On Fri, 29 Nov 2002 00:25:01 +0100, Rick Denoire
>I am in the process of setting up an Oracle DB V 9i on Redhat Linux AS
>2.1. In this case we are dealing with a rather small number of files
>having mostly a large size, some of them are being written
>sequentially (redo logs), and some almost constantly (control files).
>So the situation seems like an ideal case to use a file system that
>best suites to it. Since one can choose the file system type among a
>number of these using Linux, I am wondering which one is the more
>appropriate or offers the best performance. Besides ext3 there is
>reiserfs and xfs and...
>Any experience or recommendations?
>The most important DB files are located in a (hardware) RAID 10 (5x2
>HDs), but I think the file system does not care about that.
How brave are you feeling? You could always use raw devices. But I
wouldn't recommend it unless you're after ultimate performance.
Personally, I'd use ext2 file systems, but reconfigure them with
biggest blocksize and frag size ( -b 4096 -f 1024 ), cut the reserved
space right down ( -m 1 ), and set the number of inodes right down (
-N <100 times the number of files you expect to store there> ). For
oracle, don't forget that there's bdump and udump dirs that can have a
lot of files if tracing is switched on, so don't go too mad on the
number of inodes 'cos you can't change it later!
I expect that a number of people will suggest ext3 ( is it available
with AS? ), but, although it'll be much faster to recover after a
crash, I don't really have a feel for the performance overhead of the