recently, i have encountered hive performance issue many times. since i am going to migrate data center, so we have two CDH env, my experience is below, maybe valuable to you .
1. dont choose 4T disk, 1-2T is best.
2. don't check mapjoin as global parameter in hive.
3. don't use jvm resue function
4. open hive parallel is good choice
5. the most important is if you have got performance in HIVE, don't believe any parameter can improve your performance, tunning your HIVE SQL and check the table files.
many HIVE SQL developer even doesn't know what's MR, so sucks/