Mailinglist Archive: opensuse-amd64 (470 mails)
| < Previous | Next > |
Re: [suse-amd64] Max FS size?
- From: Andi Kleen <ak@xxxxxxx>
- Date: Thu, 6 May 2004 06:15:01 +0000 (UTC)
- Message-id: <20040506061500.GB8804@xxxxxxxxxxxxx>
On Wed, May 05, 2004 at 01:31:10PM -0500, Kevin_Gassiot@xxxxxxxxxxxxxx wrote:
>
>
>
>
> We have had problems with any filesystem larger than 1 TB. Although I have
> been assured that there are no problems with large filesystems, we can
> create a 1 TB filesystem, and things work fine... create it a megabyte
> over, and it seems to go fine, but when you start writing to the
> filesystem, it starts getting I/O errors. I seem to remember getting the
> answer that fdisk and parted cannot create LUNs larger than 1 TB, so the
> filesystems built on top of these LUNs have problems. I think I was told
> that I would have to create LUNs on the arrays smaller than 1 TB, and use
> LVM to create a logical volume to get the larger size....
>
> I also think this was supposed to be fixed (at least parted, didn't hear
> anything about fdisk) in an upcoming release ? Do you know if this was
> fixed in 9.1 ?
I am not aware of any file system problems at 1TB, and I do not
think anything specific got fixed. They have limits, but they are
much higher than 1TB: 16TB for ext3 and reiserfs, practically unlimited
for XFS and JFS which use 64bit block numbers internally.
I did test some of these with a holey loopback file and it worked
just fine.
What I am aware is that there are a few block drivers that have problems.
The generic block layer is clean in this regard at least on 64bit systems.
On 32bit there is a 2TB limit in 2.4 based kernels; 2.6 supports more
even on 32bit.
You can check if it is the driver by doing some tests without file system.
Use some tool like dd to write a pattern near and crossing the 1TB
boundary to the block device, invalidate the buffers using /sbin/blockdev
--flushbufs and then try to read it back and verify it. When that fails
already it is clearly not the fault of the file system.
Often it is not even the driver to blame, but buggy firmware in the
controller, where a firmware update may help.
One way to work around such driver problems if you are using
a RAID box is to configure LUNs <1TB and then use LVM or MD
RAID to combine them into a single RAID 0 block device at the
Linux side.
-Andi
>
>
>
>
> We have had problems with any filesystem larger than 1 TB. Although I have
> been assured that there are no problems with large filesystems, we can
> create a 1 TB filesystem, and things work fine... create it a megabyte
> over, and it seems to go fine, but when you start writing to the
> filesystem, it starts getting I/O errors. I seem to remember getting the
> answer that fdisk and parted cannot create LUNs larger than 1 TB, so the
> filesystems built on top of these LUNs have problems. I think I was told
> that I would have to create LUNs on the arrays smaller than 1 TB, and use
> LVM to create a logical volume to get the larger size....
>
> I also think this was supposed to be fixed (at least parted, didn't hear
> anything about fdisk) in an upcoming release ? Do you know if this was
> fixed in 9.1 ?
I am not aware of any file system problems at 1TB, and I do not
think anything specific got fixed. They have limits, but they are
much higher than 1TB: 16TB for ext3 and reiserfs, practically unlimited
for XFS and JFS which use 64bit block numbers internally.
I did test some of these with a holey loopback file and it worked
just fine.
What I am aware is that there are a few block drivers that have problems.
The generic block layer is clean in this regard at least on 64bit systems.
On 32bit there is a 2TB limit in 2.4 based kernels; 2.6 supports more
even on 32bit.
You can check if it is the driver by doing some tests without file system.
Use some tool like dd to write a pattern near and crossing the 1TB
boundary to the block device, invalidate the buffers using /sbin/blockdev
--flushbufs and then try to read it back and verify it. When that fails
already it is clearly not the fault of the file system.
Often it is not even the driver to blame, but buggy firmware in the
controller, where a firmware update may help.
One way to work around such driver problems if you are using
a RAID box is to configure LUNs <1TB and then use LVM or MD
RAID to combine them into a single RAID 0 block device at the
Linux side.
-Andi
| < Previous | Next > |