Home > Workflows > Parallel Processing > Common Use Cases for Parall...
The following illustrate using parallel processing to improve performance of workflows:
Build Models in Parallel
If the input source for a model is a table defined with parallel and no intervening workflow nodes generate a table that changes this state, then models are built in parallel without any changes to the workflow settings.
If the input source is not already defined in parallel, you can still build the model in parallel:
For Classification and Regression models, for Classification and Regression Models, set Workflow to Parallel; the setting for Classification and Regression will be to Split input into Tables with Parallel option
For all models, set Workflow to Parallel and insert a Create Table node before to the build node and use the created table as input to the models.
Make transformations run faster, either using parallel execution or other techniques:
Set Workflow to Parallel.
All nodes that have a form of sample input could have some benefit. If the sample size is small, then the Oracle Database may not generate parallel queries, but a complex query could still trigger parallel execution.
Add a Create Table node after expensive transformations to reduce downstream repetitive querying costs.
Add an index to Create Table node to improve downstream Join performance.
Run Graph Node only in parallel; even if no other nodes are parallel performance may improve.
Follow these steps to run the Graph node only in parallel:
Set entire workflow to parallel.
Turn off parallel on all nodes except for graph node.
Run Graph node. Graph node sample data is now generated in parallel. If the Graph node sample is small or the query is simple, the query may not be made parallel.
Run a node in parallel just once to test performance
Follow these steps to see if running the node in parallel results in improved performance:
Set entire workflow to parallel.
Run workflow; note performance.
Set Workflow parallel option off.