We have ia64 box running HP-UX B.11.23, 16GB of RAM, and we'll eventually
have 10-20 high volume clusters installed. We could definitely use all the
RAM caching DB data. On our Linux boxes, we set the cluster shared_buffer
settings quite low (10000) and defer caching to the OS, and the OS seems to
play quite nicely with us, giving up buffer memory whenever an application
needs it. On HP-UX, I'm hearing a different story: the OS does not
perform well with OS cache size > 500MB due to contention between vhand and
the buffer cache. That suggests to me that we would need to manage the
allocation of the bulk of shared memory across these 10-20 clusters to
utilize the available RAM, a DBA headache we'd like to avoid if possible.
Am I missing something? Any suggestions on making this HP-UX box work
better for our situation, short of adding RAM?