Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Oracle-managed RAID-like tablespaces

Re: Oracle-managed RAID-like tablespaces

From: Svend Jensen <svend.jensen_at_it.dk>
Date: Sun, 28 Jul 2002 18:52:26 +0200
Message-ID: <3D44214A.2070002@it.dk>


Rick Denoire wrote:

> Hello all DBAs out there
>
> Not sure, but I seem to have read somewhere that there is the
> possibility in Oracle to set up tablespaces that would span several
> independent harddisks. Of course, one can always do that adding
> additional files in the corresponding harddisks, but that is not what
> I mean.
>
> I mean that Oracle would automatically spread data of a segment over
> several independent harddisks in a way that they are not put
> physically in a sequencial order in one file/disk, but building
> "stripes" over a group of disks.

It sounds contra dictionary what you are gaining at. Or at least unclear.

It is possible to have Oracle 'spread' an object automatically over several data files. An eksample. Create a tablespace, localy managed, uniform size (say 1M) data file 'drive_1......' size 2000M. Add data file to tablespace 'drive_2.... size 2000M and so on until the tablespace got the size needed or better. Then create the object (table). If the object size is much larger than the uniform size, Oracle will create extents in a round robin fashion on all the data files in the tablespace. I believe all objects (small and large) will start with first extent/segment on first datafile and continue on second datafile, datafile three.... until the requested size is reached. This schema is seen on 8.1.7EE on windows 2000 advanced server.

rgds Svend Jensen
>
> If this feature really exists, then doing a full table scan would
> cause to distribute read requests over several disks, even if they are
> organized as JBOD ("just a bunch of disks") instead as a real RAID,
> but effectively, read performance would compare to that of a RAID. I
> am assuming that in this situation Oracle would not wait for a read
> request to complete in order to demand the next request (asynchronous
> reading). I think that for this Oracle's own stripe management, files
> put in different disks should have the same size, of course.
>
> And by the way, this feature allows a harddisk to behave as a part of
> a RAID while at the same time sequential writes to just this one
> particular disk would still be possible (for redologs, for example),
> without disturbing "RAID-like" read requests of other disks! (Remember
> that in a hardware RAID, one has no control about the physical layout
> of data on disks other than determining the stripe width and RAID
> level at setup time).
>
> Too good to be true. Am I just dreaming or what? Please correct me!
>
> I am interested in using this feature, if it exists, because we are
> setting up a cheap test environment without RAID.
>
> Thanks
> Rick
>
>
Received on Sun Jul 28 2002 - 11:52:26 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US