CentOS 5.4 vs. OpenSuSE 11.2 vs. Ubuntu 9.10 Benchmarks

[deanjo]

Thanks Deanjo. It seems (K)ubuntu also uses ordered data mode and I'll test writeback to see if it will make a difference in SQLite.

Barriers enabled does seem to have quite an effect with SQLite. You might want to check that as well. IIRC though (K)ubuntu does not have it enabled where openSUSE does.

[kraftman]

openSUSE uses barriers enabled, file extents enabled, delayed allocation enabled, mballoc enabled, ordered data mode. internal journal by default on ext4.

Thanks Deanjo. It seems (K)ubuntu also uses ordered data mode and I'll test writeback to see if it will make a difference in SQLite.

[deanjo]

What fs modes do those distros use? If CentOS uses writeback and other use ordered it probably can explain such difference in SQLite.

openSUSE uses barriers enabled, file extents enabled, delayed allocation enabled, mballoc enabled, ordered data mode. internal journal by default on ext4.

[energyman]

why gnome?

[kraftman]

What fs modes do those distros use? If CentOS uses writeback and other use ordered it probably can explain such difference in SQLite.

[h1d.]

sqlite

So, on what kernel version and file system does sqlite not get bogged down?

I'm planning to do a clean install of 9.10 Ubuntu server and sqlite will be used frequently on that.

I have upgraded one of the machine to use 9.10 server from 9.04 server, which kept the ext3 partition but the phoronix-test-suite went pretty slow on sqlite afaik, something like 150 seconds per test on a ssd with 120+mb/s read and probably half that for write.

[deanjo]

You might also want to disable AppArmor in the future as well since you disabled SELinux on CentOS and gives it a bit of an advantage.

[deanjo]

Guys,

PLEASE start filing (regression) bug reports in http://bugzilla.kernel.org/ otherwise your tests are worth nothing.

Actually that's not phoronix's job, that is the responsibility first of the distro's community to see if it is truly a kernel regression or a misconfiguration as their configs vary distro to distro. Regressions should only be submitted when running a generic kernel and most distro's do some customization to the kernel and do not use the same configs. If it is found that it is actually a kernel regression then the distro can always send it upstream.

[deanjo]

Also,

OpenSuSE 11.2 RC1 runs with the Linux 2.6.31 kernel, GNOME 2.28.0, X Server 1.6.5, NVIDIA 190.29 display driver, GCC 4.4, and uses the newer EXT4 file-system by default.

This isn't true, the default desktop is KDE 4.3.1. Gnome is a selection though.

[birdie]

Guys,

PLEASE start filing (regression) bug reports in http://bugzilla.kernel.org/ otherwise your tests are worth nothing.