Re: Asynchronous IO (UNIX)

I recently had the opportunity to play with asynch_io on a sequent se20 platform. You are correct in your assumpition that asynch_io needs to use raw disk partitions. According to the Oracle documentation for the sequent platform, you need to make a few kernel modifications, set asynch_io to true in the init.ora file, and limit the # of db_writers to one. However, performance was about 25% worse in this configuration than using unix file systems, WITHOUT the asynch_io kernel modifications and 32 db_writers.

results may vary on other systems, but it seems that on the sequent (with 8 pentiums and 4 scsi channels) file systems with 32 db_writers worked exceptionally well.

In article <44ba6u$7nk_at_mailhost.qsp.co.uk>, Steve_Dodsworth_at_qsp.co.uk says...
>
>Hello All,
>
>After reading numerous books and having countless arguments, I assume the
 following to be true.
>
>If the unix operating system supports asynchronous IO then set an
init.ora flag called use_async_io to true ONLY if the datafiles etc are to be stored on
>
>If the datafiles are stored on 'normal' file systems then set this flag
to false (default) and instead set at least as many db_writers as there are disks
>
>An exception to this is AIX, which can use kernal settings to achieve
asynchronous IO. In this case set use_async_io to true and db_writers to 1.
>
>
>If anyone knows any different or indeed knows any more exceptions - can
 they please reply
>Thanks
>
>Steve Dodsworth
>
>
> T.C.
> ___________________________________________________________
>| |
>| All opinions expressed are my own and should not be |
>| taken as being the opinions of my employer. |
>| I accept full responsibility for my thoughts and actions. |
>|___________________________________________________________|
> disclaimer-in-a-box
Received on Mon Oct 02 1995 - 00:00:00 CET