[arch-general] Adding new partition to system from unpartitioned space - any reason to not use ext4??
Guys, This is just a quick request to make sure I don't screw myself playing with new technology. I have a server 5 existing ext3 partitions on a dmraid array with several hundred gig of unpartitioned space. I have created the new partition and I'm about the create the filesystem. With ext4 now available, is there any reason I shouldn't go ahead and create the filesystem as ext4? It will simply be another data partition to supplement /home that was created rather small to begin with. Any thoughts or gotchas on using ext4 instead of ext3 for this?? -- David C. Rankin, J.D.,P.E. Rankin Law Firm, PLLC 510 Ochiltree Street Nacogdoches, Texas 75961 Telephone: (936) 715-9333 Facsimile: (936) 715-9339 www.rankinlawfirm.com
Am 07.01.2010 18:26, schrieb David C. Rankin:
Guys,
This is just a quick request to make sure I don't screw myself playing with new technology. I have a server 5 existing ext3 partitions on a dmraid array with several hundred gig of unpartitioned space. I have created the new partition and I'm about the create the filesystem. With ext4 now available, is there any reason I shouldn't go ahead and create the filesystem as ext4? It will simply be another data partition to supplement /home that was created rather small to begin with.
Any thoughts or gotchas on using ext4 instead of ext3 for this??
Delayed allocation might lead to data loss of recently created or overwritten files in case of power failure or system freeze. The most common case of overwriting files in badly written programs has been worked around in ext4, but theoretically it might still happen (witnessed by a file with size 0 after power loss or otherwise unclean shutdown). The same is the case for xfs or btrfs though. You can disable delayed allocation, but that will negate one of the performance improvements in ext4 compared to ext3.
Thomas Bächler wrote:
Delayed allocation might lead to data loss of recently created or overwritten files in case of power failure or system freeze. The most common case of overwriting files in badly written programs has been worked around in ext4, but theoretically it might still happen (witnessed by a file with size 0 after power loss or otherwise unclean shutdown). The same is the case for xfs or btrfs though. You can disable delayed allocation, but that will negate one of the performance improvements in ext4 compared to ext3.
Thomas, Thank you. That's exactly the type information I was looking for. Since this Arch box is my home server I'll give ext4 a shot. Since this thing functions under a light load and has about 30 minutes of battery backup I should be fine. I'll pass along any anomalies -> hopefully you won't see any ext4 message subjects from me later... -- David C. Rankin, J.D.,P.E. Rankin Law Firm, PLLC 510 Ochiltree Street Nacogdoches, Texas 75961 Telephone: (936) 715-9333 Facsimile: (936) 715-9339 www.rankinlawfirm.com
Thomas Bächler wrote:
Am 07.01.2010 18:26, schrieb David C. Rankin:
the create the filesystem. With ext4 now available, is there any reason I shouldn't go ahead and create the filesystem as ext4? It
Delayed allocation might lead to data loss of recently created or overwritten files in case of power failure or system freeze. The most common case of overwriting files in badly written programs has been worked around in ext4, but theoretically it might still happen (witnessed by a file with size 0 after power loss or otherwise unclean shutdown).
Kind of late response, but I strongly agree with Thomas. I've had my share of truncated files after unclean shutdowns (every single time), some kernel modules being among the victims. If you decide to decide to use ext4, then it's almost mandatory to use the 'nodelalloc' mount option, unless you're positive that unclean shutdowns are unlikely in your setup or you don't care about lost data. -- X.
On Wednesday 13 January 2010 21:57:04 Christos Nouskas wrote:
Kind of late response, but I strongly agree with Thomas. I've had my share of truncated files after unclean shutdowns (every single time), some kernel modules being among the victims. If you decide to decide to use ext4, then it's almost mandatory to use the 'nodelalloc' mount option, unless you're positive that unclean shutdowns are unlikely in your setup or you don't care about lost data.
Isn't that option auto_da_alloc? -- Shridhar
On 01/13/2010 08:06 PM, Shridhar Daithankar wrote:
On Wednesday 13 January 2010 21:57:04 Christos Nouskas wrote:
Kind of late response, but I strongly agree with Thomas. I've had my share of truncated files after unclean shutdowns (every single time), some kernel modules being among the victims. If you decide to decide to use ext4, then it's almost mandatory to use the 'nodelalloc' mount option, unless you're positive that unclean shutdowns are unlikely in your setup or you don't care about lost data.
