Two popular traditional join algorithms and their parallel versions are introduced. When designing join algorithms in serial computing environment, decomposing inner relation is considered as the right direction to sa...Two popular traditional join algorithms and their parallel versions are introduced. When designing join algorithms in serial computing environment, decomposing inner relation is considered as the right direction to save disk I/Os. However, two different decomposition algorithms are compared, such as inner vs. outer decomposition first algorithms for tuple-based and block-based nested loop joins, showing that the proposed approach is 20% better than general approach. Also lemmas are proved, when we have to use the outer decomposition first parallel join algorithms.展开更多
Based on the analysis of the task sizes and the load, this paper discussps the granularity ofrelation spliting in the spliting phase taking acount of task load being less than average load,andprobes into the relations...Based on the analysis of the task sizes and the load, this paper discussps the granularity ofrelation spliting in the spliting phase taking acount of task load being less than average load,andprobes into the relationship between the granularity and load balancing. The minimum number ofbuckets is determined on the basis of relation spliting granularity,and the maximum number ofproduct tuples of each node is induced under the prerequisite for ensuring load balance in paralleljoin.展开更多
基金supported by the National Research Foundation (NRF) of Korea through contract N-14-NMIR06
文摘Two popular traditional join algorithms and their parallel versions are introduced. When designing join algorithms in serial computing environment, decomposing inner relation is considered as the right direction to save disk I/Os. However, two different decomposition algorithms are compared, such as inner vs. outer decomposition first algorithms for tuple-based and block-based nested loop joins, showing that the proposed approach is 20% better than general approach. Also lemmas are proved, when we have to use the outer decomposition first parallel join algorithms.
文摘Based on the analysis of the task sizes and the load, this paper discussps the granularity ofrelation spliting in the spliting phase taking acount of task load being less than average load,andprobes into the relationship between the granularity and load balancing. The minimum number ofbuckets is determined on the basis of relation spliting granularity,and the maximum number ofproduct tuples of each node is induced under the prerequisite for ensuring load balance in paralleljoin.
