Fedora Logical Volume Manager Benchmarks

[chithanh]

I wonder why the article then claims that

Fedora doesn't use write barriers with LVM

A check of dmesg should decide the matter.

[yesterday]

What a typical phoronix thread. An unsubstaniated comment about a distribution accepted as fact to fuel nerdrage

[liam]

I wonder why the article then claims thatA check of dmesg should decide the matter.

I don't care what the article said, I'm telling you what the fedora site said. I trust the guys who wrote this stuff more than Michael, in this case, especially since I haven't been able to find any evidence that fedora disables barriers at any level.
For the record, this F15 laptop says

EXT4-fs (dm-1): mounted filesystem with ordered data mode

[gilboa]

Does anyone has any type of official link stating that Fedora 15 indeed doesn't support write barriers on ext4 over LVM?
(I have no problems remounting ext4 with mount / -o remount,barrier; by default ext4 is mounted in ordered mode)

- Gilboa

[energyman]

I don't care what the article said, I'm telling you what the fedora site said. I trust the guys who wrote this stuff more than Michael, in this case, especially since I haven't been able to find any evidence that fedora disables barriers at any level.
For the record, this F15 laptop says

EXT4-fs (dm-1): mounted filesystem with ordered data mode

you are looking at the wrong stuff. Even if ext4 tries to use barriers if lvm does not support them you are SOL.

Does anyone has any type of official link stating that Fedora 15 indeed doesn't support write barriers on ext4 over LVM?
(I have no problems remounting ext4 with mount / -o remount,barrier; by default ext4 is mounted in ordered mode)

- Gilboa

which will do nothing, if lvm does not support barriers.

[liam]

you are looking at the wrong stuff. Even if ext4 tries to use barriers if lvm does not support them you are SOL.



which will do nothing, if lvm does not support barriers.

Unless I'm mistaken that WAS the lvm message (EXT4-fs) since lvm is the one who would be responsible for the device management (hence dm).
If you can find some documentation that shows otherwise and where I should look I would appreciate it.

[energyman]

>EXT4-fs (dm-1): mounted filesystem with ordered data mode

is the mount message from ext4. Filesystem, underlying device, mount options.

Nothing from lvm.

[liam]

>EXT4-fs (dm-1): mounted filesystem with ordered data mode

is the mount message from ext4. Filesystem, underlying device, mount options.

Nothing from lvm.

That's my point: the message should be from LVM since it is doing the management of said device.
Per this message, https://bugzilla.kernel.org/show_bug.cgi?id=9554#c0, syslog should report if barrier's are enabled but not supported by lvm.
I don't see the message, q.e.d., error not ocuring.
Do you have any link to documentation, or command I can run, that would say otherwise?

[Zapitron]

I stumbled upon another weird case where someone saw LVM be faster than directly using partitions. Turned out the LVM overhead (or maybe just the tester's whims) made a 1TB disk become slightly less than 1TB and apparently that allowed XFS (without inode64 mount option) keep inodes closer to their data. Life can be full of surprises.

Michael's tests were with ext4, though. And it's more of a size thing than an LVM thing, but if using LVM changes your sizing habits, then some performance differences could seem LVM-related...

[mercutio]

not using barriers = I don't care about data = fedora is unfit for every even slightly serious setup.

I don't think fedora has ever meant to have been for serious setups. It's more a distribution to try things out, before putting into RHEL.