Thread (4 messages) flat view 4 messages, 3 authors, 2021-02-04

Re: Re: put 2 hard drives in mdadm raid 1 and detect bitrot like btrfs does, what's that called?

From: <hidden>
Date: 2021-02-04 17:44:47
Also in: linux-btrfs

--- Ursprüngliche Nachricht ---
Von: Andy Smith [off-list ref]
Datum: 04.02.2021 11:54:57
An: linux-btrfs@vger.kernel.org
Betreff: Re: put 2 hard drives in mdadm raid 1 and detect bitrot like btrfs  does, what's that called?

Hi Cedric,

On Wed, Feb 03, 2021 at 08:33:18PM +0100,   wrote:
it's called "dm-integrity", as mentioned in this e-mail:
https://www.mail-archive.com/linux-btrfs@vger.kernel.org/msg93037.html

If you do this it would be very interesting to see performance
figures for the following setups:

- btrfs with raid1 meta and data allocation
- mdadm raid1 on raw devices
- mdadm raid1 on dm-integrity (no encryption) on raw devices
- mdadm raid1 on dm-integrity (encryption) on raw devices

just to see what kind of performance loss dm-integrity and
encryption is going to impose.

After doing it, it would find a nice home on the Linux RAID wiki:

    https://raid.wiki.kernel.org/index.php/Dm-integrity

Cheers,
Andy

Hey Andy,

I would rather see performance figures for these setups:
A) btrfs with 2 (or more) hard drives and one SSD in writeback bcache configuration (unsafe against failure of the ssd):
+-----------------------------+
|      btrfs raid 1 /mnt      |
+--------------+--------------+
| /dev/Bcache0 | /dev/Bcache1 |
+--------------+--------------+
|   bcache writeback Cache    |
|           /dev/sdk1         |
+--------------+--------------+
| Data         | Data         |
| /dev/sdv1    | /dev/sdw1    |
+--------------+--------------+

B) btrfs with 2 (or more) hard drives and two SSD's in dm-raid 1 writeback bcache configuration (unsafe against corruption of any of the ssd's): 
+-----------------------------+
|      btrfs raid 1 /mnt      |
+--------------+--------------+
| /dev/Bcache0 | /dev/Bcache1 |
+--------------+--------------+
|   bcache writeback Cache    |
|           /dev/dm0          |
+--------------+--------------+
| 2x SSD in mdadm raid 1      |
| /dev/sdk1       /dev/sdl1   |
+--------------+--------------+
| Data         | Data         |
| /dev/sdv1    | /dev/sdw1    |
+--------------+--------------+

C) Full stack: btrfs with 2 (or more) hard drives and two identical SSD's in dm-raid 1 with dm-integrity writeback bcache configuration (safe against any failed drive):
+-----------------------------+
|      btrfs raid 1 /mnt      |
+--------------+--------------+
| /dev/Bcache0 | /dev/Bcache1 |
+--------------+--------------+
|   bcache writeback Cache    |
|           /dev/dm0          |
+--------------+--------------+
| 2 x dm-integrity devices    |
| in mdadm raid 1             |
+--------------+--------------+
| SSD hosting  | SSD hosting  |
| dm-integrity | dm-integrity |
| /dev/sdk1    | /dev/sdl1    |
+--------------+--------------+
| Data         | Data         |
| /dev/sdv1    | /dev/sdw1    |
+--------------+--------------+

D) Full stack: btrfs with 2 (or more) hard drives and two SSD's (one slow, and one very fast) in dm-raid 1 with dm-integrity writeback bcache configuration (safe against any failed drive):
+-----------------------------+
|      btrfs raid 1 /mnt      |
+--------------+--------------+
| /dev/Bcache0 | /dev/Bcache1 |
+--------------+--------------+
|   bcache writeback Cache    |
|           /dev/dm0          |
+--------------+--------------+
| 2 x dm-integrity devices    |
| in mdadm raid 1             |
+--------------+--------------+
| SSD hosting  | SSD hosting  |
| dm-integrity | dm-integrity |
| /dev/sdk1    | /dev/sdl1    |
+--------------+--------------+
| Data         | Data         |
| /dev/sdv1    | /dev/sdw1    |
+--------------+--------------+

In all these setups, the performance of the hard drives is irrelevant, because the speed of the setups comes from the bcache SSD.

Cheers,
Cedric

---

Take your mailboxes with you. Free, fast and secure Mail &amp; Cloud: https://www.eclipso.eu - Time to change!

Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help