[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On 02/12/11 23:39, Simon Waters wrote: > On 02/12/11 19:25, Gordon Henderson wrote: >> He said it was Debian Lenny - that's a few year old now and Reiser was >> still in-vogue back then... > s/in-vogue/the most mature journalling file system in the kernel/ ;) > > Which I'm guessing will still be true till they drop support. > If I remember correctly, back around kernel 2.6.33 (?) they kludged around some of the crappiness of reiser's infamous BKL (big kernel lock) stupidity by switching to recursive mutex instead: sadly, it didn't help a lot because reiserfs is architecturally based heavily around BKL and serious re-writing would be required to sort out the mess. Long story short, reiserfs has always, and will always, scale badly on SMP systems due to concurrency problems. Version 3 is inherently crippled on modern systems (or any older, bigger multi socket/core server gear), a production-ready version 4 will probably never see the light of day. Reiserfs did admittedly have some small benefits years ago when ext3 was a limiting factor as it scaled bigger and was faster for many/smaller file type workloads, and was quite efficient. Now we all have SMP boxes with multi-terrabyte disks and ext4 though, and even SuSE dropped it five years ago as the default FS. Reiser3 also had it's share of particularly idiotic problems (which I have personally had to fix many times in case you're wondering why I'm so bitter): some asynchronous operations on the many/small files case that reiserfs is touted at being so good at would result in data corruption frequently - postfix being the worst. Reiser fsck was broken. And I could go on... As for the most mature journalling file system in the kernel, I can only guess you're joking right? That would be XFS. Older, bigger, faster and with more features and stability. I can definitely understand why people *used* to use reiserfs back in the days, but those days are very, very long gone. I wish to god that either BTRFS would reach stable or someone could beat some sense into Oracle so we could have ZFS in kernel - preferably both if at all possible. Cheers, Mat -- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq