Hi all, I'm currently going through one of those "I told you so" traumatic events with my RAID 5, and given that I'm probably going to buy all new hardware with the exception of my disks, I'm considering moving away from my RAID 5 on my crappy old highpoint 3030. Any storage experts have any thoughts on the options in the title? I understand that conventional wisdom in storage these days says that RAID 5 is too risky due to URE's, and that RAID 6 is kind of acceptable so long as running under a 20TB array (which I will be until about 2016-2018). RaidZ has evidently been surpassed by Z2 but I'm not totally up on that, while unRAID seems really nice and simple but seems to rely on backing up just as much as RAID 5 but just generally be convenient in its expansion capabilities. I like data security and I don't need speed so I'm leaning towards ZFS and some iteration of RAID-Z despite being completely ignorant about linux, but I'd welcome any experiences or input others have. I tend to run a 3-8 disk array, generally upgrading disk sizes entirely when I hit 6 or 7 disks, and it's a headless home file server running purely for data storage so throughput and so forth aren't hugely relevant (capped by gbit anyway), data survivability and ease of use are the two biggies, in that order.
RaidZ is pretty painless on nas4free. You need quite a lot of memory although this is somewhat tuneable. I use raidz not z2 as I have a mirror setup on another nas4free with a nightly rsync. You do need to set up cron jobs for scrubbing the array once setup - this I believe is part of the error correction etc process. These intervals being dependent on with the drive is doing any error checking itself or not I believe from my hazy memory. One of the cool things about zfs other than protection against the odd flipped bit is the snapshot ability. So take a snapshot before doing something silly and you can restore the filesystem to an earlier time. You can also improve performance and I believe resilience somewhat by using read cache ssds and writes with a zfs intent log. But that is way beyond the scope of my home usage, so I have never bothered reading too much about it. Personally I can't see the point in hw raid... But you wait, somebody will wade in with 10x more knowledge and experience than me a beat me with their battery backed up £600+ cards; how amazingly redundant they are, until I submit and beg for mercy. Some articles about zfs recently on arstechnica: http://arstechnica.com/information-...-and-atomic-cows-inside-next-gen-filesystems/ And a walkthrough: http://arstechnica.com/information-...h-using-the-zfs-next-gen-filesystem-on-linux/
Ah yeah I'm aware of the battery back up crew, but ignore them. I'm not going to die if I can't access photos from 2009 for a few hours. What sort of parity options do you have with raidz, or do you not bother and just literally do nightly mirrror? Surely that's very expensive).
ZFS is an awesome filesystem. You can have RaidZ1 (one drive worth of parity), Z2 (two drives) or Z3 (three drives). Which one you go for really depends on what data you are storing on it and how much availability you need. Any BSD based operating system (FreeNAS is often mentioned) is the best to use with ZFS, but you can get it working well on Linux, albeit with lower performance. Btrfs is another file system that you might want to look at in the near future. It is based on ext4, but works in a similar way to ZFS in that it has integrated RAID, snapshotting, sub volumes, etc. It is however currently highly experimental, especially the RAID aspect of it, although it is stable enough for production use. As for hardware RAID - in modern systems they are a hindrance, not a benefit. If the RAID card dies, you need to find a compatible card, if not the same card. Right now that might not be a problem, but imagine if it dies 8 years down the line - how are you going to find a RAID card from 8 years ago? Besides, modern CPUs have more than enough power to do the parity calculations for parity RAID without impacting normal processing.
What concerns me with ZFS and raidZ (besides linux and non-NTFS file systems being scary in general) is that from what I've read, you need huge amounts of both RAM and processing power to stop the whole thing grinding along. This is part of my desire to move over now, as I set up this system 8 years ago. While the falling apart mobo and ancient CPU can be easily replaced in any system, the RAID card I'm using hasn't been manufactured in years. Seems like Z2 with backups to an external might be the best idea medium term, so long as it allows for OCE.
I have 3 2tb drives in the zfs raid 5 equivalent raidz. This gives me 3.6tb of space along with the ability to lose one drive. I have a forth drive of 4tb in a separate atom powered machine running just a normal volume with rsync that my nightly backup runs to. That way I can add a mirror to my atom later. For zfs I have 8gb of ram. Just sat there at the moment my usage is at 87% with approx 2tb of data. Now you can tune zfs for lower amounts of memory. I've had no experience of it myself, but my mate used it with 4gb ram and 2tb of disk space with nothing to report negative and wasn't even aware of the tunables available.
Great info, ta. As I've just ordered two extra drives (might be possible to recover my data from the RAID but I need somewhere to dump it) I'd be up to 6, so I guess 4 data 1 parity 1 backup would maybe be the wisest choice at this point. Obviously that only allows for partial back up since all disks are equal in size, but it's a start and it's better than nowt. I've heard the best bet is to go 1 GB ram per 1 TB. I don't get why it needs so much RAM (to contrast, my file server right now only has half a gig.
i believe it is to do with using it as a cache for reads and writes. TBH though they are tunable, and supposedly from what I've read before it mostly benefits concurrent use of smaller files. It may even be that freeBSD based systems don't need any tuning (according to that arstechnica post.) Page 3 of that article is v.interesting.
As far as I'm aware, ZFS uses as much memory as it can. It generally needs quite a bit to work efficiently, but RAM is so cheap nowadays anyway it would seem silly not to get at least 8GB.
Providing you aren't using deduplication, it should be OK. Dedupe is one of those features that may only benefit enterprise environments.
Only just seen this so maybe pointless now however One of the main strong points I like about unraid (which I've used without issues for several years now) is it can survive a single drive failure without data loss, the parity drive can die along side a single data drive without data loss and in a situation where you're unlucky enough to have two data disks die - you only loose data on those two drives. I can't speak about any other systems as I've not used them but unraid is very simple to setup, very easy to expand and in the time I've had it I've had one data drive fail and getting the system back to fully protected was a piece of cake. Also there support forums are one of the best I've used which is an added bonus Its not for everyone but as media storage for steaming around the house and for photo storage its been great.
