Re: block_size for large DB

From: Anton Dischner <dischner_at_med.uni-muenchen.de>
Date: Wed, 14 Sep 1994 16:21:38 +0200
Message-ID: <dischner-1409941621380001_at_gkc12a.klch.med.uni-muenchen.de>


In article <355j9v$e39_at_potogold.rmii.com>, bromberg_at_hobbes (Rich Bromberg) wrote:
Hi,

some time ago i did a little bit of benchmarking to decide 4096 or 8192.

I gained up to 10% more performance with blocksize 8192. We have tables up to 3Gig data + 2 Gig index.

But be aware, installing with 8192 is tricky (was so with 7.0.12.1.0 and IBM RS6000). Have a look into pctfree and pctused (i am using pctfree 5 and pctused 90).

i hope this helps,

Toni

> I have a question regarding the db_block_size parameter for large
> decision support type databases (> 100GB). We have found a number
> of good reasons for increasing the block size to 8K on our system,
> such as a larger number of extents allowed and increased row-size per
> block. It would also stand to reason that the increased block-size,
> coupled with a large db_file_multiblock_read_count parameter would
> increase performance for the large table scan queries which the
> typical DSS user submits.
>
> One argument against the larger block size has been that the cache
> hit ratio would be reduced, since there would be a fewer number of
> larger blocks in the SGA at any given time, thus a lower probability of
> another query finding the same data. This does not make sense to me,
> since in a large database, where many users are submitting queries which
> process thousands of blocks, the cache hit ration is basically meaningless
> for anything but the dictionary and look-up tables - most of the data
> from the large tables will not be retained in cache.
>
> At any rate, is there anyone out there with 'real' experience in tuning
> the db_block size for a 100GB (or larger) database. Any comments would
> be appreciated.
>
> Thanks in advance.

+--------------------------------------------------------------------+
| Anton Dischner, DBA and system-programmer    Phone: +49 89 70953202|
| Institut fuer Klinische Chemie               Fax  : +49 89 70958888|
| Klinikum Grosshadern                         Home : +49 89 6254060 |
| Ludwig Maximilians Universitaet Muenchen     Handy: +49 172 8388880|
| 81366 Muenchen                                                     |
| Marchioninistr. 15                                                 |
| Germany                          Mail: dischner_at_med.uni-muenchen.de|
 +--------------------------------------------------------------------+
Received on Wed Sep 14 1994 - 16:21:38 CEST

Original text of this message