Originally Posted by Hillskill
RAID10 would definitely be of benefit to you if you network can make use of the bandwidth. Just be aware that you will lose around 50% of your storage capacity with RAID10 as apposed to 25% with RAID5 (is 25% right bob?). RAID10 is a bloody expensive option but if you images are important then protection against two drive failures would give you piece of mind. I don't see video from Usenet and/or torrents as critical data. Some users will but I am not of that school of thought.
Personally I just made a little WIndows 8 box that does the NAS duties but also does the sickbeard, couchpotato and utorrent management. Its a 1156 dual core box so has enough muscle to unrar and par easily enough but is also low powered enough as not to ring up massive power bills. I used 5 x 3TB WD Reds in a RAID5 array. All data is automoved from a "landing pad" drive to locations on the RAID array. The box pretty much runs itself and I love it. I don't even have a monitor hooked up. I just VNC in when needed. I have it all living in a Fractal Array, a case not to dissimilar to the one in tycoonbobs signature.
RAID10 you will loose half of raw storage (2/n), where RAID5 you will loose 1 drive of raw storage (1 - 1/n). For example, 4 2TB drives in RAID10 yield ~4TB storage. 4 2TB drives in RAID5 yield ~6TB storage. RAID5 will allow for 1 drive to fall out of the array, but the second will destroy the array. Parity calculations will also decrease your writes, which you will see when copying over several RAW images at a time. RAID10 will allow for 2 drives to fall out of the array, and the third will kill it (caveat being 2 drives can fail, but must be from different subarrays), but rebuilding a RAID10 is so much easier since there is no parity calculations. If you had 8 3TB drives, you would get a RAID5 of ~21TB, and a RAID10 of ~12TB. Yes, that is a huge difference. However, if you loose a drive from a RAID5 with 8 drives and 21TB of storage, statistically speaking, the array will encounter a URE (Unrecoverable Read Error) before that rebuilt can finish. Most consumer level drives have a URE rate of 10^14, which is ~9TB. During a RAID array rebuild (when using a RAID level that uses parity such as 5, 6, 50, or 60), ALL bits have to be reallocated throughout the ENTIRE array, meaning all drives will be working hard and moving around every bit of data. That means every bit of data on your ~21TB array will be read, and assuming you had ~15TB of data on that array at that time, statistically speaking you will will encounter 1.5 or so UREs. Bad bad bad, and not worth the risk in my opinion. RAID6 can mitigate this, but you will experience even slower write speeds with those RAW images.
Thus, RAID10 is the best suggestion, if you can look past the decreased capacity. Seriously though, you don't design a storage solution for larger than the amount of data you think you will accumulate over the next 5 or so years. Most consumer drives and hardware have a warranty of 3 years or so, and most consumer HDDs will (statistically) fail within 5 years. Also, in 5 years we could have 6TB or even 8TB drives, and some new storage method (to replace RAID), so keep that in mind as well. When designing a storage solution, you do so for the short term. If you need long term storage, look toward DVDs and/or Tapes.
I have a feeling that 8 3TB (or even 4TB) in RAID10 should do most people for the next few years, and you have the option to start out with 4 drives and expand (2 at a time) as needed. Most decent RAID controllers (MegaRAID 9260-8i, for example) will cost you about $300 or use used, if you went that route. Expanding a RAID10 array is low risk since there is no parity calculations. Simple mirror and stripe. I, personally, use a 9261-8i controller with an HP SAS Expander in a Norco RPC-4224 chassis, and am currently running a RAID5 (yes, I know -- I just said that was bad), but am in the process of building my new sig rig (parts come in today!) then I will be buying enough drives to start a 6 drive RAID10 with 3TB drives (Toshiba DT01ACA300) and moving my data to that. I will feel so much safer about my data at that point.
Also, it's worth mentioning about cloud/offsite backups. AltDrive and CrashPlan both offer unlimited storage for around $5/mo (AltDrive is a little cheaper at ~$45/year). Unlimited storage, bandwidth, and file size. Your only real limits are your upload speeds. Honestly, after the initial data upload, it's not all that bad (I use AltDrive myself).
I do agree that videos from Usenet/Torrents are not "critical" data, even though I back all mine up to AltDrive. I have a VM on one of my C1100s that runs SAB/SickBeard/CouchPotato, so it does all the heavy lifting. I have a 200GB VHDx connected to that VM via iSCSI, which is the download directory. Post processing will extract, rename, and move it to the appropriate directories outside of that iSCSI volume. Music on the other hand, I do consider critical. I have ~15,000 songs with ~98% being FLAC rips. All my music is properly tagged with embedded artwork, and if I lost all that work, I would never use a computer again. I have local backups I use for my critical data (music, personal docs, images, etc), while all data is uploaded to AltDrive. Critical data, there are 3 copies while non-critical data there are 2.
Hardware is completely your call though. When it comes to ZFS, best case scenario is running hardware that supports ECC RAM, and using ECC RAM (~1GB of RAM per 1TB of data being stored on ZFS). If you go hardware RAID, I'd suggest a server board (ASUS, SuperMicro, etc) for better support of a RAID controller, and usually better onboard NICs. CPU wise, I use a Xeon 1220v2 in my storage box (along with 16GB of RAM). I wish I would have went with the 1230v2, so I could have gotten hyperthreading for $10 more, but honestly, that CPU averages around 5% utilization. It has no function other than storage.
Hopefully this gives you some insight and thoughts leading to more questions. I take storage very seriously, as should everyone.