So I have recently learned a painful lesson: just because data is replaceable, doesn't mean you actually want to replace it if things go bad. Needless to say I will keep backups, but I'm done with my whacky ZFS-Virtual Box with physical disks file server setup (and am in the process of recovering what didn't get corrupted right now).
But I have a new conumdrum: what do I replace it with?
This is a home server, with no UPS, so I'd like to start with the assumption that at some point there will be a power failure.
The plan is to run software RAID6, on 8 1.5TB drives, under Ubuntu. But I don't know what filesystem to replace it with. The internet suggests ext3, XFS and ZFS. ZFS is out since I can't run OpenSolaris natively, and well, there seem to be some pretty nasty failure modes there.
So I'm wondering what the current thing with XFS seems to be (ext3 being the fallback). The internet suggests its reliable, but I see sporadic reports of it going down under heavy NFS loads and the like (but that's 2006 stuff).
Does anyone have any experiences or recommendations? I'm looking for the minimum of hassle.
Posts
I'd ask for his source and post again later.
Also I would imagine the resources available for things that do go wrong are fairly extensive.
I mean, I can understand why many home PC's don't really want to bother... but a UPS that would cover the server would cost about as much as one of those 1.5 TB drives, and would avoid the entire frequent power loss problem
Also I remember reading somewhere native ZFS was available for Linux if you're willing to build it yourself. Don't know how mature or useful it is though.
Confined to a tiny spit of sand, unable to escape,
But tonight, it's heavy stuff.
You can also just use ext3, it's almost the same thing. You can always convert it to ext4 later.
XFS and ext4 both have rumours of "data loss", which stem from the same source: ext3. Some Linux developers got used to a particular quirk of ext2/ext3 called "ordered data writes", and wrote their applications to depend on it, instead of using fsync(). Of course, more modern filesystems don't use ordered writes, so un-synced files written by broken applications were often corrupted during sudden power loss.
Anybody who ran into this problem using XFS blamed the FS, since it's somewhat obscure, and went back to ext3. When ext4 was released and had the same problem, the app devs were mostly forced to fix their shit. Unless you're using old versions of low-quality applications, you probably don't have to worry about data loss in either XFS or ext4.
Also: I tried XFS a year or so ago, and was happy enough with the FS itself, but tool support is poor. It's an unusual FS choice, so many utility developers don't bother testing with it. I ended up completely stymied trying to resize some XFS partitions, gave up, and went back to ext(3/4).
I've heard lots of good things about FreeNAS. If you're wanting the minimum of hassle then freenas is definitely the way to go, I've only played about with it in a vm since the only raids I've set up recently were for high end servers with lots of fancy hardware. It seemed dirt easy to setup and maintain.
---
I've got a spare copy of Portal, if anyone wants it message me.