Re: page_in values in vmstat output

Sybrand Bakker <gooiditweg_at_nospam.demon.nl> wrote in message news:<bphd4v81d9bhbvtt3mos99ssela2ueq2lf_at_4ax.com>...
> On 9 Feb 2003 09:49:09 -0800, mkbobba_at_yahoo.com (Bob) wrote:
>
> >We are not able to understand why the server is paging when around 80%
> >of the RAM is not reserved for any other application.
> >
> >
> >
> >
> >Our server is heavily IO bound and IO is not distributed equally among
> >the 16 disks available in this server. Does this fact has any remote
> >connection with our above problem.
>
>
> Some things to look into
> - You might suffer from poorly written applications committing every
> single individual record
> - Your redo log buffers look way too big, and if your online redo logs
> are tuned improperly, that will also make you suffer. Search for
> 'checkpoint not complete' messages in your alerts
> - before Solaris 2.8 kaio doesn't work properly. This means when
> asynchronous io is enabled, these requests will fail and will get
> resubmitted as synchronous disk requests
> - Which kind of file system? ufs? In that case you need to mount your
> disks with the forcedirectio option, otherwise Solaris will use most
> of the RAM as filecache, and the filecache is *paged*.
> When vxfs: You will be able to turn off the filecache with the
> Advanced Edition of Veritas only, the stripped Veritas sw which ships
> with Solaris doesn't have this option.
>
> A fellow sufferer...
>
>
> Sybrand Bakker, Senior Oracle DBA
>
> To reply remove -verwijderdit from my e-mail address

We are using UFS file system and I am not sure if they have used forcedirectio option while mounting the disks.

Thanks for throwing the light on IO. So filecache may be part of the problem. Received on Wed Feb 12 2003 - 06:58:31 CST