File server planning

[Stilez]

I'm looking for input in rethinking my SOHO file server approach, and specifically what platform to go for.

The server has 2 x 3tb and 2 x 4tb data drives. They're mirrored - speedy is nice but the bigger priority is protecting against disk failure, especially with warranties down to 1-3 years. It's pure RAID-1 for the same reason - if a 4TB disk goes, there isn't a striped 3TB that needs resyncing so vulnerability is reduced. Also means I can power down the server and plug one of the mirror into my desktop for copying or as read-only to grab a critical file, as happened when the server's motherboard failed last year.

As well as fileserving, the server does duty as a windows update server, media player, and a few other low-effort things. Currently it's been using Win7 x64 but I've been trialling Server 2012, jury's out though as it's had minor glitches and it needed a few drivers and apps beaten into submission to persuade them to run on a server version of Win8 (*cough*Intel 82579V*cough*desktop antivirus*cough*). Overall both are viable.

But I'm not really happy. There's a lot of old data and bit rot/chkdsk/data loss in resync etc are a concern. ReFS has some concerns still (some reports of issues) so I'm wondering about a ZFS distro, with VMware or VirtualPC on it for the few Windows functions needed. Specifically, I'm considering a mature open source file server distro (Nextenta or OpenNAS or equivalent I guess?). The platform is 100% Intel + ICH10R + i5-2500S + 8GB so it's got good NICs and stability, plenty of pulling power and some spare, deliberately specced to be able to run as a VM server if needed. Getting the last 10% of speed isn't too important, anyway. (My desktop on the other hand... )

I have no *nix experience but a decent server distro should be intuitive or well documented. I also have no experience of ZFS (although I know its reputation), I'm used to NTFS speeds only. I have a spare couple of 1TB drives if more storage is any use for parity etc. Financially I'm limited in terms of offsite backup which is a risk (!(money + trees) ? 'take chance' : 'relax').

I'd like to hear views on my different options, and if I'm making the best for this situation.

I don't know what questions I need to ask, but there should be enough info there to give a good idea of the situation, and if what I'm doing is best.

Thank you for your time!

 

[BecauseScience]

I have no *nix experience but a decent server distro should be intuitive or well documented.

With zero unix experience, it would be a bad idea to put important files on a diy unix nas. There are nas gui's for unix systems but they don't (can't) possibly cover everything you might need to do. I would buy a commercial nas and get started on your unix skills. Do a big diy nas after you know your way around linux/solaris/freebsd.

 

[JoeComp]

What sort of files are you storing on this file server? Is it mostly media files and/or files that are read-only after they are added to the file server?

If so, you should consider snapshot RAID, using either SnapRAID or FlexRAID.

 

[kac77]

Even with no experience there are tons of step by step instructions out there. FreeNas in particular gives you a GUI right away. If you are starting from scratch I see no reason not to go with something like FreeNas, ZoL or OI.

 

[_Gea]

If you care about data security and availability, you should

- not use anything beside ZFS
- use a Raid Level that can survive at least two disk failures (3way mirror, Z2 ot Z3)
- do backups and keep snapshots of old date (in case of copying bad data over good data in backups)

if you are not experienced with ...X and ZFS, use an appliance distribution (ready to use web-managed storaga server) like :

FreeBSD based
- FreeNas
- Nas4free
- ZFSGuru

or Solaris based (Illumian, OmniOS, OpenIndiana, Oracle Solaris)
- napp-it

In case of my napp-it, you can even buy a HP Microserver (180 Euro, or some more highend SuperMicros),
add up to 4 disks and a USB stick, download and clone a preconfigured USB stick image to stick and your NAS or SAN is ready to use via Web-UI.

 

[hotcrandel]

What you have sounds like a pretty good bet, especially if you're comfortable with the operating system you're using. Simple RAID1 is great. I prefer file-level syncing (rsync or equivelent) when its two disks and you don't necessarily need the uptime.

We would always encourage you to make sure you have backups, RAID is not a backup.
If you're having bit-rot backups may have prevented you from losing files/data.

