Balancing resources and performance

Maximum parallel performance requires multiple CPUs and multiple I/O devices to achieve I/O parallelism. As with most performance configuration, parallel systems reach a point of diminishing returns, and a later point where additional resources do not yield performance improvement.

You need to determine whether queries are CPU-intensive or I/O-intensive and when your performance is blocked by CPU saturation or I/O bottlenecks. If CPU utilization is low, spreading a table across more devices and using more worker processes increases CPU utilization and provides improved response time. Conversely, if CPU utilization is extremely high, but the I/O system is not saturated, increasing the number of CPUs can provide performance improvement.