Scale Out of Union Operators with IQ 16 SP10!
We have introduced an important performance enhancement with IQ 16 SP10 – parallelism of union operators.
Union operators occur commonly in analytics queries for two reasons: combining similar heterogeneous data sources, and partitioning data for ILM (information lifecycle management) purposes. Both use cases can flow large amounts of data through union operators during query processing.
IQ has supported MPP processing for most query operators since the 15.3 release. However, if a query contained unions, there were many cases where the processing could not be parallelized. In SP10, union operators can be fully parallelized and distributed regardless of the number of unions, union arms, or operator types! (Note that some type of reduction such as a Group By, OLAP, or filter is required for good scaling.) This feature further strengthens IQ’s position as a high-performance and cost-effective analytics database.
Unions can be partially parallelized even if some arms cannot allow parallelism or are not worth parallelizing on a cost basis. Union queries are parallelized automatically. No schema changes or tuning required. In-house performance testing has demonstrated linear scale-up with CPU cores (typically 12x-64x) for queries that flow large amounts of data through unions.
This is a completely free and transparent feature that can provide significant performance improvements over previous releases on the same hardware. You can read about all the performance optimization features of IQ 16 SP10 here: