We've found its a big problem juggling cost vs desired resilience - after all if money is no object, just build 2 big RAID 10 servers connected to a massive distributed Storage Area Network across fiber linked sites and run Myriad using teamed network cards with dual switches etc
Fine if you are financed by a nice rich Sheikh, not so much for the rest of us...
In practice we're found something much more modest is reasonable.
Internally here we used to use RAID5 with a spare drive all in hot swap caddys (as the caddys are pretty cheap when used with SATA), but to be honest we're now dropped to straight mirrors (some servers have hot spares) as with 2TB drives it's more than enough storage space for our needs.
It's one of those questions that you really need to ask alongside the question of "Great, but what if the box itself dies (i.e. motherboard etc.)" and then just to confuse matters : "What if the building burns down/gets smoke damaged/floods/server is stolen"
When you start to look at that set of situations it can change you're thoughts - afterall, for a fixed budget you may well get better Disaster Resilience (DR) by spending a little less on that one server, and investing in a cheap backup server or large USB drive, and a pair of USB Drives that you rotate off-site.
I appreciate that our budgets here are a little different from others (more in some cases, a LOT less in the cases of some of our customers!) but if it helps anyone, here's how we do it on our 'live test' servers. (I know I'm not really answering the question about drive setup, but I do think that thinking about the wider recovery situations may change your mind on what you want to achieve!)
1) Primary Server has a pair of mirrors in it, one small pair for the operating System (100Gb of usable space or so), the second pair gives us a larger 2TB usable mirror. All 4 drives are in caddys so can be swapped in the event of a failure. We then use Intel Hardare Raid (on the motherboards) to run the mirrors. This gives us a very fast setup with good resilience.
2) Secondary server has a similar drive setup, but isn't such a beefy processor as it's a slightly older server. It does have caddys, but could easily not have them to save a few pennies as it's only being used (in theory!) in an emergency. We use Robocopy to copy data from the primary every 30 mins or so.
3) Backup Server - ok this is where we do start to get a BIT belt and braces I'll admit. We have a very large server in the building running Microsoft DPM Server 2010 that backs up ALL the servers in the building and gives us the last 10 days "snapshots" of every file so we can go a recover any file from any time in the last 10 days. It also would be used to "barebones" recover to a new server if we ever suffer a catastrophic failure of both primary and secondary servers. Most places obviously don't have this, so I'm just including it for completeness.
4) USB Drive : (Very important!) The Secondary Server has a USB backup drive plugged into the back of it - a cheap off the shelf Toshiba unit I think - that stores 2 TB of data on. These drives are Bitlocker (from Microsoft) encrypted, and Robocopy runs every night to copy the contents of the server onto the drive so there is always a copy there. This drive is then taken off site to a secure location once and week and swapped with it's twin which runs for the next week. The key bit is that we NEVER have both of these drives on site together, and where possible we don't transport them at the same time or even have them in the same postcode
These are the drives that give us the "Site Disaster Recovery" - if the building was inaccessible for some reason (fire/flood/pestilence etc.) then we can plug these drives into a suitable PC, obtain the decryption key from our key holder and then we are up and running, at worst we've only lost up to the last 6 days of data. Some people actually use 3 drives for this to keep down the likelihood of ending up with both drives in the same postcode at the same time!
5) What about up to the last 7 days of data? For this we use an backup service which we backup to every day (more often for some critical data) over the internet. This is a slow process as obviously it's limited by the speed of your internet connection, but it means that we are never TOO far from the "current data" It only needs to be the frequently changing data as the USB drives covers the rest, but we are actually aiming to backup the lot in time.
So thats how we do it - and yes it's massively overcomplicated in some ways
If I was doing a backup setup for a radio station and had a bit more limited funding I would build a big serer with a mirror and a hot spare. Then buy 2 or 3 USB drives, backup onto them each night and rotate them on and off site as often as practicable - once a week say.
If a drive dies, the hot spare takes over and I replace the failed drive (make sure you setup the email notifications, otherwise how will you know if it fails!).
If the server dies, I plug the USB drive into my studio computer and change Myriad's file locations.
If the building is out of action, I plug the backup USB drive that I took off site last week into my Laptop, call Helen or Sarah here and get a new license code for Myriad and I'm backup and running before the smoke has cleared.
Anyway, hopefully that will help someone!
(Note: as a side note, some of the data is actually replicated using Microsoft DFS but this won't copy open files so is only used for areas which don't have persistent open files)
------------
Peter Jarrett, Technical Director
Broadcast Radio Ltd.Bill Bailey: No win, no fee, no basis in reality. Just a room above a minicab office in Acton and a steady stream of greedy simpletons whose delusion is only matched by their clumsiness