Parallel Query Optimizer
-
Parallel Query Optimizer builds a query execution plan employing all system resources on the basis of data-partition knowledge, it can easily employ few tens of processors to execution a single query with high degree of parallelism. Both data-parallel and data-flow paradigms are used to achieve this goal
-
Support complex CPU-intensive or I/O-intensive queries for DSS
-
near-linear or even super-linear speedup can be achieved (due to larger effective caches)
-
Major focus for research --- optimized parallel algorithms needed
-
carried out by a query coordinator on the RDBMS server (the query coordinator generates the optimal parallel plan rather than having the best serial plan simply executed in parallel. This optimization is performed by the coordinator without the need for any external information (DB2 and Oracle7 use different approaches)
-
most relational operators are parallizable, including full-table scan, sort, aggregate operators, join, merge
-
System/Server tuning plays an important role in achieving optimized performance in a parallel RDBMS --- benchmarking is an effective means of performing tuning
|
