Review Western Digital Red Pro 20TB HDD Review: OptiNAND on a Platter

[Admin]

Western Digital’s Red Pro NAS line of high-capacity hard drives (HDDs) utilizes NAND flash with its OptiNAND technology to reach 20TB and beyond. It’s CMR technology without the drawbacks of SMR.

Western Digital Red Pro 20TB HDD Review: OptiNAND on a Platter : Read more

 

[jkhoward]

this drive sucks dude… why so many stars?

 

[JarredWaltonGPU]

this drive sucks dude… why so many stars?

"Sucks" is perhaps too harsh. It's a large capacity HDD, at a relatively reasonable price. But we did have the wrong score initially. 3-stars is the correct score now.

 

[atomicWAR]

this drive sucks dude… why so many stars?

I have two myself and would disagree with your summation. Their not perfect and certainly have some dings but they are solid high capacity HDDs to run in your system. My biggest complaint is they are louder than my other drives when they first spin up by a lot. My wife asks at least once a week what that sound is (as they spin up) when watching a movie or gaming.

"Sucks" is perhaps too harsh. It's a large capacity HDD, at a relatively reasonable price. But we did have the wrong score initially. 3-stars is the correct score now.

They don't suck. I agreed with Shane's review and while I may have scored it a half star more at 3 1/2 stars. Just like I thought your review of the 4060 Ti deserved a half star less, you can't make everyone happy with reviews! Thanks Shane for the time and effort you put into the review.

 

[domih]

I use 3 x 10 TB WD Red (not pro, not plus) 5400 RPM for 3 rsync'ed copies as my NAS.

IMHO, the 5400 RPM are a better use for home NAS, they are quasi silent when spinning, and they spin down after a few minutes idle (therefore 100% silent, no power consumption.) They are CMR.

They have been running 24 x 7 for 4+ years using ODROID HC2, https://ameridroid.com/products/odroid-hc2, running Ubuntu Server. No issues so far. Yes, I have a rsync'ed copy in a different geolocation.

With the 3 x ODROID HC2, power usage is minimal (a few watts per unit).

Alternatively, you can go with a HC4, https://www.tomshardware.com/news/odroid-hc4-nas whose drive interface is PCIe native, no USB to SATA bridge. If you do not like the toaster version, then go with the ODROID-HC4-P KIT, https://www.hardkernel.com/shop/odroid-hc4-p-kit/. Similar power usage.

You can have a pretty good enough always on home NAS using ARM-based Small Board Computers (SBC). No need for a full x86 system.

My guess is that the 7200 RPM WD Red Pro are more intended for small businesses or SOHO's where you have more users than a normal home.

 

[tamalero]

Wouldn't that many platters give this hard disk way more points of failure?

I remember back when the first 1.5 TB drives were out. There was a huge failure rate on certain models that had 3+ platters.
And every review site recommended to avoid like a plague

 

[eldakka1]

My concern lies around the longevity of the NAND.

Is the NAND likley to wear out before the useful platter life of the drive, such that a non-OptiNAND drive will just last longer?

What happens when the OptiNAND does wear out on an otherwise fully functional drive?

1. Fails entirely?
2. falls back into a read-only mode (that require no meta-data updates)?
3. fallback to a 'traditional' mode where it has to start storing metadata on the platters with the user data?

Those are pretty important questions for my usage of SATA drives where I'd be using them for years, maybe decades - I'm still using some 1.5TB drives in some arrays that I've had god knows how long (along with 3TB and 8TB drives, and am looking at 16+ TB ones to start the shuffle-down of drives).

 

[JarredWaltonGPU]

My concern lies around the longevity of the NAND.

Is the NAND likley to wear out before the useful platter life of the drive, such that a non-OptiNAND drive will just last longer?

What happens when the OptiNAND does wear out on an otherwise fully functional drive?

1. Fails entirely?
2. falls back into a read-only mode (that require no meta-data updates)?
3. fallback to a 'traditional' mode where it has to start storing metadata on the platters with the user data?

Those are pretty important questions for my usage of SATA drives where I'd be using them for years, maybe decades - I'm still using some 1.5TB drives in some arrays that I've had god knows how long (along with 3TB and 8TB drives, and am looking at 16+ TB ones to start the shuffle-down of drives).

Assuming standard wear leveling algorithms, and the fact that this NAND is only used for tables and such rather than for actual files, I suspect the NAND will probably last several lifetimes in theory. The rest of the drive is far more likely to fail. Also looking at performance, I don't even know that the OptiNAND really helps that much. 🤷‍♂️

 

[helper800]

Also looking at performance, I don't even know that the OptiNAND really helps that much. 🤷‍♂️

I agree, its like they tried to reinvent the wheel and ended up with another wheel...

 

[abufrejoval]

Honestly, I see a scandal cooking, like the SMR/CMR fiasco!

The least problematic effect of that flash would simply be what we see in the tests: it doesn't do any harm.

It has some extra vaguely "premium" sounding extra attribute called OptiNand and that's it. Just lures some naive buyers into choosing this over an otherwise indistinguishable competitor.

If the flash was good enough to serve as an in-line write-back cache, the benefits of that are usually too limited at the capacity vendors can support, but NAND is getting cheaper...

But on a CMR hard drive, there simply is no need nor benefit to handle larges amounts of "metadata": apart from some defect management data is pretty much direct mapped, so what's there to cache?

Well of course these days you can't even exclude the possibility that the HDD is actually doing some content analysis on your personal data, puts that into the NAND and WD sells it to the highest bidder via some back-channel... We are way too far along in this data economy craze to exclude some product managers wanting to sell that: feasability might still be another matter...

