monkeyboy said:
use: medical research data, including MR imaging data, genomic sequencing data, microscopy imaging data, etc...
So your data would cost quite a lot of man hours/$ to re-generate then? That data sounds kind of important, if it was me, I wouldn't be trying to cheap out on it too much - I'd focus on attempting to provide the most reliable solution possible and if the budget is too big, then work out ways it can be trimmed. But yes, unless you can answer the question of approximately how much your data costs to generate, determining that protecting it is "too expensive" is premature. I'd get the bean counters involved to work out how much it would cost if the data were to go away - then work out some figures on hardware failure rates, other catastrophic events (building fire, etc.) and ask management what they want to protect against - and outline the exclusions in your project scope. You don't want to be the one holding the ball when it all goes pear-shaped and they say "Why didn't you prevent this?". Much better if you can reply that they signed off that protecting against that particular risk was deemed too expensive. Don't forget downtime due to time-to-recover from disaster - it's all well and good to back up to tape or what have you, but if you will go out of business before you can recover from failure, there's no point.
If you go in low-ball initially, they'll try and cut it anyway and you'll end up compromising too much. Every time IT wants to spend money it is always too much unless management are convinced properly
In my experience, if you present the business case appropriately most sane management want to protect themselves. - but may not have sat down and actually thought about the risks (we pay IT to just make magic happen line of thinking).
NetApp et al... I don't think the budget is there for systems like those... I've been looking at Synology, which is probably doable budget-wise...
Who's setting the budget? This a business decision of risk vs. cost to mitigate. If you're comparing NetApp or EMC vs. roll-your-own, are you including the wages of a full-time on-call storage administrator to build and maintain the system (including testing of new gear before rolling to production, etc.)? If being storage administrator is not your job, this will be a significant impact on your work - which is a real cost.
Do you actually need the 500TB to be live, or is a lot of it archive-able? Are there any performance requirements?
Not necessarily saying don't go ZFS on FreeBSD, but if you're doing comparisons, you need to compare like for like... if you get hit by a bus (or go on holiday or whatever) will the business be able to support the solution in your absence? Will it have redundant controllers? Etc.
Whether it is NetApp, EMC, ixSystems, or whoever, there will likely be costs beyond the cost of purchasing the hardware that should be taken into account.
edit:
I realise I haven't answered much here, but this is because a lot of it is a case of "well, that depends...".
Hopefully some of the questions above have given you some things to consider and ask management about - at the end of the day, it's their data. If they do not value it, and don't want to spend, then so be it. So long as you make them aware of the trade-offs.
Backups: restoring 500 TB from tape will take weeks. It also won't be cheap (I'm guessing what... $3000-5000 in tapes alone for a single full backup). If that time-frame is unacceptable, you'll be forced to do disk to disk, preferably to another dataset in a different building, cloud service offering, etc.
