Comparing the filesystems

[milomak]

Finding a comparison is very difficult and confusing since everyone uses different tests and arguments.

One such failure of most online tests is that they just test the dead standard default file system, no optimization. For xfs this is a HUGE drawback since by default it is a bit useless.

What would be really nice if someone here volunteered to do a test comparing ext3, reiserfs, reiserfs4, ext4 and xfs. I think those are the mayor contenders on the desktop today. This way we can have an even comparison and tweak the tests as we see fit for everyday use and have real world results in performance.

So let's lay down how we want to do this. i have some partitions available for testing. I can do the tests using one of gentoo, debian, or arch. What you will have to give me are the parameters:
1. mount options for each of the partitions
2. which kernels to test
3. kernel options (provide me the .config files to compile the kernels)
4. testing methods – will probably use the phoronix tools

There are probably other things I haven’t thought of. So let’s flesh it out. Once we are in agreement I’ll start the test.

 

[risingtide]

This was recently posted.
http://www.ibphoenix.com/main.nfs?a=ibphoenix&s=1264048227:903849&page=ibp_file_linux_firebird
I don't know if that is of any general help or only applies to Firebird databases.

 

[MyWorld]

I'll gather a bit of data on file system optimization later tonight.

I had a long hard though about this, what we need is a pure test, in other words, etx3 copy to ext3 paste, etc. I have no idea if this is a factor but it sounds logical to me to do it this way.
This will complicate things a bit since you now have to have 2 partitions of said file system almost maxing out your entire partition table (ext3, ext4, xfs, reiserfs, reiserfs4 = 10 partitions). This means that the hd/hdds used should be a spare or free one lying around.
Also, are we going to use one single hd? Or do a hd to hd test to get more accurate read/write figures?

What I was hoping for is that we could create a test bed and later on as arguments against the testing comes in, then we could easily change a few parameters and test again, but this is going to be nigh impossible for us normal desktop users.

Anyway, have a look at iozone, seems like the tool for the job.

 

[MyWorld]

Here is optimization for
Ext3:
http://en.opensuse.org/Speeding_up_Ext3

Ext4:
It seems the only option is the nobarrier option but in a real world usage that would be very unwise.

More reading here:
http://ext4.wiki.kernel.org/index.php/Ext4_Howto

xfs:
mkfs.xfs -l size=64m
mount -o noatime,nodiratime,logbufs=8

Some more reading:
http://dforbes.net/?p=21
http://everything2.com/index.pl?node_id=1479435

Reiserfs:
The following needs to be set in fstab:
data=writeback,noatime,nodiratime,notail

Why?
http://www.ubuntugeek.com/how-to-increase-ext3-and-reiserfs-filesystems-performance.html

Reiserfs4:
The options here seems to be:
defaults,noatime,nodiratime


If anyone have some other options, then I'm all ears since it seems that there are many, many options to sift through, like here:
http://linux.die.net/man/8/mount

 

[MyWorld]

Is this still on?
I need to get myself a new hd in the coming months, only then will I be able to try anything sporty.

 

[milomak]

my free partitions are on the same hdd. i'm still dealing with a newborn so my computer time is severly limited.

I'll be formatting each partitoin with mkfs.<partition type> /dev/<partition>