And somewhat between these two scenarios lies another, where WD is actually adding a mapping or indirection layer somewhat like what's used on SSDs for wear levelling.

Except that for lack of wear in this case it could actually be used to serialize a number of concurrent write activities and use the mapping table to manage that, pretty much like a write log on a relational database.

Now if that worked, it should show somewhere in less physical seeks for logical seeks; perhaps not on synthetic benchmarks but hopefully on application traces.

Could be a good thing to do on a video drive...

But eventually you'd want to do some housekeeping or even 64GB might not be enough to manage enough redicrection.

But since those trace benchmarks don't show any indications of "lost seeks" that doesn't seem to be the case.

And then there is the chance that they are in fact hiding an SMR drive behind a NAND cache...

Unless someone at WD comes up with a good explanation as to what's going on and how that's good, I'd stay away.

 

[JarredWaltonGPU]

Honestly, I see a scandal cooking, like the SMR/CMR fiasco!

The least problematic effect of that flash would simply be what we see in the tests: it doesn't do any harm.

It has some extra vaguely "premium" sounding extra attribute called OptiNand and that's it. Just lures some naive buyers into choosing this over an otherwise indistinguishable competitor.

If the flash was good enough to serve as an in-line write-back cache, the benefits of that are usually too limited at the capacity vendors can support, but NAND is getting cheaper...

But on a CMR hard drive, there simply is no need nor benefit to handle larges amounts of "metadata": apart from some defect management data is pretty much direct mapped, so what's there to cache?

Well of course these days you can't even exclude the possibility that the HDD is actually doing some content analysis on your personal data, puts that into the NAND and WD sells it to the highest bidder via some back-channel... We are way too far along in this data economy craze to exclude some product managers wanting to sell that: feasability might still be another matter...

And somewhat between these two scenarios lies another, where WD is actually adding a mapping or indirection layer somewhat like what's used on SSDs for wear levelling.

Except that for lack of wear in this case it could actually be used to serialize a number of concurrent write activities and use the mapping table to manage that, pretty much like a write log on a relational database.

Now if that worked, it should show somewhere in less physical seeks for logical seeks; perhaps not on synthetic benchmarks but hopefully on application traces.

Could be a good thing to do on a video drive...

But eventually you'd want to do some housekeeping or even 64GB might not be enough to manage enough redicrection.

But since those trace benchmarks don't show any indications of "lost seeks" that doesn't seem to be the case.

And then there is the chance that they are in fact hiding an SMR drive behind a NAND cache...

Unless someone at WD comes up with a good explanation as to what's going on and how that's good, I'd stay away.

That's... some deep level conspiracy stuff. LOL. I'm sure WD isn't going to try and log personal data to NAND and sell that to the highest bidder. That's just way too far out there, jumping the shark while doing a triple back flip!

Most likely scenario in my mind is that WD was trying something new with NAND and hoped to see bigger benefits. Maybe they were going to try a cache and it ended up not working well, or having other problems, so rather than ditching it completely they pivoted to caching meta data.

I wouldn't expect it to be an attempt to hide SMR, but I suppose anything is theoretically possible. Probably just an attempt to make an HDD more attractive to buyers that ultimately fell flat.

 

[abufrejoval]

That's... some deep level conspiracy stuff. LOL. I'm sure WD isn't going to try and log personal data to NAND and sell that to the highest bidder. That's just way too far out there, jumping the shark while doing a triple back flip!

If drive manufacturers had not started to phantasize about doing data analytics stuff as extended drive functionality at one point, I wouldn't have gone there myself: making all kinds of memory and storage "smarter" on the cheap is a dream more persistant than some storage.

When Martin Fink went from doing memristors to eliminate all other storage classes at HP to Western Digital as CTO, he evidently took some of his outlandish ideas with him.

The next thing we heard was that he was doing RISC-V based controllers with all kinds of extra smarts, perhaps native Ethernet interfaces and analytics.

He seems to have retired since and as you quote a Broadcom chip as the controller used, those initiatives might have gone the way of the memristor but the question remains: what on earth are they doing with that flash?

Most likely scenario in my mind is that WD was trying something new with NAND and hoped to see bigger benefits. Maybe they were going to try a cache and it ended up not working well, or having other problems, so rather than ditching it completely they pivoted to caching meta data.

But hard disks traditionally just don't have a lot of metadata to manage. Some SMART data and defect management won't use dozens of gigabytes (especially since I remember HDDs with 5 or 10MB total capacity from my earliest PDP-11/34 days). So where would the metadata explosion come from and why is it not a universal phenomenon?

64GB of "industrial grade" flash may be rather cheap today but with nothing to show for it in features or performance it still isn't something that WD would put into millions of units...unless we are at a point where smaller flash chips cease to be cheaper.

The more I think about it, the more I'd want a really good explanation, especially since they call it "opti" meaning best: the best use of flash clearly isn't doing nothing with it.

I wouldn't expect it to be an attempt to hide SMR, but I suppose anything is theoretically possible. Probably just an attempt to make an HDD more attractive to buyers that ultimately fell flat.

Well honesty seems wanting and we've seen how they pushed out SMR without telling.

If you wanted to do SMR 'better', larger DRAM and flash would help to hide the write amplification to the point (beyond 64GB) where few would notice.

So I believe the question needs to be raised with WD and for the moment I can't imagine any explanation that would satisfy me, including "useless dark flash".

 

[wbfox]

1. Stop lauding them for 512MB of cache. Its insulting to the surviving optanes of the world.
2. Good. Proprietary storage details. Exactly how will this monkey wrench my ZFS resilver this time WD?
3. WD Device Analytics is the next up and coming public apology. Making normies have the thought their drives need to be replaced when they don't...oh here come class action attorneys now. Fast buggers aren't they?