Recently, after a major upgrade of Database and Application version, and also to a new server, we had a Cache Hit Ratio of 99.62%. GOOD, you say ?
Very bad it was. That 99.62% also meant extremely high latch waits ("cache buffer chains") and very high Logical Reads.
With tuning, we brought Latch Waits down from 79.6% to 27.5% of the total Response Time, Logical Reads per DataBlock Changed down from 223 to 87 and an overall 74% reduction in Response Time.
BUT, our Hit Ratio [even after increasing the Buffer Cache by 50%] was down to 97.45%.
I was happy.
Very fortunately, we had Oracle Consultants who knew better than to look at the Cache Hit Ratio -- to focus on Logical Reads. Bring in people with the right attitude and setup the right KPIs and you will focus on the correct issues.
Updated 23-Feb : So, what was the problem ? The "cache buffer chains" latch waits was the indicator -- hot blocks, very high logical reads (reading the same blocks repeatedly) giving us a false Hit Ratio.
Updated 05-Mar : Latch waits are now down to less than 4% of total Response Time. Further tuning will be undertaken before we start looking at Physical Reads.