Isn't that option auto_da_alloc?
Well, I thought I would close the thread with a follow-up on how it went and what I ultimately went with. I ended up expanding the extended partition to encompass the remainder of the disk. I then created 2 ext4 partitions, a 6 gig for private data that can be mounted for use then unmounted and the remainder as a big data drive. Both are dmraid mirrored partitions. The current mount options are: /dev/mapper/nvidia_ecaejfdip12 /data ext4 defaults 0 1 /dev/mapper/nvidia_ecaejfdip11 /home/david/pvt ext4 defaults 0 1 are you guys saying I should add 'auto_da_alloc' as an option for all ext4 partitions? -- David C. Rankin, J.D.,P.E. Rankin Law Firm, PLLC 510 Ochiltree Street Nacogdoches, Texas 75961 Telephone: (936) 715-9333 Facsimile: (936) 715-9339 www.rankinlawfirm.com
On Thu, Jan 07, 2010 at 11:26:05AM -0600, David C. Rankin wrote:
Guys,
This is just a quick request to make sure I don't screw myself playing with new technology. I have a server 5 existing ext3 partitions on a dmraid array with several hundred gig of unpartitioned space. I have created the new partition and I'm about the create the filesystem. With ext4 now available, is there any reason I shouldn't go ahead and create the filesystem as ext4? It will simply be another data partition to supplement /home that was created rather small to begin with.
Any thoughts or gotchas on using ext4 instead of ext3 for this?? No. There is no reason not to use ext4. It's a good all-around filesystem. Google are in the process of moving from ext2 to ext4.
Now, give me a reason why you should listen to me?
On 01/07/2010 04:01 PM, arch@nezmer.info wrote:
On Thu, Jan 07, 2010 at 11:26:05AM -0600, David C. Rankin wrote:
Guys,
This is just a quick request to make sure I don't screw myself playing with new technology. I have a server 5 existing ext3 partitions on a dmraid array with several hundred gig of unpartitioned space. I have created the new partition and I'm about the create the filesystem. With ext4 now available, is there any reason I shouldn't go ahead and create the filesystem as ext4? It will simply be another data partition to supplement /home that was created rather small to begin with.
Any thoughts or gotchas on using ext4 instead of ext3 for this?? No. There is no reason not to use ext4. It's a good all-around filesystem. Google are in the process of moving from ext2 to ext4.
Now, give me a reason why you should listen to me?
Guys, Partitioning is done, Always a bit more fun with dmraid. To begin with, I had simply used fdisk and mkfs to create dm-11 (6G) out of unpartitioned space at the end of my extended partition. I had then created a 516G primary partition out of the remaining unpartitioned space which was created as dm-13 also with an ext4 filesys. The whole setup seemed sloppy to have 1/3 third of the array as an extended w/logical partitions and 2/3 of the disk as a big primary, so I thought I would do what I thought looked better and delete the new logical and primary partitions I had created, grow the extended to encompass the entire disk and then re-add new logical partitions with ext4 filesystems instead (one 6G for an encrypted partition and then the remaining 581G as the data partition I wanted to supplement /home. One issue worth noting -- gparted recommended using 0.4.6-1 until a few bugs get ironed out with 0.5.X so I downloaded a fresh live cd iso and used that. I ran into an issue where gparted froze on the first attempt. My initial preparation resulted in 5 transactions for gparted to execute. gparted completed 2.5 of them and froze creating the 6 G extension which required a power-off hard reset. (always something that will cause your ... to pucker during the middle of writing partitions to disk) On restart, gparted reported that the 6G partition was of "Unknown type", but other than that the array was perfectly fine. So I set it up to delete/recreate the 6G and create the 581G which completed without any issue. I don't know why gparted caused the system to freeze the first time, but be aware there may be some issues with gparted right now that are being investigated according to the gparted site. The impressive part of the whole endeavor was the fact that gparted handled the power-off hard reset without scattering my entire drive :p Thanks for all your feedback on the ext4 issue. -- David C. Rankin, J.D.,P.E. Rankin Law Firm, PLLC 510 Ochiltree Street Nacogdoches, Texas 75961 Telephone: (936) 715-9333 Facsimile: (936) 715-9339 www.rankinlawfirm.com
participants (5)
-
arch＠nezmer.info
-
Christos Nouskas
-
David C. Rankin
-
Shridhar Daithankar
-
Thomas Bächler