Next: Related Work [cont'd.]
Up: Related Work
Previous: Related Work
- They have presented a taxonomy of skew in the parallel databases
Attribute Value skew : Inherent in the dataset .
Partition skew : Occurs due to imbalance in the load between the nodes.
AVS leads to partition skew , other factors include the following :
(i) TPS : Tuple placement skew i.e. initial placement of tuples between nodes may vary.
(ii)SS : Selectivity skew may vary from node to node for different predicates.
(iii) RS : Redistribution skew occurs when tuples are redistributed in preparation for the actual join.
(iv) JPS : Join Product skew occurs on individual nodes' varying selectivity leading to difference in number of output tuples.
- Walton et al used an analytical model to compare the scheduling hash join
algorithm and the hybrid hash-join algo of Gamma .
- Result: Scheduling hash effectively handles RS while hybrid hash degrades and eventually becomes worse than scheduling hash as RS increases( significant skew only then ).
DBMS
1999-03-11