I’ve noticed they are FAR cheaper than anything else right now - just €12/TB.
But they are desktop drives, with volatile reliability (in the past), rated for 8hr/day use and with only 2yr warranty. But load/unload cycle count is identical as with enterprise drives ( 600K cycles).
What are your experiences that you’d care to share ?
How much of it really matters and how much is just manufacturer’s ass-covering?
Is there really big difference in spindle bearing quality ?
FWIW, I intend to place them in DIY harness with significant mass, isolated from each other (vibrations,shocks etc) and cooled to a relatively constant, low temperature 24/7, so they shouldn’t endure anything like enterprise drives see in datacenters…
The barracuda is most likely a smr drive. Smr drives and raid do not play together well (terrible write performance). Smr stands for shingled magnetic recording which lays down the data similar to shingles on a home. This allows for more data per platter and thus is cheaper to produce.
Cmr (conventional magnetic recording) is the older standard which works well with raid.
The primary ass covering’s Barracudas are specified for sustained read. Haven’t worked with any of them over 8 TB but the smaller models’ll drop down to ~30-40 MB/s write.
Ironwolf and Exos are rated for sustained transfer rate and the larger ones’ll hold a steady 250+ MB/s write across much of the patters.
No data. Closest thing’s Backblaze’s migration off 1-4 TB consumer drives to 8+ TB enterprise drives, on which the spread’s from parity with enterprise models to complete drive mortality in three months. Anticipating an order of magnitude higher failure rate’s probably reasonable.
If it’s low WRL, you don’t care about write perf, and the drives are spun down most of the time, might be ok.
I suspect that might be, at least in some cases, conservative.
I’ve been using a bunch of old cheap desktop Samsungs 2TB for probably more than a decade by now. All running 24/7/365, albeight rarelyunder load.
IIRC one or two have died during that time, but even that was, I suspect at least partially my fault.
Rest of them are running just fine, albeight some of them accumulated some SMART failures due to unstable power supply, frequent outages and occasioonal indirect lightning strike damage on other elcheapo HW.
It’s not complicated. Seagate’s specs on the 16, 20, 24 TB Barracudas are clear about 120 TB/year WRL and 9x5 power on. Since you’re intending to operate outside at least one of those parameters, expect elevated failure rates.
Yes, but those are for typical office conditions.
I plan to run them in far better conditions. Much more constant temperature and humidity, fare less virbrations, practically no power cycling etc.
Those desktop 2TB Samsung that I’ve been using for far more than a decade have had similar specs I suspect. And yet they are running just fine.
With any non-NAS, non-enterprise drives, you’re going to need to make sure you set the TLER/SCT/ERC timeout for use in RAID. Years ago, some desktop drives didn’t allow changing the default. This official Seagate rep said a few years back you can set it, but need to do so after every reboot.
I tried an 8TB Seagate Barracuda years back, and found it massively stalled out if it was busy reading/writing data when a probe requesting the drive temperature came in. That was bad…
I’d go for a used/refurb Ultrastar drive with 5yr warranty from somewhere reputable like goharddrive.com before I’d consider messing around with (new) desktop drive.
For which there’s no evidence of making a difference, within reasonable limits. Looks like this just a thread seeking confirmations to buy Barracudas, though, so carry on. Realistically you’re likely to pay one way or another.
Yeah, same. It’s only a few hours of lost productivity to take out a ~€8/TB savings, so the refurb/recert drives I can get’d be in the black pretty quick just on write speed.
Not at all. WRT vibrations and temp shocks, well it worked for me.
before I did that I had plenty of failures. After that, MUCH less.
What’s so unbelievable about HDD being susceptible to those ?
FFS, there is a YT video with a guy disrupting work of a bunch of HDDs in a server just by YELLING loudly into them.
Given the track geometry, but size etc its totally understandable.
Manufactuers don’t boast about head stabilization novelties, especially in enterprise models for nothing.
Also, it seems that a differrentiation between lines WRT maximal number of drives in a bunch is also there for a reason.
Also I have reservations about posted statistics:
Those boxes don’t represent my environment, where shocks are possible. One disks are in a gigantic box in a rack, shock are much less of an issue.
They usually don’t keep the same disk in the box for 10 or 15 years.
SOme models have horrible death rate. Who’s to say it’s not because high temps and vibrations ?
Those are THEIR studies.
What makes you think they are objective ?
And even if so, what makes you think you can seamlessly transfer their use case into yours ?
BTW, how many datacenters have humidity and temperature variations as homelabs ?
Last but not least: Datacenters order and get HDDs by the shitton, carefully delivered in a container and packed in boxes with original, plentiful padding. Your order of few drives was probably delivered by FedEx and packed in improvized package while their disgruntled, underpaid employees played soccer with the package.
I suppose I can’t prove they are being completely objective, but they do present real world data in a non-suspicious way that says vibrations negatively affect HDDs which aligns with my real world experience.
The datacenters are pretty good at keeping humidity and temperature constant which can help, but the super dense LFF chassis they tend to use probably subject the drives to more vibrations than a “typical” home lab environment would do to the drives.
^^ people don’t factor this into their homelab’s reliability enough. I’m at the point where I won’t order less than ~12 HDDs at a time to ensure that they come in the factory carton.
Might be the other way around, given the ones I can think of offhand mass 1+ kg/drive, leading to 100+ kg arrays. It’s Ironwolf and Red where vibration sensors and unlimited bay counts remain speced, not Exos and Ultrastar. Or Barracuda.
Be interesting to load both a small box NAS and rackmount chassis with sets of NAS or datacenter drives and see if there’s any measurable difference. We’ve got most of those combinations and haven’t had any trouble sustaining expected perf but a formal study with 32+ drives would hit some corner cases we don’t have deployed.
Most denser rack spaces I’ve been in require ear protection, so it’s not like drives don’t normally run under substantial acoustic as well as mechanical vibration. External noise is probably lower than an actuator’s own shockwave, distinction is it’s coming from outside the arm’s control loop. Which is what the sensors are for.
Seems to me if one’s planning to put up a drive array the default selection would be drives meant to run in arrays. From what I can tell, consumer drive lines probably assume single active bay, maybe two, but not RAID10, 5, or 6.
Not so often an option for us as a lot of our buys are smaller. I’d like think I’ve got purchasing to quit ordering from about half the places they were using. The drives I put in for are scoped to preferred suppliers, anyways.
This is why I love group buys. People meet on a forum, agree on a model and someone orders for a group. They split a bill and save on the postage.
On top of that group buy usually happens when someone finds especially great price.
Haven’t done it myself, but have seen people doing it with great results.
I’ve had issues with them in the past. Selling refurbs as new. The alleged new drives had a do not remove sticker on them. Peeled that back revealed a refurbished by sticker under it…
