1 |
Am 27.12.2012 01:18, schrieb Alan McKinnon: |
2 |
|
3 |
> I am *very* impressed with ZFS for this. Yes, I know, it's not really |
4 |
> there on Linux - I use it on FreeBSD (FreeNAS). |
5 |
> |
6 |
> It has everything I've wanted in a filesystem for a long time, and all |
7 |
> the crap I've stuffed into my head over many years related to storage |
8 |
> just goes away. It doesn't go to some place I don't have to deal with |
9 |
> it, it just ceases to exist. Very nice. |
10 |
> |
11 |
> There is no more weird partitions from the days of DOS, no PV/VG/LV to |
12 |
> remember the details of. There is only storage and ZFS knows what I |
13 |
> want to happen with each "chunk" of it. A "chunk" (my term) in this |
14 |
> context is a directory and everything below it. |
15 |
> |
16 |
> ZFS doesn't have partitions and filesystems. It has volumes. A volume |
17 |
> is sort of a cross between a filesystem (you mount it and can assign |
18 |
> quotas to it) and a directory (you assign permissions and ownerships to |
19 |
> it). You can overcommit storage space and quotas - you do not get "disk |
20 |
> full" errors and three days of nightmares while you figure out how to |
21 |
> deal with this. the FS just tells you it used more than the allocated |
22 |
> space and keeps telling you till you get it under the limit. |
23 |
> |
24 |
> mv'ing a few TB of video to a different FS to free up space is not fun |
25 |
> at all, but with ZFS it's like an mv on the same FS (that volume thing |
26 |
> again). It checksums every write and lets you know if things fail. It |
27 |
> has proper snapshots built in - that's proper as in copy-on-write so |
28 |
> they don't really take up space until you start modifying files. Your |
29 |
> media collection is like mine - I only add to it and seldom delete, so |
30 |
> I have months of snapshots that consume about 1% extras space. Dale's |
31 |
> rm problem cannot happen to me anymore hehehehe ;-) |
32 |
> |
33 |
> In summary, it does everything I want and does it well. It can also do |
34 |
> other things I don't want but others might (eg de-dupe). |
35 |
|
36 |
I also used ZFS for that kind of storage and it was very pleasant to |
37 |
work with. Performance is an issue if you use zfs-fuse for example, |
38 |
although there are people running zfs-on-linux on LUKS for their main |
39 |
working machine (using an SSD, that helps!): |
40 |
|
41 |
https://mthode.org/ |
42 |
|
43 |
describes such a setup (although slightly off-topic in context of media |
44 |
libraries). |
45 |
|
46 |
I had the pleasure to write a feature about that and provide a demo-vm: |
47 |
|
48 |
http://www.oops.co.at/en/publications/beitrag-fuer-linux-magazin-012013-zfs-und-luks |
49 |
|
50 |
(german feature, sorry ...) |
51 |
|
52 |
- |
53 |
|
54 |
Using snapshots with ZFS really is fun and let's you rethink stuff. It |
55 |
makes doing backups easier and you can send and receive them via pipes |
56 |
(think ssh here). |
57 |
|
58 |
ZFS thinks of storage in the way we think of RAM: plug in some more and |
59 |
everything (as in "every filesystem") is able to use it, just a pool of |
60 |
ressources (actually it uses that term all over, "a zfs pool called tank"). |
61 |
|
62 |
I'd love to use it as root-fs sometimes, but I still hesitate. Might be |
63 |
better to stay with the filesystems most linux-users use, just to |
64 |
benefit of the huge tester-group :-) |
65 |
|
66 |
For media-storage I wouldn't hesitate to run zfs-on-linux on a mirror of |
67 |
2 disks or so. With that you benefit of the so-called self-healing: |
68 |
|
69 |
If the checksum for block X on disk sda isn't correct, very likely the |
70 |
checksum for block X on disk sdb is still valid. ZFS sees that, creates |
71 |
a new block X on sda, with the correct content&checksum and drops the |
72 |
corrupted block (and doesn't reuse it, AFAIK). |
73 |
|
74 |
That is a great feature ... maybe not too important for video where the |
75 |
occasional bit error isn't that much of a problem. But good to have for |
76 |
other stuff ... |
77 |
|
78 |
Greets, Stefan |