I have a job that has lots of tiny map tasks that finish very fast (I think my max time was 9 seconds) I understand that I should change my input to avoid that... and it's difficult because these processes are using a transform script on binary data so it makes it difficult to pull off (Long story). That being said, the cluster isn't distributing the job. Only one node is running the map tasks. They are finishing quickly, but I am just wondering would cause that ... I know the job tracker and task trackers on other nodes are running, it's just odd.