Limits of Cache Fusion in Reducing Buffer Cache Contention for Large Tables in Oracle RAC 19c

Goal

Determine whether Cache Fusion reliably lowers query latency for concurrent read‑only workloads on large tables (>10 GB) in a two‑node Oracle RAC 19c environment, given that documentation notes potential benefits while empirical results show variability.

Constraints include a read‑heavy OLAP pattern, default 8 KB block size, automatic memory management, and the inability to isolate Cache Fusion effects from other interconnect traffic.

Uncertainty remains about the workload intensity at which Cache Fusion transitions from neutral to beneficial, and whether specific initialization parameters can be adjusted to guarantee a latency improvement.

• What concurrent request rate triggers a measurable latency reduction when Cache Fusion is enabled?
• How does varying the database block size affect the efficacy of Cache Fusion for large‑table scans?
• Which initialization parameters (e.g., _fusion_io_size, cluster_database_instances) can be tuned to ensure Cache Fusion provides a consistent benefit for read‑only workloads?