Linux's benchmarks & 3dmark

dan17 · 2008-03-05 01:07:53

Hi,

I would like to share with you the firsts results of the benchmarks I have done with 'many' different linux distros plus BSD distros and XP for the first test.
Prehaps, you can confirm or not, I steel open.

I'have installed the OS'es as they come - no 'tactical' optimization.

MySQL
********
My first tests were done with MySQL and a database of 1Gb (4 me is enough big).
For this test, I had to deal with the systems files too.
I made a 'simple query' and the better results were for XP (8s), next Fedora (12,7s), Ubuntu, OpenSuse and the last FreeBSD with a notable difference.
Even then probing all the different Linux/Unix filesystems (ext2, ext3, reiserfs, xfs, jfs), XP obtains the first place.
The tests I would like to do are :
- to tune the unix fs clusters sizes,
- put the database on ntfs partition because In my opinion, I think that the difference of performance is due to the system file.

3dMark under Wine+Wine-doors
*************************************
Slackware 12 - 7750 (added March 9)
OpenSuse 10.3 - 7415
Ubuntu 7.10 - 7397
Arch Core - 6281

Regarding the memory usage, Arch is the distro that uses the less memory. memory usage ordering : (Fedora>OpenSuse>Ubuntu>FreeBSD).

So, I would like to understand why Arch is not obtaining the best results because - if I understood - the binary packages are compiled for i686 ?

I hope that nobody will be felt offended, it is not my intention - just identify/understand the differences.

Regards,
Dan17

Last edited by dan17 (2008-03-09 00:25:32)

schivmeister · 2008-03-05 03:38:16

Heh, the Arch system used for the benchmark is a failure.

joephantom · 2008-03-05 05:24:11

¿All the OSs were installed in the same hard drive when you did the benchmarks?

Last edited by joephantom (2008-03-05 05:47:17)

dan17 · 2008-03-05 21:27:27

Hi,

Yes same machine, no changes.

Dan17

vogt · 2008-03-05 22:58:09

dan17 wrote:

MySQL
********
My first tests were done with MySQL and a database of 1Gb (4 me is enough big).
For this test, I had to deal with the systems files too.
I made a 'simple query' and the better results were for XP (8s), next Fedora (12,7s), Ubuntu, OpenSuse and the last FreeBSD with a notable difference.
Even then probing all the different Linux/Unix filesystems (ext2, ext3, reiserfs, xfs, jfs), XP obtains the first place.
The tests I would like to do are :
- to tune the unix fs clusters sizes,
- put the database on ntfs partition because In my opinion, I think that the difference of performance is due to the system file.

So you had the database on one ntfs filesystem (the xp driver is probably better than ntfs-3g...)?

Were these operating systems installed concurrently, or was each installed as the only operating system at the time? harddisk read speed does fall off towards the end of the disk (higher numbered partitions).

EDIT: the 3d mark discrepancy is strange: if you use the same drivers, ubuntu should be the same as arch...

Last edited by vogt (2008-03-05 23:00:15)

dan17 · 2008-03-06 00:40:35

Hi,

Only one OS at a time - no multiboot. All versions have been tested in the same conditions.

>EDIT: the 3d mark discrepancy is strange: if you use the same drivers, ubuntu should be the same as arch...

That's why I post:-)

A++
Dan

ramiblanco · 2008-03-20 07:15:03

which wine version did you use on each? video driver?

brebs · 2008-03-20 08:37:48

The differences will be hidden in hundreds of patches and different software versions. Including the gcc version, gcc patches and compiler options used to compile each package. So not exactly easy to track down

freakcode · 2008-03-20 15:52:26

brebs wrote:

The differences will be hidden in hundreds of patches and different software versions. Including the gcc version, gcc patches and compiler options used to compile each package. So not exactly easy to track down

Yep. There must be patches, Ubuntu AFAIK uses a bunch of them. I wouldn't doubt there are some to better Wine DX performance, as lot of *buntu users should use Wine to play and that stuff.

Regarding MySQL performance on XP, the test you did was mostly disk intensive. What I can think from my head is that either NTFS is faster than journalized FS from Linux (which I doubt), or your XP might give you better disk I/O than Linux because you probably installed the manufacturer driver for your motherboard on XP, while Linux might be using some generic driver.

Anyway... Arch won't top unless you tweak it. Despite being i686, I guess most configs come more default/plain than in other distros, and that gives you the difference from a fresh install.

Last edited by freakcode (2008-03-20 15:55:23)

brebs · 2008-03-20 16:20:05

It's not just wine, it's every package - from the kernel upwards.

fwojciec · 2008-03-20 16:25:48

My guess would be that kernel version is likely to be the biggest factor... There were big changes introduced in the 2.6.23 line of kernels (new scheduler, etc.) -- AFAIK Ubuntu (Gutsy) uses 2.6.22 kernel, not sure about Slackware, but I'd expect something similar.

iBertus · 2008-03-21 01:50:21

Perhaps the difference is in how Ubuntu auto configures X settings compared to Arch. Could you have missed something in your X config?

cactus · 2008-03-21 02:22:13

i am venturing a guess here, but I wonder if the major vendors ship with default configs. You said *you* didn't do any custom configuration modifications..but make sure the configs *as shipped* are actually the same across the tests... as well as make sure the versions of each app tested are the same (or close enough).

Otherwise there is such a variance in the tests that it really is hard to get much in the way of extrapolation out of them.

"There are three kinds of lies: lies, damned lies, and statistics." -- Mark Twain

dan17 · 2008-03-23 00:49:58

Hi,

Regarding 3DMark, I m pretty sure having the same config for Xorg. The main difference can come from the kernel version (and the scheduler).

So, the remark regarding the hd controler driver for the mother board is a good point. I'll investigate this subject.

Regarding, the Mark Twain citation, I agree but ly was not my goal - just open a debate and "investigate"

Regards;

Arch Linux

#1 2008-03-05 01:07:53

Linux's benchmarks & 3dmark

#2 2008-03-05 03:38:16

Re: Linux's benchmarks & 3dmark

#3 2008-03-05 05:24:11

Re: Linux's benchmarks & 3dmark

#4 2008-03-05 21:27:27

Re: Linux's benchmarks & 3dmark

#5 2008-03-05 22:58:09

Re: Linux's benchmarks & 3dmark

#6 2008-03-06 00:40:35

Re: Linux's benchmarks & 3dmark

#7 2008-03-20 07:15:03

Re: Linux's benchmarks & 3dmark

#8 2008-03-20 08:37:48

Re: Linux's benchmarks & 3dmark

#9 2008-03-20 15:52:26

Re: Linux's benchmarks & 3dmark

#10 2008-03-20 16:20:05

Re: Linux's benchmarks & 3dmark

#11 2008-03-20 16:25:48

Re: Linux's benchmarks & 3dmark

#12 2008-03-21 01:50:21

Re: Linux's benchmarks & 3dmark

#13 2008-03-21 02:22:13

Re: Linux's benchmarks & 3dmark

#14 2008-03-23 00:49:58

Re: Linux's benchmarks & 3dmark

Board footer