[HADOOP-561] one replica of a file should be written locally if possible

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.8.0
• Assignee: Dhruba Borthakur
• Reporter: Yoram Arnon
• Created: 26/Sep/06 16:50
• Updated: 08/Jul/09 16:42
• Resolved: 20/Oct/06 17:33

Description

one replica of a file should be written locally if possible. That's currently not the case.
Copying a 1GB file using hadoop dfs -cp running on one of the cluster nodes, all the blocks were written to remote nodes, as seen by fsck -files -blocks -locations on the newly created file.

as long as there is sufficient space locally, a local copy has significant performance benefits.