MapReduce vs. Relational DataBase
- scale out vs. scale up
- key/value pairs vs. relational tables
- functional programming vs. declarative query
- offline batch processing vs. online transactions(MapReduce is more like warehouse)
New API vs. Old API
- start from 0.20
- classes moved from org.apache.haddop.mapred(.lib) to org.apache.haddop.mapreduce(.lib)
- introduction of context object to unitify communication with framework
- with this way, map/reduce need not to be changed for further feature addition
- only some new method need to add to context object
- JobClient and JobConf replaced by Configuration and Job class
- Configuration class configures a job
- Job class defines and controls execution of job