Confusion about chunk/stripe size mdadm in combination with LVM stripe size

Confusion about chunk/stripe size mdadm in combination with LVM stripe size

[Caspar Smit]

Hi all,

I got a little confused about the various chunk/stripe sizes when
using mdadm in combination with LVM.
I searched the mailing list archives and googled it but couldn't
really find a definite answer to my question.

I'm searching for the most optimal chunk/stripe sizes using mdadm and LVM.


Here's an example setup:

I created 2 md sets, both raid10 containing 8 disks with a chunk size of 512Kb

This is where the confusion starts and correct me if i'm wrong here:
To my knowledge I now have a stripe size of 2048Kb on each md set
raid10 has 4 actual data disks (and 4 mirrors), so one stripe of data
is 4x 512Kb = 2048Kb


Now when using LVM:

I created two PV's of both md sets and added them to one VG.
Now when I create a LV and like to stripe across both PV's (md sets)
what would be the ideal stripe size?

2048Kb or 4096Kb?

When i read the LVM documentation about creating a striped LV, it
looks like the stripe size is what mdadm calls the chunk size.
Or is the stripe size the size of a total stripe across all PV's?

I hope someone can clarify this for me.

P.S. I'm also curious how this calculation would be if using different
raid levels like raid5 or raid6

Kind regards,
Caspar Smit

Re: Confusion about chunk/stripe size mdadm in combination with LVM stripe size

[Stan Hoeppner]

On 8/23/2012 4:04 AM, Caspar Smit wrote:
> Hi all,
> 
> I got a little confused about the various chunk/stripe sizes when
> using mdadm in combination with LVM.
> I searched the mailing list archives and googled it but couldn't
> really find a definite answer to my question.
> 
> I'm searching for the most optimal chunk/stripe sizes using mdadm and LVM.

<snip>

It would seem you're one of the countless we see here with hardware,
real or imaginary, who has no workload, or is seeking a workload for his
hardware.  This is the wrong approach.

You need to approach this list with:

1.  This is my workload
2.  This is the hardware I currently have (suitable or not)
3.  How do I best setup my storage to meet my workload requirements?

You're approaching this backwards.

-- 
Stan