Re: Bigfile tablespaces on OS filesystems?
Date: Mon, 21 Sep 2009 18:58:38 +0100
Message-ID: <7765c8970909211058k5aa73caj250e3cf6db778688_at_mail.gmail.com>
why not suggest an o/s that supports async i/o (Linux or Win64 spring to mind) :) That should give them bigger jitters than ASM!
OK so you can see that, from your description, I don't like the no ASM rule - no good reason has been given. I would however in your situation definitely favour many smaller files (don't forget ulimit) over fewer big ones both for the reason you mention and for management reasons -- assuming you use many partitions and hence it is sensible to have many tablespaces and hence many datafiles.
I was about to say that the first paragraph was only semi-joking but
actually it's not really very much of a joke at all of performance is a
concern. If you can't do the sensible performance things that have been
around for years with the OS groups policies why follow the policies at all,
why not use something that will do the job.
Niall
On Mon, Sep 21, 2009 at 6:44 PM, Lou Avrami <avramil_at_concentric.net> wrote:
> Hello all,
>
> I have an INTERESTING scenario that I would like to run by you folks.
>
> I have been assigned the task of migrating an existing data warehouse of
> approximately 2 TB to a new server running 11gR1. After much discussion ...
> the UNIX/SAN team will not OK using ASM, so I am stuck with creating
> tablespaces on OS filesystems.
>
> Originally, I wanted to use bigfile tablespaces for the larger tablespaces,
> in conjunction with ASM. Now that ASM is out the window (at least for the
> moment - maybe we will be allowed to migrate to it someday), does it still
> make sense to use bigfile tablespaces?
>
> The OS that the new data warehouse will be on is HP-UX 11.23, so I believe
> that means no asynch I/O at the filesystem level.
>
> What would the read/write performance be like for something like a 500 GB
> bigfile tablespace on a UNIX OS filesystem? My initial impression is that
> concurrent reads and writes would have horrible performance, because UNIX
> would "force" the entire 500 GB file to be accessed for each read/write
> operation.
>
> From a read/write performance perspective, would it be better to use
> smallfile tablespaces with multiple uniform datafiles?
>
> Any helpful information would be much appreciate down here in the rabbit
> hole. :-)
>
> Thanks,
> Lou Avrami
>
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>
-- Niall Litchfield Oracle DBA http://www.orawin.info -- http://www.freelists.org/webpage/oracle-lReceived on Mon Sep 21 2009 - 12:58:38 CDT