 15/Nov/17 01:03
• Resolved: 14/Sep/15 23:04

Description

  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);

The last line sets SO_RCVBUF explicitly, thus disabling tcp auto-tuning on some system.

Shall we make this behavior configurable?

Issue Links

• is related to: HDFS-10326 Disable setting tcp socket send/receive buffers for write pipelines
• is related to: HDFS-9259 Make SO_SNDBUF size configurable at DFSClient side for hdfs write scenario