Re: Veritas -- do # of spindles really matter ??

In article <387F476D.B9081B01_at_earthlink.net>,   John Ott <johnott_at_earthlink.net> wrote:
>
>
> rob van laarhoven wrote:
>
> > >>The question:
> > >>Which is faster -- (a) the current 2-spindle architecture, or (b)
if 5
> > >>separate filesystems were explicitly-defined on their own
dedicated disk?
> > >>Keep in mind both (a) and (b) would take advantage of the same
> > Veritas/RAID
> > >>configuration (striping and mirroring).
> > >>
> > >
> > IMHO there should be a 1 to 1 relation between mountpoints and
physical
> > disks. This will not make the disks faster but will make it
possible to
> > place databaseobjects on a specific physical disk, this will be
handy when
> > distributing IO over the disks..
>
> I respectfully disagree.
>
> You can manage this by having the DBA create a logical structure
> and then do the stripping and mirroring you want using Veritas
> or whatever volume management software you are using.
>
> This also makes sense for when you upgrade your hardware. Just
> backup the logical volumes, then set up the new hardware
> and take advantage of its features and present logical volumes
> to the database for the restore.
>
> In some cases you can't get a spindle to define a mount point.
> For instance if you are using a HP autoraid array or a Hitachi 7700e.
> These and other RAID hardware present logical drives to the OS.
> So you can't get a spindle if you wanted to. Many RAID systems
> keep a lot of data in RAM anyway.
>
> So with the advances in Hardware RAID defining I/O layout
> at the Database level is not as important as it used to be.
> And in many cases doesn't make sense at all.
>
> Let the DBA worry about speeding up the
> applications. Let the RAID hardware and software speed
> up I/O.
>
> later
> John
>
>

It is true that advances in RAID hardware have made the DBA's placement of datafiles more "logical". However, whether two RAID 5 spindles are acceptable for system performance DEPENDS on the performance requirements of the system, and the budget available.

The "old school" thinking is that the DBA will place each datafile in its ideal location on individual physical disks. The reality is somewhere between this statement and the scenario described above.

This sounds like a medium size database with < 100 GB data. If performance becomes a concern the following steps can be taken:

1. If you have only two "spindles" to work with, organize the datafiles such that at any given time, one spindle is reading and the other is writing. This will avoid massive writes obliterating reads stored in the physical disk cache.
2. separate the redo logs onto raw disk. Use Oracle mirroring instead of hardware mirroring because Oracle mirroring may protect you against errors being mirrored between redo members.
3. RAID 5 is great for reads, poor for writes. Add some RAID 1/0 disk for the high i/o datafiles in your application.
4. In order to take advantage of RAID 1, ie. placement of datafiles on individual disks, the DBA must have an in-depth knowledge of the i/o pattern of the application and be able to map it to individual disks.

Sent via Deja.com http://www.deja.com/
Before you buy. Received on Fri Jan 14 2000 - 21:09:26 CET