we are running NW 6.5 SP2 two node clusters.
After some time (about 1 month) the servers total cache buffers are
reduced by 20000 per day.
How can we stop this behaviour?
Can SP5 be a solution?
How about the CacheMemoryThreshold parameter of the tsafs.nlm?
Should we use the Memory Tuning Calculator from Sysop Hamish Speirs?
Where should we start?

Thanks in advance.