1 |
On Mon, 20 Oct 2008 08:54:20 +0200 |
2 |
Wolfgang Liebich <Wolfgang.Liebich@×××××××.com> wrote: |
3 |
> Hi, |
4 |
> |
5 |
> <SNIP> |
6 |
> > |
7 |
> > the howtos on gentoo-wiki worked well for me. |
8 |
> |
9 |
> I'm working with them, too. Just one question remains: I want to use |
10 |
> udev. Do I have to create the md devices or does udev that for me? |
11 |
> |
12 |
|
13 |
udev will do it for you. But make sure your initramfs init script |
14 |
unmounts /sys & /proc. On the box I'm working on setting up it |
15 |
wasn't unmounting /sys on the initramfs, so when it switched to the |
16 |
real root it thought /sys was already mounted & didn't mount /sys |
17 |
under the real root, which meant that udev didn't work - which took |
18 |
me a while to figure out. |
19 |
|
20 |
> > |
21 |
> > |
22 |
> > > - Put the root partition on another RAID1 (I thought about |
23 |
> > > putting the root filesystem into my LVM setup, too -- it is |
24 |
> > > REALLY annoying if the root partition get's to small), |
25 |
> > |
26 |
> > yeah, but if you have 20+ gb root is always big enough ;) AFAIK |
27 |
> > lvm kills barriers. You use raid for better data security. So |
28 |
> > using lvm is a bit.. contra productive. |
29 |
> |
30 |
> Sorry, I'm neither a LVM nor a RAID export - could you please |
31 |
> elaborate on that? |
32 |
> I like LVM because of the convenience it adds. |
33 |
> |
34 |
|
35 |
Write barriers are a feature to allow write caching on the hard disks |
36 |
w/out endangering filesystem integrity. Write caching helps |
37 |
performance significantly, but also allows the disk to re-order write |
38 |
requests - the disk may actually write a write-request that was |
39 |
received later before a write-request that was received earlier, |
40 |
which in some situations can lead to filesystem corruption. Write |
41 |
barriers are a special type of request that the disk is not allowed |
42 |
to reorder around - everything the disk receives before the write |
43 |
barrier must be written before anything received after the write |
44 |
barrier. But in order to work, write barriers need to be supported |
45 |
by every layer from the filesystem down to the actual disk; if your |
46 |
filesystem is on top of LVM & LVM doesn't support write barriers, |
47 |
then you won't be able to use them, and if write caching is enabled |
48 |
on the actual disks, you may be risking fileystem corruption. The |
49 |
Device Mapper kernel subsystem (dm-crypt, dm-raid, LVM, etc.) does |
50 |
not support write barriers - but neither does MD RAID except for |
51 |
RAID1, so write caching is dangerous except for filesystems directly |
52 |
on disk partitions or on RAID1 (if the RAID1 is directly on disk |
53 |
partitions). |
54 |
|
55 |
I personally decided against using LVM because from what I read it's |
56 |
difficult to correctly stripe-align LVM, and incorrect alignment can |
57 |
have a very big performance impact. |
58 |
|
59 |
|
60 |
Good luck, |
61 |
Conway S. Smith |
62 |
-- |
63 |
The only "intuitive" interface is the nipple. After that, it's all |
64 |
learned. (Bruce Ediger, bediger@××××××××.org, in comp.os.linux.misc, |
65 |
on X interfaces.) |