Re: raw space vs unix filesystem

From: Brian P. Mac Lean <brian.maclean_at_teldta.com>
Date: 1996/08/02
Message-ID: <320263D2.64DB_at_teldta.com>#1/1


Johnny A. Wedekind wrote:
>
> I am a relative newbie to the Oracle world but in the Informix and the
> Sybase worlds, raw disk space is far far far preferred for dataspace over
> using the Unix filesystem. However, I am being informed that in the
> Oracle 7 world, using the Unix filesystem is quite acceptable. I have
> yet to attend the Oracle DBA classes and I am very anxious to have this
> statement confirmed or bashed. Please either reply to the newsgroup or
> to me directly at johnny_at_hoffman.ds.adp.com.
>
> Thank you much!!!

I responded to about the same question back in April 96, so I'll just paste the origional Q&A here:

Wachs Rene wrote:
>
> Hello
>
> This summer our DEC Alpha 4000 will be replaced by a DEC 8xxxx. Today we have a 20
> GB-DB with mainly 2 large-tabs (6.5 GB-Tab with 12 mio. rows, 3.5 GB-Tab with 7
> mio. rows). Performance is satisfactory, but sometimes multiuser behaviour is
> really bad. Nobody can explain that - neither Digital nor Oracle.
>
> DEC says : on the new DEC-Box don't make Files ! Make raw-Devices ! This will
> increase performance and improve the multiuser behaviour.
>
> My questions :
>
> - is this true ? Advantages of raw-devices against files ?
> - which conditions have to be done for raw-devices ?
> - raw-devices pitfalls ?
> - practical experiences ?
>
> Thanks very much.
>
> R. Wachs

  • is this true ?
        Yes. In all tuning books I have read including Oracle's it suggests 
        using raw devices. I have also done some timings on a Sun Sparc 1000
        with Solaris 2.3 using a Sparc Array via Oracle 7.1.3. I mixed and matched
        configurations to include different levels of raid and raw/UFS. In 
        short, the best was raid 0+1 (striping+mirroring).  But as always, 
        test your configuration and select the best solution.  

  • Advantages of raw-devices against files ?
        No more overhead of the system cache (remember to lower the UNIX 
        system cache value, a large one is not necessary anymore).

        I/O is direct and therefor faster. The work that UNIX must do to
        maintain the file system (UFS) is gone.

        Most UNIX file systems like to reserve 10% of any space for overhead,
        fragmentation, i-nodes, etc. (or something like that) leaving only
        90% usable. With raw devices the entire partition, all 100% is 
        available for Oracle tablespace datafiles.

  • which conditions have to be done for raw-devices ?
        I'm not sure what you mean hear. How about this, all parts of the
        db can be raw including the redo logs (I have not tried this but
        it's supposed to be do-able) excluding the archive area.      
          
  • raw-devices pitfalls ?
        must use the UNIX 'dd' command or something at the device level to
        do backups. No more cp/compress/tar/etc.

        You must brown nose the system administrator so they will do the 
        disk partition for you.

  • practical experiences ?
        I have had positive experiences with raw devices for the last 5 years
        on 5-10g databases. 

Hopefully you will find this helpful.

brian.maclean_at_teldta.com

(not responsible for lost or stolen data)

                                 ,        ,
                                /(        )`
                                \ \___   / |
                                /- _  `-/  '
                               (/\/ \ \   /\
                               / /   | `    \
                               O O   ) /    |
                               `-^--'`<     '
                   TM         (_.)  _  )   /
|  | |\  | ~|~ \ /             `.___/`    /
|  | | \ |  |   X                `-----' /
`__| |  \| _|_ / \  <----.     __ / __   \
                    <----|====O)))==) \) /====
                    <----'    `--' `.__,' \
                                 |        |
                                  \       /
                             ______( (_  / \______
                           ,'  ,-----'   |        \
                           `--{__________)        \/
Received on Fri Aug 02 1996 - 00:00:00 CEST

Original text of this message