Raid1 + NFS performance. Which filesystem to choose?

Hi!

I’m changing hard disk on my home server (Opensuse 12.2-i586). Now I have one disk which is mountes by three clients through NFS and samba.
I’m going to change disk for 2 HDD-1TB in raid1 software ( mdadm).

I’m reading some things about the filesystem to choose but I’ not so sure. ext4 seems to be fine, xfs seems to be better but I’ve read ( File System Choice for NFS Servers | peacon) that it has issues with mdadm.

I’m not sure about btrfs: With btrfs it’s possible to create the raid1 apart from mdadm, but i’m not sure if it’s better than ext4 and it’s performance with NFS. I’m afraid it’s still a bot experimental.

anyone can tell his experience about it?

regards

Unless you want to experiment I’d stick to ext4.

I was told recently by one of the btrfs advocates that the RAID features of btrfs are not there yet. See the discussion in the beta forum

Hi fperal,

you may want a reliable file system,
so that your files don’t get crunched ?

That’s it.

See

OK, you would like to have max performance.

But can you imagine how that performance breaks down if you have errors in your file system?

I think the answer of gogalthorp was pretty good !

Good luck
Mike

fperal wrote:
> Hi!
>
> I’m changing hard disk on my home server (Opensuse 12.2-i586). Now I
> have one disk which is mountes by three clients through NFS and samba.
> I’m going to change disk for 2 HDD-1TB in raid1 software ( mdadm).
>
> I’m reading some things about the filesystem to choose but I’ not so
> sure. ext4 seems to be fine, xfs seems to be better but I’ve read (
> ‘File System Choice for NFS Servers | peacon’
> (http://blog.peacon.co.uk/file-system-choice-for-nfs-servers/)) that it
> has issues with mdadm.
>
> I’m not sure about btrfs: With btrfs it’s possible to create the raid1
> apart from mdadm, but i’m not sure if it’s better than ext4 and it’s
> performance with NFS. I’m afraid it’s still a bot experimental.
>
> anyone can tell his experience about it?

It all depends on your workload, so nobody can tell you.

I use xfs with mdadm and it works. But it’s slow on part of my workload.
I use ext4 a bit, but it’s even slower. My main workhorse is still
reiser3, which works best for me.

I use XFS, I have never had a problem with it. Also you do not have to worry about the 68 day fsck check that the exts have. And I always have to hold the Y for an hour when an ext3 box crashes or power loss, have not tried ext4.

Dave W

On 10/07/2013 10:46 AM, dwestf wrote:
>
> I use XFS, I have never had a problem with it. Also you do not have to
> worry about the 68 day fsck check that the exts have. And I always have
> to hold the Y for an hour when an ext3 box crashes or power loss, have
> not tried ext4.

You need to investigate the tune2fs command to find out how to change the days
between fsck runs.

Yes, I know that you can change the number of days. Not sure if you can turn if off. But the real problem is that I have never seen ext3 recover from loss of power with out a long manual fsck. Where as I have never seen XFS not come right back up.

Dave

ext4 recovers from power loss with no problem

On 2013-10-15 14:46, dwestf wrote:
>
> Yes, I know that you can change the number of days. Not sure if you can
> turn if off. But the real problem is that I have never seen ext3
> recover from loss of power with out a long manual fsck. Where as I have
> never seen XFS not come right back up.

I did, once.

I fact I have seen all with a royal class crash: reiserfs, ext3, and
xfs, on different occasions.


Cheers / Saludos,

Carlos E. R.
(from 11.4, with Evergreen, x86_64 “Celadon” (Minas Tirith))