[pts] Ext4 vs Ext3, ext4 very very slow with sqlite

Eric Sandeen sandeen at redhat.com
Wed Mar 11 23:38:10 CDT 2009


Bryan Quigley wrote:
> Eric,
> What were the system specs/distro you used.  How did you format it ext4/3? 

I formatted & mounted w/ defaults, and tested on an upstream kernel on
Fedora.

However, the storage I used was a raid controller on a SAN, and it
probably ignored barrier requests.

I just tested again on a local disk, and it does indeed seem to be
barriers which cause the difference.

> I did fresh Ubuntu alpha installs using the whole disk with the
> filesystem in question (no swap). Sqlite results in seconds
> Ext4: 1690.87
> Ext3: 108.77
> Ext2: 50.67
> Reiser: 105.23
> Xfs: 1196.03
> Jfs: I must have forgotten to upload this result :(

ext4 & xfs have barriers on by default; the others do not.  So this
theory makes sense.  Try mounting with -o barrier=0 for ext4, or -o
nobarrier for xfs, and see what you get.  For ext4 you can do this with
a remount; for xfs I think remount may not work.

(BTW, how does one tell phoronix to do filesystem tests on a particular
filesystem?  I just modified the sqlite tests to run the db on some
other mountpoint but surely there is a better way....)

> I also did a Fedora Alpha install using the liveCd but couldn't actually
> get it to match the above disk config.  (It wanted to set up an LVM, and
> didn't seem to like me manually specifying ext4)

yes, that was a hiccup with the alpha livecd.

-Eric

> -Bryan
> 
>     Just ran my own tests and did not see any such massive slowdown,
>     although ext4 did clock in a bit slower, which is still interesting.
> 
>     Mine was on the order of 65s for ext4, 50s for ext3.  Barriers on either
>     fs didn't make a huge difference.  xfs (w/ and w/o barriers) came in at
>     ~45s just FWIW.
> 
>     -Eric
> 
> 




More information about the trondheim-pts mailing list