Spending time learning new OS = spending money learning new OS, so be ready to invest some time in learning something like FreeNAS.

As simple as it as and as much as I love it, it has a learning curve, is far from universal about hardware, and you will get a lot more out of it being comfortable with the CLI.

Flexraid and other drive pooling solutions are also really great and might take you further to where you want to go.

 

[Stilez]

Wow, some good answers, thank you. Comments on them:

@kristof:
I'd be seeing this much like a capable router or other firmware appliance. I don't necessarily need to know what's in the box to be viable, though I do to make best use of it. Since most NAS are specialised *nix distros anyway, I can't see that a mature soft NAS distro would automatically be less likely to be usable without experience than a formal manufactured NAS.

Also with a mature, actively developed, dedicated NFS distro, a lot of the basic setup and security is already configured by design because it doesn't have to be left wide open, and the GUI usually covers the remaining choices, so there's less pressure to know it all up front. Or don't you think so?

@JoeComp:
Good question. 90% of files will be read-only - software I use, drivers, AVI/MP3/books/PDFs, desktop system images and backups, etc. But not entirely read-only, I modify files as needed too, or remove them when no possibility of needing them, or rename them... or for media, sometimes batch process them. There may also be a few VMs run on the server (eg if it's a *nix platform windows update service would need to be VM'ed) and those would be on the same disks for redundancy. Also it might help to comment on deduplication and long paths:

• There are repeated backups of "my docs", desktop/latops, adobe elements photos etc, so there is good scope for deduplication as well as checksumming/scrubbing for bit rot. It's not essential, I won't fill my current disks for a long time even without deduplication. But there are a lot of duplicates in those kinds of folders. Dedup or incremental handling/versioning would be possible, if reliable. But multiple copies are ok too.
• Some files/folders have long NTFS filenames (the "right" API uses 32k filepaths but too many progs use the "wrong" one). In some folders I use very descriptive filenames in lieu of tagging, explaining exactly their contents, which version etc they are of working files or what's working/what isn't. It's great when checking which version of an old file to revert to, but the 260+ long paths aren't ideal when nested in a backup hierarchy (\backup\user\date\...). It works so I can live with it, but not ideal.

@_Gea:
Snapshotting sounds very desirable (as I conceive it, it means that within the limits of disk capacity, files can be updated but at any time you can ask the system to show you read-only older versions of files).

I like

@hotcrandel:
Not sure I agree that external backups preserve against bitrot. Bitrot is a function of "amount of data" vs "risk of data losing magnetic state". Whether same disk or not, it's a statistical decay process. The only safeguards for bit rot are multiple copies and checksums/error-correcting storage, but those can be same disk or different disk. I also run my disks cooled and spaced out which may help (~ 28 - 37C).

Raid isn't backup, as you say. I reckon my main risks to be:

• Single disk failure (highest risk in practice, mitigated by mirroring, would have to be very unlucky not to have 2-3 hours to get data off the partner as soon as it happens),
• PSU failure killing multiple drives simultaneously (hopefully mitigated by good quality PSU - Antec Truepower 750 - that hopefully is designed to blow up without sending huge voltage spikes to all drives ), and
• Malicious activity (hopefully unlikely and sufficient precautions, can't ever tell).
• Other:
  Accidental overwrite of files/folders and
  Major user error (erasing volumes etc)
  Mass loss/gradual accumulated corruption is the issue. Losing a few individual/isolated files due to mistake is unlikely to be devastating, and I'm not too prone to user error (deleting root folders etc). So I see these as lesser risk. Multiple backups on multiple machines will probably exist of some files at least, and many files on the server are themselves backups of files already backed up on individual PCs. So to an extent independent backup already exists. More consistent offline backups would be good, but money + trees, considering the most likely risks first.

I also agree that long term, learning the platform will be best. I do plan to become more used to *nix over time. I'd never run a system I don't know enough to maintain, manage and secure - and there's a lot to learn about that on *nix. But that will take time, and maybe I'll never be able to secure a platform as well as those creating NFS distros would do. But I agree on the investing of time