Kernel 3.1 is released

[viking60]

Kernel 3.1 is released:
https://lkml.org/lkml/2011/10/24/212

Here are the improvements.

One thing to notice is that your Ext3 filesystem might get slower after upgrading. This is because the Kernel now has implemented the Red hat policy of turning on filesystem barriers by default in Ext3:

Hard disks have a memory buffer were they temporally store the instructions and data issued from the OS while the disk processes it. The internal software of modern disks changes the order of the instructions to improve performance, which means that instructions may or may not be committed to the disk in the same order the OS issued them. This breaks many of the assumptions that filesystems need to reliably implement things like journaling or COW, so disks provide a "cache flush" instruction that the OS uses when it needs it. In the Linux world, when a filesystem issues that instruction, it is called a "barrier". Filesystems such as XFS, Btrfs and Ext4 already use and enable barriers by default; Ext3 supports them but until this release it did not enable them by default: while the data safety guarantees are higher, their performance impact in Ext3 is noticeable in many common workloads, and it considered that it was an unnaceptable performance regression to enable them by default. However, Linux distros like Red Hat have enabled barriers by default in Ext3 for a long time, and now the default for mainline has been changed aswell.

In other words: if you use Ext3 and you note performance regressions with this release, try disabling barriers ("barriers=0" mount option).

As you can see it will not affect XFS, Btrfs and Ext4.

[rolf]

No mention of reiserfs, afaict, but here's an old comment, true or false, idk, that might be interesting:

extx3 disables barriers by default, but xfs and reiserfs enable it by default. If enabled there should be (if you believe the kernel people) a 30% penalty for ext3.

[viking60]

It is probably still true then. The major new thing is that Ext3 has it enabled now. The others who already have it, have been more secure at the familiar speed for some time.

[rolf]

I don't spend time chasing down benchmarks but some of the speed-related fervor for ext3 might die down if the hit is that big. I thought there were some credible claims about the superior performance of reiser4 but that is not in mainline, has required a patch from Edward Shishkin that Thomas Backlund was implementing in his kernels prior to kernel-3. The latest I have seen of whether he will continue to do that is around here:
http://www.spinics.net/lists/reiserfs-d ... 02990.html and at the link, above.

He would post those patches @ kernel.org, under people/. After the compromise of those servers, there have only been 3 devels who have re-made their accounts, one more than the last time I mentioned this, no edward... ftp://ftp.kernel.org/pub/linux/kernel/people/

[viking60]

It looks like there is not much development on ReiserFS.

[rolf]

It looks like there is not much development on ReiserFS.

There is still development, kernel patches: http://www.spinics.net/lists/reiserfs-d ... html#02990

[dedanna1029]

There is still development, kernel patches: http://www.spinics.net/lists/reiserfs-d ... html#02990

There sure is! Even if considered outdated, one will not get performance like they do from ReiserFS, and I believe that to be for all time. I'll probably be sitting in the year 2025, still running ReiserFS, because no one's come out with anything as stable.