New FreeNAS new setup questions

[geekyink]

New to FreeNAS as of this week, and I like what I have read so far! Looks like a great community of developers and engineers. I have read the manual and scoured these forums. All situations are different and I am hoping to get some specific recommendations on my needs. Mildly familiar with Linux, but let's just assume I am a newbie all around.

Needs: AFP (sometimes SMB/NFS) file servers with large amounts of storage with some redundancy. Currently I require about 70Tb of production space and then will need to backup the production and have additional snapshot/incremental backup space. My 100 users read/write from the file servers about 10 hours a day with your standard Office/Adobe applications.

Hardware: I have inherited 7 SuperMicro rack mount servers that are all about 32Gb RAM, Dual Intel® 64-bit Xeon 8 CORE with Intel 6port Ethernet cards and Adaptec 5805 RAID cards. Each have 24 4Tb Segate drives for about 90Tb of storage each. I also have plenty of HP 3500 switching equipment for possible ISCSI implementations (although not familiar with that technology either).

Questions:
I understand that I am better off using ZFS (RAIDZ2 I think) instead of the Adaptec RAID controllers. However, I haven't found a clear answer on if I should setup each server as JBOD or just use the Adaptec as a straight SATA controller instead. Any other advantage/disadvantage of the hardware raid cards I am missing, as I already have them? This is my biggest question currently as I don't want to get this part wrong as it affects everything else.

We frequently have to add space to our file servers, so I want to set them up in the best possible way to allow me to incrementally add space (SAN, ZFS?) so it is an easier process and I don't have to move so much data around each time.

As for file server needs, I have two types of data. "Production" that is about 20Tb and in use 24/7. "Archive" which is about 50Tb and more "near line" as it is read only for users to pull old content to reuse. I am not sure of the best ways to use/configure my servers to host that data. Is it better to have one giant file server with multiple share points or to have a separate server for archive and production?

We don't have a backup system currently, we just use RSYNC to clone servers nightly. Recommendations on backup options? Second server with NFS mount and nightly cloning? Does ISCSI come into play here and help at all? I have enough servers that I pretty much have a 1 to 1 match of production/backup space, so what is the best way to utilize that? I know some systems cluster (right word?) where it writes data to two places for backup purposes.

I think that is all for now. I appreciate any answers you can provide and am looking forward to learning more on FreeNAS!

 

[JaimieV]

Set the RAID controllers as straight SATA ports. The reason for your confusion is that some mfr's use "JBOD" to mean that, others mean JBOD for "concatenate all disks into one big virtual one". ZFS likes native control of the disks, so arrange them to be available with no interference from intelligent RAID cards.

If you have to add space "frequently" then either your planning or up-front capital expenditure is probably wrong. Look at that first. After, read Cyberjock's presentation in the sticky at the top of the Noobs forum - use that and your expected space increment size to design the vdev disk loadout units to work with as incremental blocks - eg six disk RAIDZ2 units, 11 disk RAIDZ3, whatever. Your hardware chassis/disk shelves will play into this as well.

I would almost always recommend multiple fileservers rather than one big one. In your case with a strict demarcation between prod and archive, definitely so. You've got stacks of them, too - best use them.

I would always recommend having a backup system, too! rsync works, but to pick a 'real' backup system is way outside the scope of these forums - too dependent on your exact needs and server+network loading, your finances, so on. However to answer the other question: FreeNAS doesn't support clustering directly - you could look into gluster and the like. You can send replicas from machine A to machine B very regularly though - not the same as a moment by moment cluster, but a max 20 minutes out of date near-line backup system is worth its weight in gold when the primary pops.