Which File System ext3 or ext4

[spectra]

Folks,
In some of the newer Linux releases, when setting up the filesystem they default to ext4. Such is the case of Ubuntu 10 0 4 LTS. Further reading suggest benefits of the ext4 filesystem.

Now, the zimbra FOSS 7 documentation says ext3.

Should I be in conflict ? Should I go with ext3 which is an older filesystem for a newer Linux server or ext4 newer filesystem and benefits of the new ext4.

So which one should be used ?

thanks

[phoenix]

Originally Posted by spectra

So which one should be used ?

You can use any filesystem you choose.

[LHammonds]

Originally Posted by spectra

In some of the newer Linux releases, when setting up the filesystem they default to ext4. Such is the case of Ubuntu 10 0 4 LTS. Further reading suggest benefits of the ext4 filesystem.

Now, the zimbra FOSS 7 documentation says ext3.

Should I be in conflict ? Should I go with ext3 which is an older filesystem for a newer Linux server or ext4 newer filesystem and benefits of the new ext4.

So which one should be used ?

I think you missed Zimbra's Ubuntu-specific guide for configuring the storage system --> Ubuntu Server Config 7.1.pdf

That document clearly shows them saying to use ext4. The only other difference is the boot partition using ext2...for compatibility reasons.

I didn't like the recommendation to have EVERYTHING stored in the root because your /opt/zimbra folder will continually grow and grow. If it consumes all free space on your system, it could make the OS completely unresponsive since the root file system is sharing its space with /opt as well.

Take a look at my installation notes to see how I designed my layout (also has a visual diagram). My design allows for flexible growth as well as automated growth of file systems with notifications (using scripts) so you can order more hard drive space far in advance of actually needing it.

LHammonds

[lytledd]

I use XFS and use the built-in support for snapshots. At this time, I don't trust ext4

Doug

[vavai]

Originally Posted by lytledd

I use XFS and use the built-in support for snapshots. At this time, I don't trust ext4

Doug

Any reason why you don't trust ext4? It's not quite stable, not enough mature or maybe experience problem regarding ext4?

Actually, I'm using ext3 on SLES 11 SP1 and ext4 on Ubuntu 10.04 LTS simply because ext3 is the default partition on SLES 11 SP1 as well as ext4 on Ubuntu 10.04

[lytledd]

Any reason why you don't trust ext4? It's not quite stable, not enough mature

A little bit of both. I'm on several mailing lists that bring the subject occasionally, where the fix has been to move from ext4 to 3. Mostly to do with databases and how the partition is mounted. I've also read articles that always are saying, "It's almost as good" or, "It's catching up".

Since I've been using XFS for years, I see no compelling reason to move.

Before that, I was using RiserFS.

Doug

[LHammonds]

I used LVM to overcome the lack of snapshot support in Ext4. It may not be the best performance option for database servers but I have powerful enough systems that can overcome that problem. My main goal was the ability to automate certain tasks to allow human intervention only when necessary...such as the system automatically expanding the file system when needed and knowing when expansion is getting close enough that administration needs to be notified that more storage needs to be purchased and configured. LVM allowed me to do that.

Here are some good articles to read up on the matter:

Ext3 vs Ext4 vs XFX vs Btrfs
Wikipedia articles: Ext3, Ext4, XFS, Btrfs, Comparison of File Systems

[bofh]

Ahm First
Dont use any Filesystem you want if you use virtualisation

in addition if you use ubuntu in a kvm enviroment DO NOT USE EXT4 i repeate do not use it

Do not use it for the host - where your image reside
do not use it on the guest - where zimbra lives

this is realted to an IO Problem
Also xfs is sorta problem


it would be to much to explain it here completely but oyu can google that issues
its about how kvm handle io request and how does ext4

Soo since you should use virtualisation - you should use ext3


in addition peformance tip for ubuntu - ubuntu has some kind aweird IO problem as a guest anyway spcially for cacheds reads

so best you can do is - set guest cache to NONE
do not use writeback or writhe trough
do not use native - use threads

thers another bug if you use native which causes filesystem corruption if the imageblocksize isnt the same as the host blocksize

lvm is an alternative - you could pipe it through - howevery im personally not a frind of this - is kinda wierd virtualisation if i use a real partition - but ok this might be also kinda thing of taste

[bofh]

PS: ext4 is overrated
its useful for desktops but for server its to young

and its already predicted to be EOL
many people and distributions think the future will be in btrfs

so the real jump should be form ext3 to btrfs

however btrfs is still beta and work on it goes on for years now - so dont think its useable in production for at least antoher 2-3 years

btw ext4 isnt nessesarly faster than ext3 - still kernel development going on and since its first prod. release its getting slower because of new safegards in the kernel

[LHammonds]

I'm running Ubuntu Server 10.04 LTS on top of an ESXi 4.1 server (IBM 7870AC1 Blade). The data is stored on a DS3400 storage system running SAS drives in RAID5 and connected via a FiberChannel.

The file systems I have configured are Ext4 encased in LVM. Design can be seen in this image.

To get a quick performance test, I run this command:

hdparm -t /dev/sda

And the results are:

Timing buffered disk reads: 1320 MB in 3.01 seconds = 438.96 MB/sec

And that is about the average performance...sometimes a bit better or a bet less but not by much. It's not bad and I have yet to run into any problems. Been running a Zimbra and mysql servers for a couple of months now.

EDIT:

On my test servers, I use the same kind of IBM blade, but a different storage system...IBM DS3200 using slower SATA drives in RAID10 configuration. The performance on that system is not nearly as nice as the SAS system, however, ALL my servers are sending every byte written to their SAS hard drives to the remote SATA drives (offsite server sync/backup)...so they tend to stay busy all the time but performance can vary drastically depending on the global amount of data going on. Here are the results of the same test run on a server using the SATA storage system:

The average score of the best results (during off-hours):

Timing buffered disk reads: 1712 MB in 3.00 seconds = 558.33 MB/sec

The average score of the lowest results (during peek hours):

Timing buffered disk reads: 244 MB in 3.03 seconds = 80.53 MB/sec

EDIT: Installed Debian 6.0.3, here are the hdparm results:
The average score (during peek hours):

Timing buffered disk reads: 2020 MB in 3.00 seconds = 672.81 MB/sec

LHammonds