Re: Oracle on NT with raw devices

Uhmm, as someone who is alleged to have done such testing on "other" dbms's (yes, published/audited), I've seen that raw is much faster than 2% over ntfs.

You actually have more than the obvious gains to benefit from; for example, ntfs does a form of synchronous i/o per processor per file. It's most noticeable over 4p, but it is there nonetheless.

Anyway, I don't have the data handy to back it up at the moment, but you are better off with raw vs ntfs from a performance standpoint if you can manage it (big if, I grant). If you are really concerned with manageability, you probably don't want to rely on oracle anyway - use a vendor toolset like those offered by EMC or Sun or HP or whatever. OPS is another beastie of course.

Thomas Cox <thomasbcox_at_nospam.home.com> wrote in message news:37094B2A.C6F93A3E_at_nospam.home.com...
>
>
> "Christopher M. Day" wrote:
>
> > If you have performance figures for the gain in using RAW over NTFS then
> > please make them known to the rest of the group. I'm sure lots of people
> > are interested.
>
> > Chris.
>
> > > Using Raw FS increases performance by 10 - 15% for NT, but decreases
> > > manageability by 50%.
>
> > > Stilian
>
> According to folks who do TPC-C and -D benchmarks for a living, they
> find Raw gives about 2% better performance for them than NTFS. Since
> TPC benchmarks completely ignore manageability, they use Raw to get that
> 2% gain, and damn the torpedoes.
>
> For me, I would never use Raw unless I had to (i.e. Oracle Parallel
> Server). I really do care about manageability.
>
> Cheers.
> -Tom
> --
> See also my DBA white papers at <http://home.att.net/~tbcox>
Received on Wed Apr 07 1999 - 22:40:05 CDT