Do Not Chase the Hit Ratio
Old books say 'keep the buffer cache hit ratio above 95%'. That number is misleading: a badly written query that reads the same blocks millions of times makes the ratio look wonderful. Use advisors and wait events instead.
SGA Advisor
estd_db_time is the estimated total database time at each SGA size. Look for the size where the curve flattens. A bigger SGA beyond that point gives no real benefit.
Buffer Cache Advisor
estd_physical_read_factor = 1.0 is your current size. A value of 0.5 at a bigger size means disk reads would be cut in half.
PGA Advisor
estd_overalloc_count is 0 and the cache hit percentage is close to 100%.
Parse Health
A healthy OLTP system has few hard parses compared to total parses. Many hard parses mean applications are not using bind variables.
Open Cursors
If sessions get close to open_cursors, you will see ORA-01000. Often the application leaks cursors and the real fix is in the code, but you can raise the limit as a short-term step.
Quick Health Table
| Check | Healthy sign |
|---|---|
| Hard parses | Small share of total parses |
log file sync waits | Short average, a few ms |
| PGA advisor | estd_overalloc_count = 0 |
| Swapping at OS level | None. Swapping ruins Oracle performance. |
Conclusion
Use the SGA, buffer cache and PGA advisors to size memory, and check hard parses and open cursors for application problems. Ignore the old hit-ratio rules.

