LWN: Comments on "The 2.6.29 kernel is out"

The 2.6.29 kernel is out

Spudd86 — Fri, 27 Mar 2009 14:59:13 +0000

It moves responsability for large amounts of graphics hardware state out of X and into the kernel, this lets all kinds of things happen.
- Mode only gets set once as you boot (no flicker when X starts)
- Kernel can display panic messages while X is running
- X doesn't need to be root anymore

The 2.6.29 kernel is out

bojan — Wed, 25 Mar 2009 06:42:42 +0000

But it was so much fun! ;-)

The 2.6.29 kernel is out

thedevil — Wed, 25 Mar 2009 06:28:18 +0000

I am confused by the excitement about kernel GPU modesetting.

Wasn't this exactly what the "use framebuffer for modesetting" option in the X server has done for many years now?

I am a very light user of graphics - 2D only, no games, etc. That may partly excuse my ignorance here :)

The 2.6.29 kernel is out

muwlgr — Tue, 24 Mar 2009 20:46:11 +0000

Google do not use fsync. They use triple redundancy for their data and handle single storage node faults as usual in-process working events (i.e. bring the redundancy back to 3, and replace the failed node with a new empty operational one).

The 2.6.29 kernel is out

tialaramex — Tue, 24 Mar 2009 20:45:30 +0000

Ted's talking about ext4 there, not ext3. Combining delayed allocation with data journalling clearly doesn't make much sense. But point taken.

The 2.6.29 kernel is out

cortana — Tue, 24 Mar 2009 19:59:21 +0000

The data=journal mode protected data, didn't it?

Apparantly not. I think you want the 'sync' mount option in this case.

The 2.6.29 kernel is out

tialaramex — Tue, 24 Mar 2009 18:19:01 +0000

The data=journal mode protected data, didn't it?

But yes, some people (and Google are apparently enough people to count) want the best possible FS features minus journalling.

Also fsck is probably /faster/ in ext4 than ext2, so Google would be wise to choose ext4 w/o journalling over ext2 even if they did care about fsck performance. Sparse inode structures iirc were a particularly big fsck win.

The 2.6.29 kernel is out

nix — Tue, 24 Mar 2009 15:45:23 +0000

Journals only protect metadata, not data. Google don't care how long fsck takes (they never run it), but they *do* care about I/O rates. So they don't want to use a journal.

The 2.6.29 kernel is out

Tuxie — Tue, 24 Mar 2009 14:02:58 +0000

Btrfs is cool but I'm waiting for it to get Z-RAID equivalent functionality before I'll start playing with it.

The 2.6.29 kernel is out

clugstj — Tue, 24 Mar 2009 13:04:57 +0000

Let's not start this again!

The 2.6.29 kernel is out

epa — Tue, 24 Mar 2009 12:21:25 +0000

Yes, obviously if your application requires the data to be stored permanently it should call fsync()!

The 2.6.29 kernel is out

kragil — Tue, 24 Mar 2009 11:40:57 +0000

The patch is from Google (they still use EXT2) because they just reimage a machine when it crashes.

Sometimes you don't need a journal but you want the added speed and features.

The 2.6.29 kernel is out

ssam — Tue, 24 Mar 2009 11:34:46 +0000

Great to see the 'Ext4 no journal mode'.

I guess this is to guarantee data loss after hard crashes and power outages. POSIX FTW