I believe the last comment on that issue by Owen O'Malley is still valid:
"Most of these things except for the jvm reuse have been done in other patches. I've only left this bug open because of the jvm reuse issues, but the patch is currently useless because the code has changed so much in the last year. In fact, it would be relatively tricky to get right with the capturing of stdout/stderr from the tasks."
Personally I'd like to see the option of the tasktracker starting one jvm per job and run multiple tasks in that instead of one per task. But as Owen says, it would be a pretty big change.
/Johan Camilo Arango wrote:
Hi, I saw this interesting patch in Jira that improves performance: http://issues.apache.org/jira/browse/HADOOP-249 does the latest version of hadoop already include this fix? Thanks, Camilo A.
