On 19/02/2013 19:05, Reindl Harald wrote:
Am 19.02.2013 20:02, schrieb Gordan Bobic:
what exactly do you need to align on the partitions?
For a start, making sure your RAID implementation puts the metadata
at the end of the disk, rather than the beginning.
"my RAID implementation"?
LINUX SOFTWARE RAID
and this is how the raid-partitions are looking
no problem since years
Disk /dev/sda: 2000.4 GB, 2000398934016 bytes, 3907029168 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x0000ae2c
Device Boot Start End Blocks Id System
/dev/sda1 * 2048 1026047 512000 fd Linux raid autodetect
/dev/sda2 1026048 31746047 15360000 fd Linux raid autodetect
/dev/sda3 31746048 3906971647 1937612800 fd Linux raid autodetect
[root@srv-rhsoft:/downloads]$ sfdisk -d /dev/sda
# partition table of /dev/sda
unit: sectors
/dev/sda1 : start= 2048, size= 1024000, Id=fd, bootable
/dev/sda2 : start= 1026048, size= 30720000, Id=fd
/dev/sda3 : start= 31746048, size=3875225600, Id=fd
/dev/sda4 : start= 0, size= 0, Id= 0
That's the MD partition alignment, not the alignment of the FS space
within the MD device. The two are not the same.
MD superblock versions 0.9 and 1.0 are the the end of the block device.
Versions 1.1 and 1.2 are at the beginning of the device. I don't recall
off the top of my head whether the superblock size and offset ensure
that the /dev/mdX device's space is actually aligned to a 4KB boundary
on the raw underlying device.
Throw in the SSD erase block sizes and RAID chunk size and you actually
have to think through your alignment throughout the entire stack or
you'll get it wrong. And if you are using ext*, make sure you adjust
your block group size (-g) so that the block group beginnings rotate
around the disks - default values more often than not lead to all the
block groups starting on the same disk which then becomes a massive
bottleneck.
See here for details:
http://www.altechnative.net/2010/12/31/disk-and-file-system-optimisation/
The article is old enough that it refers to fdisk in CHS rather than LBA
mode (CHS is still default in EL6), but the general principle still
applies in exactly the same way.
Gordan
--
users mailing list
users@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org
