Re: Max SGA size for 32-bit 9.2.0.6 on Solaris

I assumed that the discussion of this thread is focused on dedicated Oracle database server, not a home PC.

As to safe range, it depends on when your OS starts scanning for free memory. Solaris starts scanning at 1/64 remaining free. You definitely want to avoid that point. So you need to run your database overtime and collect OS stats to decide what your stable safe point is.

Usually I leave 2/64 during peak usage.

I am not sure what you mean by "too much housekeeping". Each group of blocks have their own dedicated bucket which is managed by dedicated cache buffers chain latch. The bigger the buffer cache the more child latches it creates. The ratio is the same. If there is an overhead managing the excessive latch children, it must be something very minimal as I haven't observed any yet.

We standardize all DB buffer cache to 48 GB now. We did not do this without careful assessments. We collect all statspack and OS data for before and after comparison. Turst me, if the problem out-weighs the benefit we wouldn't be doing this.

The bottom line is: Very large SGA does save CPU cycles (up to a certain level of course) which is the most expensive resource in your box. Memory is getting cheaper and cheaper. It is a fair trade. Received on Thu Jul 07 2005 - 11:55:43 CDT