Re: Details about dbms_resouce_manager.calibrate_io
Date: Fri, 02 Mar 2012 14:40:58 +0100
we've executed it several times. It ran for about 12 - 18 minutes in our environment.
One important point that was not so obvious for me is that it appears that you'd better
run it when you have allocated sufficient space for the database - the tablespaces have to
be big enough (i.e. if the allocated space is less than or equal to the cache of your SAN,
you'll get certain figures, if it goes beyond that size - you'll get different results).
I realize that this sounds logical now, but it wasn't so intuitive for
me when we executed the
procedure for the first time (the SAN here has 64GB of cache).
On the first run, right after the database creation (with just few small tablespaces - system, sysaux, undo etc.) it read everything from the SAN's cache (95-100% hit ratio).
After that we allocated few 3TB tablespaces (this a new DWH setup) - the
storage team reported 0 cache hit ratio
on the SAN side ... (hence my question for details).
I found some old threads here, in the mailing list archive, that mention Exadata best practices and setting the values manually to fixed numbers (there's a MOS note about a bug that explains the procedure).
On 02/03/2012 14:16, boulayp_at_free.fr wrote:
> Initiates an I/O calibration, could take a long time ... as they say
> anyone around with some metrics about this ?
> ----- Mail original -----
> De: "Dimitre Radoulov"<cichomitiko_at_gmail.com>
> À: "ORACLE-L"<oracle-l_at_freelists.org>
> Envoyé: Vendredi 2 Mars 2012 12:32:42
> Objet: Details about dbms_resouce_manager.calibrate_io
> Hi all,
> does anyone know what exactly the dbms_resouce_manager.calibrate_io
> procedure does?
> I mean, it's obviously wrapped proprietary Oracle code, but the storage
> team asked for details.
> I could trace its execution, but I would prefer some official guide (I
> could have missed something
> while searching the official documentation, MOS and Internet).