* Thomas Bächler [21.04.2010 09:41]:

Am 21.04.2010 09:34, schrieb Jan de Groot:

...

On Tue, 2010-04-20 at 20:52 -0400, Caleb Cushing wrote:

...

has anyone installed arch on one of these drives? thinking about buying a pair and running them mirrored... but not sure if everything will be working optimal. note: since I'll have to migrate off an existing drive I can use the tools installed on that to partition it first. Is there anything I should know first?

You will have to align partitions on a 4K boundary, performance will be miserable without that. I think parted is the best tool to do partitioning in this case, as parted supports aligning.

I think the very very latest util-linux-ng will have fdisk/cfdisk that'll handle it just fine.

I can second that, I don't have Arch installed on a 4k drive but I have an external drive, which I partitioned with fdisk -c -u /dev/sdX and there's no speed drop at all, so you can use this command line without any worries

On Wed, Apr 21, 2010 at 3:40 AM, Thomas Bächler wrote:

I think the very very latest util-linux-ng will have fdisk/cfdisk

is that shipping on any of the install disks? -- Caleb Cushing http://xenoterracide.blogspot.com

On Thu, Apr 22, 2010 at 7:12 PM, Thomas Bächler wrote:

You should google more to see if 4k sectors are really well-supported now.

I've actually done quite a bit of research. But most if it came back upstream. So I thought I'd better ask in an arch specific context. According to one of the latest upcoming kernel developments there will be more 4k improvements in 2.6.34 what I was unable to determine is if that was something that would require me to do something after the kernel is released. In fact some of this stuff I was confused as to whether... do I have to run fdisk in a certain way? is cfdisk supported? how can I check to make sure it is working? and the obvious even if the kernel and tools are now ready will an arch disk be ready (obviously I already asked and has been answered)... -- Caleb Cushing http://xenoterracide.blogspot.com

On Mon, Apr 26, 2010 at 9:22 PM, Marti Raudsepp wrote:

However, this WD disk crashed after barely 3 days of operation and I replaced it with a normal 512-byte sector Seagate.

eek! hope the ones I ordered today don't do that. New article on developer works 'bout them http://www.ibm.com/developerworks/linux/library/l-4kb-sector-disks/index.htm... -- Caleb Cushing http://xenoterracide.blogspot.com

How did you get two drives into RAID5 and if so, why? On Apr 29, 2010 7:38 PM, "bardo" wrote: 2010/4/29 Caleb Cushing :

On Mon, Apr 26, 2010 at 9:22 PM, Marti Raudsepp wrote:

...

However, this WD disk ... Just two days ago I built an Arch NAS/HTPC with two 2TB WD Caviar Green WD20EARS (the ones with 4k-sectors). I used the latest iso from build.archlinux.org, the one that Thomas suggested, and everything went happily fine. I used parted's interactive mode to partition, followed by mdadm and mkfs.

Parted automatically warned me about an unoptimal sector alignment when I tried to put the first partition at 0 (the first partition should start at 2MB), but if you want to be extra-sure just check the the '-a optimal' parameter. About the disk reliability I can't tell much, but I already transferred some 100GB to it, and of course I had the first RAID5 build, which took a few hours and ran through the whole disk. Also, smartctl doesn't have anything bad to say. HTH, Corrado

On Fri, Apr 30, 2010 at 2:37 AM, bardo wrote:

Parted automatically warned me about an unoptimal sector alignment when I tried to put the first partition at 0 (the first partition should start at 2MB), but if you want to be extra-sure just check the the '-a optimal' parameter.

Great to hear that this has changed. :)

About the disk reliability I can't tell much, but I already transferred some 100GB to it, and of course I had the first RAID5 build, which took a few hours and ran through the whole disk. Also, smartctl doesn't have anything bad to say.

I run weekly SMART "long" tests on all my disks because, in my experience, that has been the best indication of impending disk failure. It performs a full surface scan, so on a 2TB disk this can take 6 hours (maybe more on green disks). Something like this in smartd.conf: /dev/sda -a -H -l error -l selftest -f -m marti@juffo.org -s L/../../3/11 checks for attribute changes and performs a long test every thursday (3) at 11 o'clock and emails me about problems. Regards, Marti

On 04/30/2010 10:13 AM, Thomas Bächler wrote:

Am 30.04.2010 01:37, schrieb bardo:

...

Parted automatically warned me about an unoptimal sector alignment when I tried to put the first partition at 0 (the first partition should start at 2MB), but if you want to be extra-sure just check the the '-a optimal' parameter. About the disk reliability I can't tell much, but I already transferred some 100GB to it, and of course I had the first RAID5 build, which took a few hours and ran through the whole disk. Also, smartctl doesn't have anything bad to say.

I'm still curious about the status of 4k support in the latest util-linux-ng's (c)fdisk. Did anyone try that?

From my tests not too long ago cfdisk does not play well with the current 4K sector disks, at least when they report a 512B sector size as is the case with the WD EADS disks. It also seems (to me) that cfdisk will not play well with partition schemes that are not created in "compatibility" mode so unless it gets updated it will soon be a nonstarter for disk partitioning.

fdisk will align all partitions if you start it with 'fdisk -uc' and accept the suggested partition start and use +***M to specify the size of partitions, I didn't try specifying the end in sectors and check what it would do to the next partition though. I have tried this with an external WD Elements disk and fdisk. If using the whole disk with one partition, fdisk will set exactly the same start and end sectors that WD uses in the partition they define and format in the factory. I did not try parted but I've read that it will also align the partitions properly, but it may require some extra switch to disable the alignment to cylinders. -- Mauro Santos