I once had a junior admin whom I would not wish on any IT department. Before
he left, we had a consultant in who tuned our systems for maximum performance,
but I've spent the last few years looking for booby-traps. While upgrading
memory last week, I found that he'd reset BufPoolSize to 50-75% of memory.
While correcting that, I found he'd set LogPoolSize to 262K on all our machines.
Now documentation says the max is 8192. On all machines, BufPoolSize is set to
1/4 of available memory, so what is a good rule of thumb for the log buffers?
Give them all the max, regardless of available memory? Scale them per
available memory, i.e., 8192 for the 16TB boxes and scale on down for the
smaller machines? If this is an option that requires a reboot, I can't afford
to experiment since I can only reboot once a week.
|