RE: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

RE: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Jean-Eric Cuendet]

> We use LVM + Reiserfs on 2 production boxes. One is a 5x36 gig Raid 5
> (Dell (AMI) PERC Controller) while the other is just two 18 gig drives
> LVMed together. So far there have been no problems.

So you have a HARD Raid5 controller with LVM and Reiser?
Did you already had a crash? How was the filesystem after that?

-jec

Re: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Patrick Boutilier]

Jean-Eric Cuendet wrote:

> > We use LVM + Reiserfs on 2 production boxes. One is a 5x36 gig Raid 5
> > (Dell (AMI) PERC Controller) while the other is just two 18 gig drives
> > LVMed together. So far there have been no problems.
>
> So you have a HARD Raid5 controller with LVM and Reiser?
> Did you already had a crash? How was the filesystem after that?
>
> -jec

We had one crash on the box with the HARD Raid5 controller and the file
systems were fine when it came backup.

We have had more crashes on the box with the 2 drives and the filesystems
have also been fine.

Re: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Stephen C. Tweedie]

Hi,

On Mon, Jul 03, 2000 at 06:16:27PM +0200, Andi Kleen wrote:
> On Mon, Jul 03, 2000 at 03:05:24PM +0100, Stephen C. Tweedie wrote:
> > The only known problem with journaling and software raid is to do with
> > the raid array rebuild code in raid1/5.  The rebuild could cause I/O
> > to be done in the wrong order for the filesystem, causing clean blocks
> > in the buffer cache to be flushed to disk as part of the rebuild.
> > That is fixed in the 2.4 soft raid.  raid0 and LVM were never affected
> > by this as they have no rebuild operation.
> 
> Note this applies only to the RAID1 in Raid 0.90 or in 2.4.
> Unpatched 2.2 does Raid1 reconstruction in user space only and should be 
> safe for journaling.

I think the user-space rebuild still has the same problem --- it
accesses the underlying raid device through the buffer cache, causing
clean buffers to be written to disk.  That's enough to kill
journaling.

Cheers,
 Stephen

Re: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Andi Kleen]

On Mon, Jul 03, 2000 at 03:05:24PM +0100, Stephen C. Tweedie wrote:
> The only known problem with journaling and software raid is to do with
> the raid array rebuild code in raid1/5.  The rebuild could cause I/O
> to be done in the wrong order for the filesystem, causing clean blocks
> in the buffer cache to be flushed to disk as part of the rebuild.
> That is fixed in the 2.4 soft raid.  raid0 and LVM were never affected
> by this as they have no rebuild operation.

Note this applies only to the RAID1 in Raid 0.90 or in 2.4.
Unpatched 2.2 does Raid1 reconstruction in user space only and should be 
safe for journaling.


-Andi

Re: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Stephen C. Tweedie]

Hi,

On Wed, Jun 28, 2000 at 11:21:30AM +0200, Jean-Eric Cuendet wrote:
> 
> > > 	Currently you cannot put journaled filesystems on a software 
> > > raid I believe, because they use the buffer cache in incompatible 
> > > ways.  If I am wrong, someone correct me.  Ext2 (what I use) works
> > > fine. 
> > 
> > I'm using etx3 on top of 20GB raid0 for over an year with no 
> > problem...
> 
> In fact, if you have ext3 and a crash, in the worst case, journal will be
> out of sync (In the case of writes were not like the filesystems thought due
> to LVM)

No, LVM never lies in any way which will confuse a journaling
filesystem.  The block remapping which happens inside LVM does not
result, ever, in early I/O completion notifications.

The only known problem with journaling and software raid is to do with
the raid array rebuild code in raid1/5.  The rebuild could cause I/O
to be done in the wrong order for the filesystem, causing clean blocks
in the buffer cache to be flushed to disk as part of the rebuild.
That is fixed in the 2.4 soft raid.  raid0 and LVM were never affected
by this as they have no rebuild operation.

Cheers,
 Stephen

RE: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Jean-Eric Cuendet]

> > 	Currently you cannot put journaled filesystems on a software 
> > raid I believe, because they use the buffer cache in incompatible 
> > ways.  If I am wrong, someone correct me.  Ext2 (what I use) works
> > fine. 
> 
> I'm using etx3 on top of 20GB raid0 for over an year with no 
> problem...

