Re: ORACLE AND UNIX QUESTION

Steve - Schow <sjs_at_shell.portal.com> wrote:
>We did one test with two very similar machines, one using Raw devices and
>the other using UFS files and when we monitored disk i/o from UNIX, we
>noticed that the Raw device was only performing about 30% the number of
>writes that th UFS disk was.

How were they different? Same Controller Types, Disks ...

>This is probably due to inode updating and so forth.

Not unless you keep your news directory in the same fs as your database files. If your in a filesystem with only datafiles I would have expected very little update to the inodes and no additional lookups after everything was cached on startup.

>Still, you can't ignore the fact that there were 1/3 the number of disk writes
>with the raw device. Plus UNIX buffer cache was not being used, which means
>more memory available for Oracle buffer cache.

Ok fair enough. But what caused the increase. different blocksizes? unbalanced file distribution? init.ora parameters? differenct Kernel settings.

I'm not trying to argue with you. But as you noted 2/3 is a big number. And I have trouble beleiveing that this behavior is inherent to UFS I can beleive these number for an out of the box UFS but one that has been tuned for a Oracle database?

Am I wrong? Any body else seen this in a side by side test.

-- 
Regards, 

Lee E. Parsons                  		
Systems Oracle DBA	 			lparsons_at_world.std.com