Re: Kernel parameter recommendations for HP-UX 11.11 in regard to SGA size

On Jun 6, 3:16 pm, joel garry <joel-ga..._at_home.com> wrote:
> On Jun 5, 5:30 pm, hpuxrac <johnbhur..._at_sbcglobal.net> wrote:
>
>
>
>
>
> > On Jun 5, 7:53 pm, joel garry <joel-ga..._at_home.com> wrote:
>
> > > On Jun 5, 2:39 pm, Swoop <s..._at_doesnotwork.com> wrote:
>
> > > > sybrandb wrote:
> > > > > On Jun 5, 12:01 pm, Swoop <s..._at_doesnotwork.com> wrote:
> > > > >> I'm knew to HP-UX and was just after some advice in regards to some
> > > > >> kernel parameters. I've done a fair bit of research but some of the
> > > > >> things I've read seem to conflict and because lower versions of the OS
> > > > >> have different settings and because the amount of RAM on the boxes (that
> > > > >> they are configuring) differ it is hard to peg it down.
>
> > > > >> For example, a metalink document recommends setting the system buffer
> > > > >> cache to a maximum of 128M but it says this for a low amount of RAM,
> > > > >> however, it doesn't specify what it means by low :(
>
> > > > >> I'll be testing it out first on a test box.
>
> > > > >> 2 processor box, HP-UX 11.11
> > > > >> 4Gb RAM
> > > > >> Running 9.2.0.8
> > > > >> vxfs filesystem
> > > > >> Running four databases (largest is 30Gb, others are 2Gb) as a pure
> > > > >> database server
>
> > > > >> Just wondering how others have configured the following parameters
>
> > > > >> bufpages
> > > > >> dbc_max_pct
> > > > >> dbc_min_pct
> > > > >> nbuf
>
> > > On a 4 processor 8G system with 2 databases (one test) I have:
>
> > > bufpages 0 (nbuf*2)
> > > dbc_max_pct 50
> > > dbc_min_pct 5 (see the kernel help for determining, once you
> > > get to a steady state it doesn't matter so much)
> > > nbuf 0 "This parameter is for backwards
> > > compatibility and should be set to zero because dynamic buffer cache
> > > is preferred."
>
> > Do you really have dbc_max_pct set to the default of 50?
>
> Yes.
>
>
>
> > Sounds like you probably have 4 gig of os memory serving as a file
> > buffer cache ... why?
>
> Nothing before now ever stimulated me to look at it. Also, not using
> direct I/O. Unable to test when it was new, due to RAID-5 salespeople
> having their way.
>
> This crashconf thingy tells me I'm using 1.6G. I'm hoping that's
> usage rather than just some configuration, but I don't see any such
> value in the .h file. The man is a bit unclear on that.
>
>
>
> > Most oracle hpux installs should have the two parameters dbc_min_pct
> > and dbc_max_pct set to very close if not the same values
>
> No argument there, although seeing comments elsewhere about setting it
> so you have good hit ratios doesn't give me the warm'n'fuzzies.
>
> Being able to say "hey, want a couple of free G on the production
> server?" right before a new division is added might get some notice,
> even without any business performance issues, so thanks!
>
> jg

Ok you have me a little confused. You have a server with 8 gig of RAM and you are running oracle on it and you aren't doing direct IO so you are going through the os buffer cache?

Well if so then you want to proceed very carefully here in thinking about any changes.

We run raw here and that of course has always not used the os buffer cache.

You may want to read this ...

http://www.oracle.com/technology/deploy/performance/pdf/TWP_Oracle_HP_files.pdf Received on Fri Jun 08 2007 - 04:19:28 CDT