1 |
On Tue, Apr 28, 2015 at 2:11 PM, Neil Bothwick <neil@××××××××××.uk> wrote: |
2 |
> On Tue, 28 Apr 2015 13:38:55 -0400, Rich Freeman wrote: |
3 |
> |
4 |
>> > The same is also possible with BTRFS, including built in RAID. RAID5 |
5 |
>> > in btrfs is expermiental, but its RAID1 is like RAID5 in some ways, |
6 |
>> > such as giving the capacity of n-1 disks and tolerating a single disk |
7 |
>> > failure. |
8 |
>> |
9 |
>> btrfs raid5 is still fairly experimental (though now it supports |
10 |
>> recovery) and works more-or-less how you'd expect raid5 to work. |
11 |
>> Raid1 on btrfs gives you the capacity of n/2 and not n-1 disks, |
12 |
> |
13 |
> You're right, I was clearly confused (an oxymoron?) when I wrote that. |
14 |
> |
15 |
> So RAID1 gives less capacity than RAID5 on BTRFS, but it is stable (in |
16 |
> btrfs terms). |
17 |
> |
18 |
|
19 |
Correct. |
20 |
|
21 |
For drives of identical size and not using compression, I'd expect |
22 |
space use on btrfs to be equivalent to the same raid level on |
23 |
mdadm+lvm+ext4. With mixed drives you will potentially get more space |
24 |
on btrfs, and compression will of course get you more space. |
25 |
|
26 |
As far as data security goes there is a tradeoff. Btrfs is still |
27 |
immature and I seem to have issues with it 1-2 times per year (but |
28 |
I've yet to have unrecoverable data loss). On the other hand, btrfs |
29 |
does do full data checksumming which means you're less likely to lose |
30 |
data due to issues with the physical storage than with mdadm - as with |
31 |
zfs it always checks the checksum and will recover from another disk |
32 |
if possible, and in the event of raid disparity it always knows which |
33 |
(if any) of the copies is right. |
34 |
|
35 |
I'm hopeful that at some point I'll be able to recommend it without |
36 |
reservation. Right now, that isn't entirely the case. I'm still |
37 |
patching the 3.18 kernel series so that it actually mounts my root |
38 |
partition when whatever is causing it to panic (probably also btrfs) |
39 |
does so (the patch is in the queue, but hasn't made it to 3.18 yet for |
40 |
some reason - I believe it has been in the other stable series for a |
41 |
release or two now). |
42 |
|
43 |
-- |
44 |
Rich |