Alignment and performance

Question

hey, i'd really like to get the best out of ssd but the linux documentation is patchy and on average is contradictory and at worst is destructive.
here's the deal, i have a 50gb ocz vertex which i am more than willing to thrash to death to discover the best alignment parameters for ssd under linux and can be applied (roughly) for my next ssd.
i've already tried numerous techniques for partitioning an ssd the best result came with a nothing special install which culminated in a top seek of just under 200mb.sec. the ocz website tells me sustained througput can be 250+... that is my aim.
so an initial suggestion on how to align 4 primary partitions. there will be no swap!
part 1 = /boot and some extra for safe storage = 1 gb = ext2
part 2 = /root =10000 gb = ext4
part 3 = /home = 20000 gb = ext4 or maybe xfs
part 5 = storage = the rest = ext4
(for some reason sda4 is only used for  extended/logical)
i know how to align a single entire disk partition with info i found here http://www.ocztechnologyforum.com/forum … post373226 but once more partitions are created the  fstab is altered and thus becomes mis-aligned. so, it might be helpful (possible?) to align four separate partitions as they are created or post created.
so, if i created the aforementioned partitions with fdisk or parted does anybody at all know the command line parameters to create and format all 4 partitions and alignment occurs?
i have gone over the man for fdisk but it remains a bit unclear;
will partitions 235 be automatically aligned if part 1 is aligned or will part 235 have to be resized to have a length of 0 at the respective partitions opening cylinder? i digress
anyway, fdisk setup creation for ssd suggestions very welcome and unless they are completely outlandish i'll give 'em a whirl.
thx

Answer 1

This is how i've aligned my Intel X25-M:
Disk /dev/sda: 80.0 GB, 80026361856 bytes
32 heads, 32 sectors/track, 152638 cylinders
Units = cylinders of 1024 * 512 = 524288 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x929cf67c
Device Boot Start End Blocks Id System
/dev/sda1 * 2 513 262144 83 Linux
/dev/sda2 514 2049 786432 82 Linux swap / Solaris
/dev/sda3 2050 12289 5242880 83 Linux
/dev/sda4 12290 152638 71858688 5 Extended
/dev/sda5 12290 152638 71858672 83 Linux
Nr AF Hd Sec Cyl Hd Sec Cyl Start Size ID
1 80 0 1 1 31 32 512 1024 524288 83
2 00 0 1 513 31 32 1023 525312 1572864 82
3 00 31 32 1023 31 32 1023 2098176 10485760 83
4 00 31 32 1023 31 32 1023 12583936 143717376 05
5 00 31 32 1023 31 32 1023 1024 143717344 83
It's based on assumptions / tips i've found on the net:
* use fdisk to partition in cylinders instead of sectors so that the partitions automatically align to cylinder boundaries. I've used fdisk -H 32 -S 32, obviously.
* partition with an alignment of at least 32-64kb (ideally to fit the erase block size) to minimize unnecessary writes. 32*32*512b gives you 524288b = 512k alignment which many assume is the erase block size of the intel ssds.
* If you use an extended partition you have to manually change the offset because it's set to 32 by default. Choose alignment *2 instead.
* When using 4k block size, tell ext4 to stripe times 128, so that the FS aligns to 512k, too.
Feel free to add and/or correct me, if i'm wrong.
But, badly aligned, well aligned, JFS, XFS, EXT4, BTRFS or even x86 vs x86_64 - I tried it all and you really don't _feel_ a difference. It always just feels very responsive and fast. I didn't bother with benchmarks, but hdparm -t always gave me ~250MBps.
Last edited by demian (2010-06-10 10:21:30)