Adaptive Cursor Sharing Dynamics Under Parameter Skew in Oracle SQL Engines

Abstract

Parameter skew in Oracle SQL workloads can cause significant performance variability when a single
shared execution plan is applied to bind values that produce very different data selectivity conditions.
This article investigates the behavior of Adaptive Cursor Sharing (ACS) in Oracle Database 19c using
a trace-based execution analysis framework, examining how cursor states evolve from bind-sensitive
to bind-aware as runtime feedback accumulates. The results show that ACS effectively stabilizes
performance when skew is pronounced, generating specialized child cursors tailored to distinct value
distributions. However, in cases of moderate skew, ACS adaptation occurs only after multiple
suboptimal executions, indicating a detectable lag between performance deviation and optimizer
response. The study concludes that ACS provides reliable long-term stability but may require
complementary tuning strategies to accelerate plan differentiation in latency-sensitive or rapidly
shifting workloads.