Re: Forcedirectio In Solaris

One thing I couldnot understand is why it is not correct for large table full scans ? Doesnot Oracle buffer anyhow ? Why should we not turn on forcedirectio option for such tablespaces ? After all we don't need kernel buffering, do we ? Thanks

In article <8smmj3$m4q$1_at_news.syseca.fr>,   Philippe.Michel_at_syseca.thomson-csf.com (Philippe Michel) wrote:
> In article <8slefp$qnd$1_at_nnrp1.deja.com>, EnderW <ender29_at_mydeja.
 com> wrote:
> >Hi,
> > Quick question about mounting a ufs with forcedirectio option... I
> >would always think this should be on when the fs contains Oracle
> >datafiles. Is that true ? Thanks
>
> IMHO it is almost always true. A possible exception may be tablespaces
> containing tables of reasonable size, accessed via full scans fairly
 often,
> but not explicitly cached.
>
> My experience with Solaris' directio is that :
> - using it for online redo logs is a must ; log file write waits
> (from v$system_event) are considerably decreased.
> - db file write waits decrease is less obvious, but using it on
> datafiles frees plenty of memory for a better use.
> - you don't need to mount filesystems with forcedirectio option, you
> may prefer to set directio on individual files with something like :
>
> #include <sys/types.h>
> #include <sys/fcntl.h>
> #include <errno.h>
>
> int main(int argc, char *argv[])
> {
> extern int errno;
> int fd;
>
> errno = 0;
>
> if ((fd = open(argv[1], O_RDWR)) < 0)
> perror("open"), exit(1);
>
> if (directio(fd, DIRECTIO_ON))
> perror("directio"), exit(2);
>
> return 0;
> }
>
> Moreover, this kind of program (and another with DIRECTIO_OFF) can be
 used
> to experiment more easily than mounting filesystems with varying
 options.
>

--
Ender Wiggin


Sent via Deja.com http://www.deja.com/
Before you buy.