In fact, if you have ext3 and a crash, in the worst case, journal will be
out of sync (In the case of writes were not like the filesystems thought due
to LVM)
Then you'll have a 15 minutes fsck (like if it was ext2).
In the best case, journal is OK, and you have a 3 seconds fsck which is
fine.

Agree?

Bye
-jec

Re: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Patrick Boutilier]

[-- Attachment #1: Type: text/plain, Size: 960 bytes --]

We use LVM + Reiserfs on 2 production boxes. One is a 5x36 gig Raid 5
(Dell (AMI) PERC Controller) while the other is just two 18 gig drives
LVMed together. So far there have been no problems.

Jean-Eric Cuendet wrote:

> Hi,
> I'd like to build such a system:
> - Kernel + root on non-LVM Raid0/ext2 partition (no problemo)
> A big partition (100Gb) which lokks like:
> - Raid5 (linux md)
> - LVM
> - filesystem
>
> It's a classic schema, but my question is:
> Does anyone already tried it?
> Any success? failure? Drawback?
> Does XFS + LVM or JFS + LVM or ReiserFS + LVM be stable enough for
> production?
> Do they all have resize tools?
>
> Thanks
> -jec
>
> _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
> Jean-Eric Cuendet
> Linkvest SA
> Av des Baumettes 19, 1020 Renens Switzerland
> Tel +41 21 632 9043  Fax +41 21 632 9090
> http://www.linkvest.com  E-mail: jean-eric.cuendet@linkvest.com
> _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/x-pkcs7-signature, Size: 1941 bytes --]

Re: [linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Brian Kress]

I have almost exactly this system.  ( I don't have the
raid0 boot partition)  Works great.


Jean-Eric Cuendet wrote:
> 
> Hi,
> I'd like to build such a system:
> - Kernel + root on non-LVM Raid0/ext2 partition (no problemo)

	This gets a little tricky.  You need both disks BIOS
addressable (preferably hda and hdb or sda and sdb).  You also
need a LILO that speaks Raid0.  I think RH6.2 comes with one.

> A big partition (100Gb) which lokks like:
> - Raid5 (linux md)
> - LVM
> - filesystem
> 
> It's a classic schema, but my question is:
> Does anyone already tried it?

	/dev/md0 is a raid5 of /dev/hdb, /dev/hde, /dev/hdg, and 
/dev/hdl.  /dev/md0 is a PV for a VG that has 7 LVs in it.  I 
even have root in LVM.  (this requires MD autodetect and 
an initrd for LVM) 

> Any success? failure? Drawback?
> Does XFS + LVM or JFS + LVM or ReiserFS + LVM be stable enough for
> production?

	Currently you cannot put journaled filesystems on a software 
raid I believe, because they use the buffer cache in incompatible 
ways.  If I am wrong, someone correct me.  Ext2 (what I use) works
fine. 
	Raid 0.90 and LVM are both quite stable for production use.

> Do they all have resize tools?

	Ext2 has one I think.  I don't generally resize LVs.


	If you'd like any help setting this up, let me know.


Brian Kress
kressb@icp.siemens.com

[linux-lvm] ReiserFS/XFS/JFS, LVM, Raid5

[Jean-Eric Cuendet]

Hi,
I'd like to build such a system:
- Kernel + root on non-LVM Raid0/ext2 partition (no problemo)
A big partition (100Gb) which lokks like:
- Raid5 (linux md)
- LVM
- filesystem

It's a classic schema, but my question is:
Does anyone already tried it?
Any success? failure? Drawback?
Does XFS + LVM or JFS + LVM or ReiserFS + LVM be stable enough for
production?
Do they all have resize tools?

Thanks
-jec

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
Jean-Eric Cuendet
Linkvest SA
Av des Baumettes 19, 1020 Renens Switzerland
Tel +41 21 632 9043  Fax +41 21 632 9090
http://www.linkvest.com  E-mail: jean-eric.cuendet@linkvest.com
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _