Re: A Real Stinker to Solve - EXPERTS APPLY WITHIN (DB Crashing)

Johne_uk wrote:
> Johne_uk wrote:
> > > This is for db blocks. log_buffer is the parameter we need to know.
> > > Since you didn't list it in the init.ora parameters, it must be
> > > default, which is 512 KB or 128 KB * CPU_COUNT, whichever is greater.
> > > See
> > > http://download-west.oracle.com/docs/cd/B10501_01/server.920/a96533/memory.htm#38479
> > > to see if you need to change it.
> > >
> >
> > Hi, yes the log_buffer is curently at the default size of 512k. I'm
> > reading the article you attached as regards resizing.
> > Thanks
>
> The 'redo buffer allocation retries' statistic is 4336. This should be
> nearly zero - thereore, I'm going to increase the size of log_buffer to
> 2M.

Somewhere I've seen advice from Oracle that over 1M is generally useless. For one particular OLTP production system of mine, 1.5M with 200M redo is perfect. YMMV, but start with 1M. Also, consider what your maximum log switch rate is (the times listed in the alert log), you may want to resize the logs if there are times when they are constantly switching, and that change would affect the usage of this buffer. More thought may be necessary if you have times of slamming in mass quantities of data or large scale updates.

jg

--
@home.com is bogus.
"You're probably a sociopath who finds solace in exact measurement.
Join the club." - Dr. Science