I have a "combative attitude"? Maybe, if you find yourself on the side of being wrong.
I'm trying to sound "more informed than I actually am"?
Well. Let's see. I've been doing this big storage stuff like forever. I'm well known in the FreeBSD community for disk related stuff. As a FreeBSD developer, see
http://www.freebsd.org/doc/en/articles/contributors/contrib-develalumni.html, I contributed things such as the first implementation of the noatime flag on FreeBSD, did extensive testing and debugging of many things related to huge DAS arrays on FreeBSD, well documented on the mailing lists, and contributed tons of performance tuning advice to the community, much of which remains the gold standard, such as the comments in ccd regarding interleave. I'm well known as the primary developer behind Diablo, the Usenet news package used by numerous service providers, who compete with each other on how many petabytes of storage they currently offer. It's not too hard to check my references.
So here are the facts.
1) Shelves of disk installed by major array vendors will come populated with a single manufacturer's drives, if you allow them to do that to you. Now, if you're small, they may simply choose to ignore your business if you insist on heterogeneous storage. But big customers can and do do it. Even if you don't, it isn't that unusual to find that your storage vendor was decent enough to use drives sourced from two or more batches in building your array, which would appear to be tacit acknowledgment that using all one drive from a single batch is a really bad idea.
2) As I previously said, no, homogeneous arrays do not "constantly lose customer data as my opinion would suggest", which I
didn't actually suggest. The industry *knows* that there are high failure rates among certain models and vintages of hard drives. I can point to numerous discussions of this. If this year's Barracuda has some bad runs, your array built out of WD Black's isn't going to be more unreliable because of it. But out of the population of arrays that are
made out of Barracudas, you are likely to have an artificially high rate of data loss, because some sets of drives are statistically more prone to fail.
3) Even the storage vendors have researched this. Consider for example, a commonly quoted authority on the subject, NetApp, who submitted the following a while back.
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1285441
4) We're talking about FreeNAS best practices, here, not vendor-supplied arrays. For a FreeNAS deployment, the majority of users are free to pick and choose their hardware. Many of them cannot afford to be building excessive amounts of backup and redundancy into their systems, and have come to FreeNAS in order to protect their data with ZFS. While it is very unlikely that a user who has a small quantity of drives in RAIDZ2 will lose enough drives to render the pool useless, the fact of the matter is that you can reduce that risk further through careful selection of hardware. Some of these guys are buying 24-drive chassis like the Norco, and at that point, it's definitely worth looking at doing anything and everything you can to protect your data.
5) FreeNAS users themselves report a lot of familiarity with certain brands/batches of drives failing. I mean, geez, read the forums. Check out
http://forums.freenas.org/showthread.php?577-And-another-2TB-drive-dies-Samsung-F4-this-time for example.
Or perhaps I'm uninformed, Network Appliance's uninformed, Google's uninformed, FreeNAS users are uninformed, and you're here to enlighten us as to how our ways are stupid and we're wrong to buy a mix of drives so that it's easier for us to .. uh, how did you put it? "There's the same number to call for a replacement."
I guess I'm stupid as I can't even begin to figure out how that's meaningful. I would have thought that the user would pull the drive, look at the manufacturer's name, go to their web site, and fill out an RMA request. Since you're not RMA'ing the *other* drives in the array, the utility of having the "same number to call" for all your drives seems irrelevant. Please enlighten us as to how this is relevant.
(well, hey, you did accuse me of a combative attitude, I aim to please.)
But seriously, now, my point is to avoid giving people advice that's commonly understood to be bad. Telling someone to buy drives of all the same kind because it's *easier* (your word, not mine), well, it'll work out often enough, it's a minor risk, except for when you actually do experience a multiple failure. I mean, the number of times my life has been saved by a seatbelt in my car is zero. Therefore, I don't need to wear a seatbelt, right? It's that same sort of thing. You can likely survive just fine buying homogeneous arrays of matched drives. But I'm going to advise people that this is risky, and I'm certainly going to speak up when someone actually suggests it as though it's a best practice.
You might find that going to JBOD or passthru mode "fixes" whatever the problem is, anyways, so it's worth a shot going that route.
