[HADOOP-8069] Enable TCP_NODELAY by default for IPC

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.23.0, 2.4.1
• Fix Version/s: 2.6.0
• Component/s: ipc
• Target Version/s: 2.6.0
• Hadoop Flags: Reviewed
• Release Note: This change enables the TCP_NODELAY flag for all Hadoop IPC connections, hence bypassing TCP Nagling. Nagling interacts poorly with TCP delayed ACKs especially for request-response protocols.
• Assignee: Todd Lipcon
• Reporter: Todd Lipcon
• Created: 14/Feb/12 04:11
• Updated: 28/Jan/16 20:54
• Resolved: 29/Jul/14 03:00

Description

I think we should switch the default for the IPC client and server NODELAY options to true. As wikipedia says:

In general, since Nagle's algorithm is only a defense against careless applications, it will not benefit a carefully written application that takes proper care of buffering; the algorithm has either no effect, or negative effect on the application.

Since our IPC layer is well contained and does its own buffering, we shouldn't be careless.

Issue Links

• is duplicated by: HADOOP-7421 Performance Patch: Disable Client-Side Nagle By Default
• is related to: HADOOP-8071 Avoid an extra packet in client code when nagling is disabled
• relates to: HBASE-1177 Delay when client is located on the same node as the regionserver
• relates to: HDFS-9700 DFSClient and DFSOutputStream should set TCP_NODELAY on sockets for DataTransferProtocol