Re: About the support of MapReduce task scheduling schemes

[Owen O'Malley]

On Feb 13, 2008, at 9:39 AM, Jaideep Dhok wrote:
Hi,
AFAIK right now they are only doing FCFS scheduling. You can read  
the code
in org.apache.hadoop.mapred.JobTracker.java. I think the code in
"getNewTaskForTaskTracker" method.

It actually prefers a host where the data is local. On large jobs, we  
get 95% of the maps running locally (assuming hdfs cluster == map/ 
reduce cluster). You also adding rack locality into the scheduling in  
HADOOP-1985.

-- Owen