prod-consumer overlap factor affects how much pipelined parallelism can be created in a query plan. The default value is 20%, which means that if two operators in a parent-child relationship are run by separate worker processes, there is a 20% overlap. The remaining 80% of the operation is sequential. This affects the way in which Adaptive Server costs two plan fragments. Consider the example of a scan operator under a grouping operation. In such a case, if the scan operator takes N1 seconds and grouping operations take N2 seconds, the response time of the two operators is:
0.2 * max (N1, N2) + 0.8 * (N1 + N2)
In setting this parameter, consider the number of online engines on which Adaptive Server is running and the complexity of the queries to be run. As a general rule, use thread resources to scan on multiple partitions first. Then, if there are unused thread resources, use them to speed up vertical pipelined parallelism. Do not exceed a value of 50.
