Hi all,

I'm trying to spec a backup system for our department. 4TB drives have come down in price, so it makes sense to load a 24 disk enclosure with them. (I understand this doesn't equate to 96TB for several reasons.)

The system will live mostly with several concurrent rsync/ssh sessions, mostly writing, and rm'ing.

* What's a reasonable amount of memory for such a system?
* What's the best raid level -- 50, 60, other? Or does it make sense to have multiple volumes?
* What's the best filesystem -- ext4 or is the native Linux ZFS now stable enough to use in a critical system?
* Is there any benefit to using two hardware RAID controllers, or just one high end?

Thank you!

Jameson

Hello Jameson,

Backup solutions vary greatly in how they are used.

Keep in mind -- RAID IS NOT A BACKUP.... Be sure that really critical data is still synced offsite.

1) Amount of memory will vary, if you are doing a nightly back-up on a cron job and the data isn't being accessed on a regular basis and you aren't writing huge chunks of data at a time you could probably get by with a considerable small amount of RAM (4-8GB). Memory though is fairly inexpensive these days, and if you are wanting the best performance I would probably go with 24GB or more.

2) If I was doing this I would use a hardware RAID card and go with a RAID 6 array. Software RAID is great but a hardware RAID card with monitoring functionality would be ideal for use in a business scenerio.

3) I would probably go ext4, others might argue but I would say that ZFS likely isn't stable enough to use in a critical system.

4) I would do a single high-end hardware RAID controller -- or at least that is what I did for the last back-up server I built.

You'll get a ton of opinions on this one .. personally I'd probably go with:

16GB RAM
4 x 6 RAID5 volumes (you'll only lose 4 disks worth of capacity)
If the enclosure supports it go with 2 raid controllers, double the bandwidth is better if you have multiple writers
Filesystem choices may depend on the nature of the files you're backing up, ext4 is pretty good all round but there's also XFS


good luck

ZFS is perfectly stable but you will need a ridiculous amount of memory (in the 100s of GB) for it to work well.
I would have said raid6 is a good way to go but the rebuild time on a 24 disk array of 4TB disks will be very long. I would suggest splitting up in to two 12 disk arrays, or even four 6 disk arrays! Hardware Raid6 oR ZFS is definitely the way to go though, just ensure you spec ZFS with sufficient memory because you'll need a lot of it. Also, if you are looking at ZFS you probably shouldn't be looking at linux, the native implementation is on BSD/Solaris. If you are looking for something similar on linux btrfs may be of interest, although its probably not stable enough yet. Also I believe the implementation of raid6 on btrfs was only released on Monday in kernel 3.12-rc2