Re: Sophisticated Performance Tuning: Need Wizard!

From: Virag Saksena <vsaksena_at_us.oracle.com>
Date: 1996/01/15
Message-ID: <4dcfb0$gpu_at_inet-nntp-gw-1.us.oracle.com>#1/1


You have to determine what is causing your INSERTs to be slow. What is your transaction rate ? What is the application waiting for ? Have you verified your assumption that the DBWR is the bottleneck. If the application is waiting for the DBWR to catch up then you'd see a lot of free buffer waits.

The _db_block_write_batch is still used as one of the mechanisms for triggering the DBWR to write dirty buffers but increasing it will not cause you to go to async io. I would not recommend playing with it unless you really know what you are doing and can balance it with _db_block_max_scan_cnt Among the side effects you can get because of playing with it are severe latch contention or IO bottlenecks.
If DBWR is a bottleneck then you can get performance gains without going to raw by using multiple DBWRs.

Virag


Virag Saksena                                         vsaksena_at_us.oracle.com
Senior Consultant                                     415.506.5087
System Performance Group
Oracle Services

In article <4bo6m6$s6a_at_aladdin.iii.org.tw> Morris Chen <root_at_corolla.tatung.com.tw> writes:
>Hi Wizards,
> We recently have to tune the fat ORACLE 7.1 on DEC Server 2100(alpha)
>/OSF1 3.0 and have the following questions:
>
> 1. We currently use file system and find it's very slow to INSERT,
> any techniques found on ORACLE manuals are carried out but in vain,
> should I 'kill' the UNIX buffer cache to bypass ORACLE synchronous
> write or decrease the UNIX system flush rate?
> 2. How can I know the exact size of the System Global Area that will
> not be paged out? Can ORACLE use any 'resident' shared memory like
> Informix?
> 3. ORACLE 6.* use (db_block_write_batch/2) as a 'high water mark'
> (threshold) for db writer to write dirty database blocks to disks.
> Is this still true for ORACLE 7.1? BTW, if I set db_block_write_batch
> large will Oracle automatically use asynchonos(parallel) I/O?
>
>ANY HINTS ARE APPRECIATED. Sorry for my poor Taiwanese English.
>
> Morris Chen
> ORACLE keeper
> TATUNG COMPANY (over 25,000 employees)
> root_at_corolla.tatung.com.tw
> 1995. 12. 26.
>--------------------------------
>The service of Oracle Taiwan Inc. is NO SERVICE. (:-<~ #_at_$%!
>
>
>
Received on Mon Jan 15 1996 - 00:00:00 CET

Original text of this message