1 |
> I thought of creating 5 200mb paratitions of raid1. So 5 will be |
2 |
> mirrored. |
3 |
> What do you think of this solution? |
4 |
|
5 |
I don't think it will help much... Sorry. Partial writes are partial writes |
6 |
|
7 |
The main way you can kill flash drives is to power off while they are |
8 |
doing an internal write. On many devices if you do this at the right |
9 |
moment then the flash drive completely disappears and is completely gone |
10 |
(not recoverable as far as I know, never can be used or data accessed |
11 |
ever again...) |
12 |
|
13 |
I believe some very new drives has small capacitors to finalise a write |
14 |
if the power goes off suddenly, however, I think on everything else your |
15 |
flash is likely toast... The issue is corruption of the internal flash |
16 |
drive filesystem, NOT the filesystem that you create on top of that (I |
17 |
have one dead flash drive and I think it may well have been killed by |
18 |
this? Allegedly the latest SLC flash drives from PC Engines are somewhat |
19 |
resistant to this problem...) |
20 |
|
21 |
With regards to corrupting the user filesystem due to sudden power loss, |
22 |
ie as per normal spinning drives, then yes, any journalling filesystem |
23 |
seems like a good idea. Ext4 is chosen by google (albeit without a |
24 |
journal...). XFS is an odd ball choice and may have massive write |
25 |
amplification problems that I don't know about, but what is nice is that |
26 |
it seems resiliant to corruption on powerloss and avoids a fsck step. |
27 |
This is only "better" in the sense that it doesn't require user input |
28 |
and for embedded, this may keep the device going longer between visits |
29 |
from support... However, bottom line is that if the power goes off |
30 |
during a write then at some point you loose data... |
31 |
|
32 |
|
33 |
Good luck |
34 |
|
35 |
Ed W |