[HDFS-281] Explore usage of the sendfile api via java.nio.channels.FileChannel.transfer{To|From} for i/o in datanodes

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Won't Fix
• Assignee: Unassigned
• Reporter: Arun Murthy
• Created: 30/Nov/07 08:43
• Updated: 17/Jul/14 18:54
• Resolved: 17/Jul/14 18:54

Description

We could potentially gain a lot of performance by using the sendfile system call:

$ man sendfile

DESCRIPTION
       This  call  copies  data between one file descriptor and another.  Either or both of these file descriptors may refer to a socket (but see below).
       in_fd should be a file descriptor opened for reading and out_fd should be a descriptor opened for writing.  offset is  a  pointer  to  a  variable
       holding  the input file pointer position from which sendfile() will start reading data.  When sendfile() returns, this variable will be set to the
       offset of the byte following the last byte that was read.  count is the number of bytes to copy between file descriptors.

       Because this copying is done within the kernel, sendfile() does not need to spend time transferring data to and from user space.

The nio package offers this via the java.nio.channels.FileChannel.transfer

{To|From}

apis:
http://java.sun.com/j2se/1.5.0/docs/api/java/nio/channels/FileChannel.html#transferFrom(java.nio.channels.ReadableByteChannel,%20long,%20long)
http://java.sun.com/j2se/1.5.0/docs/api/java/nio/channels/FileChannel.html#transferTo(long,%20long,%20java.nio.channels.WritableByteChannel)

From the javadocs:

     This method is potentially much more efficient than a simple loop that reads from this channel and writes to the target channel. Many operating systems can transfer bytes directly from the filesystem cache to the target channel without actually copying them.

---

Hence, this could well-worth exploring for doing io at the datanodes...

Issue Links

• is duplicated by: HDFS-2246 Shortcut a local client reads to a Datanodes files directly
• is related to: HADOOP-3164 Use FileChannel.transferTo() when data is read from DataNode.
• relates to: HDFS-916 Rewrite DFSOutputStream to use a single thread with NIO