Abstract:

A cache-conscious MapReduce star join algorithm was presented,each column of fact table was separately stored,and dimension table was divided into several column families according to dimension hierarchy.Fact table foreign key column and corresponding dimension table was co-location storage,thus reducing data movement in the join process.CC-MRSJ consists of two phases:firstly each foreign key column and the corresponding dimension table were joined; then the intermediate results were joined and random accessed measure columns,and so got the final result.CC-MRSJ read only the data needed,and cache utilization is high,so it has good cache conscious feature; it also takes advantage of late materialization,avoiding unnecessary data access and movement.CC-MRSJ has higher performance comparing to hive system based on SSB datasets.

Key words: star join, MapReduce, cache conscious, storage model