Running Hive on top of the Mapreduce layer significantly hampers performance, often lagging behind parallel database systems by a factor of 50 to 100. Traditional Hadoop stacks are generally less efficient, leading major proponents like Cloudera and Facebook to develop alternatives such as Impala, which bypasses the Hadoop layer entirely for improved speed and efficiency.
