| 1 |
On Friday 10 December 2010, Stefan G. Weichinger wrote: |
| 2 |
> Am 09.12.2010 18:21, schrieb Volker Armin Hemmann: |
| 3 |
> > Put /boot on raid1, / on raid6. Don't bother with lvm - it is just |
| 4 |
> > another layer that can go wrong. |
| 5 |
> |
| 6 |
> You mean "don't use lvm for /" ? ... for other stuff it's very useful, |
| 7 |
> isn't it? |
| 8 |
|
| 9 |
no, I mean don't use lvm. |
| 10 |
It is a very compley, easily broken layer reducing data safety. |
| 11 |
|
| 12 |
There is no need for lvm with bind mounting and ln -s. |
| 13 |
|
| 14 |
> |
| 15 |
> > If Raid6 is like raid5 you should be |
| 16 |
> > able to have the kernel auto assemble everything, so no initrd is |
| 17 |
> > needed. Pay attention to the metadata format when creating the raid. |
| 18 |
> > Also have a look at stripe sizes. And stripe_cache_size. |
| 19 |
> |
| 20 |
> Thanks for your reminders. |
| 21 |
> What exactly do you think of with stripes and sizes? |
| 22 |
> You point at the performance-impact? large files vs. small files etc? |
| 23 |
|
| 24 |
stripe size has nothing to do with big and small files. But choosing the wrong |
| 25 |
stripe size can impact your performance very, very badly. We are talking about |
| 26 |
abysmal performance, Challenger depht abysmal. XFS and ext4 - for both is a |
| 27 |
lot of documentation available about choosing the right stripe size. |
| 28 |
|
| 29 |
stripe_cache_size can be set in /sys and has a 4-5x performance impact on |
| 30 |
raid5 (where I tried it). A good size for me is 8192. For example. |
Archives