Lvm – Optimizing ext2 filesystem for use on LVM + RAID device? Stride, stripe-width, LVM IO size considerations

I'm building a new database server and am trying to make sure I have the setup as correct as possible. The database storage is 22x 15.7k RPM Western Digital SAS drives in RAID10, in an external SAS enclosure with two RAID controllers. Initially, I created just a single volume on the SAS array, formatted with ext2. I calculated the Stride and Stripe-Width as follows:

• chunk_size = 128 Kb
• block_size = 4 Kb
• total_disks = 22
• data_disks = 11
• stride = chunk_size / block_size = 128 Kb / 4 Kb = 32 Kb
• stripe-width = stride * data_disks = 32 Kb * 11 = 352 Kb

I gave the Stride and Stripe-Width values to mkfs.ext2 when I created the filesystem.

I then tried another setup, which is where my question begins. I created two volumes on the SAS array, each with a different "primary" controller. When exported to the host, I used LVM2 to initialize them as Physical Volumes, created a Volume Group, and then passed '–stripes 2' to lvcreate in order to stripe the volumes. The purpose of this was to spread the IO load across both controllers, for hopefully higher performance.

There is a section in the "best practices" documentation for the SAS array that states, in reference to LUN segment (chunk) size:

When using a volume manager to collect multiple storage system LUNs into a Logical
Volume Manager (LVM) volume group (VG), the I/O stripe width is allocated across all of > the segments of all of the data drives in all of the LUNs. The adjusted formula becomes:
LUN segment size = LVM I/O stripe width / (# of data drives/LUN * # of LUNs/VG)

Now, the Volume Group on the SAS array has 22 drives, 11 of which are data drives, and two LUNs. The Volume Group on the host has two physical volumes (the LUNs), and both LUNs access the same data drives. So given a LUN segment size of 128 Kb, would I calculate LVM I/O Stripe Width as

• 128 = LVM_Stripe_Width / (11 * 2)
• 128 * 22 = LVM_Stripe_Width
• 2816 = LVM_Stripe_Width

or

• 128 = LVM_Stripe_Width / (11 * 1)
• 128 * 11 = LVM_Stripe_Width
• 1408 = LVM_Stripe_Width

Then this leads me into the Stride and Stripe-width question for mkfs.ext2: Is it calculated the same way as in the initial setup, or is it different now because of the LVM layer? Would it become

• chunk_size = 128 Kb
• block_size = 4 Kb
• total_disks = 2
• data_disks = 2
• stride = chunk_size / block_size = 128 Kb / 4 Kb = 32 Kb
• stripe-width = stride * data_disks = 32 Kb * 2 = 64 Kb

where

• total_disks = 2

comes from the '–stripes 2' to lvcreate?

Thanks,

Kendall