What does my CPU util drops every time I bounce my DB ?

If I bounce my DB after being up for a long time ( several months ), CPU utilization drops 20-25% right after that then it gradually catches up in serveral weeks.

I know you think it is because I cleared shared pool fragmentation and therefore reduced shared pool latch contention and CPU spin count. But I doubt that is the reason. First, I don't heavy latch wait, and surely not shared pool latch - which I believe you will for sure see if your freelist gets too long. Also, CPU util drop  occurs to ALL of my production databases, just a degree of difference.

But it looks like the higher the commit rate, the more difference I see after bounce.

The only other thing I suspect is buffer cache size. We have very large buffer caches ( 24 GB - 48 GB ) that to most site it is almost unheard of ( even to Oracle ). I don't think buffer cache has fragmentation issue, but could it be that when your cache gets filled up, it costs more to manage it ?

The high commit rate means that more cloned blocks are generated so it could potentially has to do with the oversized buffer cache and CPU overhead to maintain it. But this is just my uneducated guess.

Can anybody out there validate if this theory is true ? Does any of you have similar observation ?

Thanks in advance for your response.

BTW, this occurs to both 8.1.7 and 9.2 in 8K blocks ( under Solaris 2.8, if it matters ) Received on Fri Mar 26 2004 - 17:21:50 CST