Scan benchmark is more than 1.5X slower(almost 92% increased) in 0.23 than Hadoop-0.20.204 on 350 nodes size cluster.
Tasks may take upto 3 seconds to exit after completion
MR AM's ContainerAllocator is assigning the allocated containers very slowly
Counters aggregation slowed down significantly after MAPREDUCE-3749
Data Locality suffers if the AM asks for containers using IPs instead of hostnames
RackResolver should maintain a cache to avoid repetitive lookups.
Counters are getting calculated twice at job-finish and delaying clients.
NPE in FileOutputCommitter when running a 0 reduce job
last split generated by FileInputFormat.getSplits may not have the best locality