WD Red & avoiding the SMR pitfall

[ttabbal]

They tried to hide it because they know that anyone working with arrays would avoid them like the plague. And rightly so. They are so bad that many array systems will mark them as failed. When the drives are marketed specifically for NAS usage, that's downright false advertising as far as I'm concerned. Many of us don't need SSD speed, and won't pay SSD prices. We want online storage with reasonable performance that keeps the costs down a bit. Selling SMR as archive drives or for security cameras etc. makes sense. As a general NAS drive, not at all. Even used SSDs are significantly more expensive than new HDDs in this size range. New enterprise grade 4TB HDDs are listed from a number of sellers on ebay for under $80/ea. There aren't many SSDs listed in that size range, but it looks like the cheap ones are almost $500/ea.

Don't get me wrong, SSDs are awesome. I won't use a HDD for booting or running workstations. But as mass storage for a NAS, I'm not willing to pay a 5x premium for it. I'm a home user with at most 4 clients hitting it simultaneously. For an enterprise server with lots of clients, SSD is the only way to go, no question. Or at the least a tiered storage setup.

For me, SMR serves the same niche as tape. Slow, offline storage for backups etc.. And the price and marketing better reflect that or I won't buy it. I can't believe someone there thought false advertising was a good idea in this market. I know people that STILL won't touch a Seagate drive, even an SSD, due to the problems they had in the 90s.

 

[alfredborges]

I've investigated this a bit further and came up with the following conclusions. Western Digital is not transparent with any of this information - so this is based just on what I found

1. The WDx0EFRX drives appear to be the older model - I purchased WD Reds in 2013 and they match the 2013 datasheet. Ditto for some 3TB Reds I bought in 2016. As recently as the 2018 datasheet, WD listed WD40EFRX drives in their NAS datasheet. However, this was the first appearance of the WDx0EFAX drives in 10 and 12 gig sizes

2. Their latest datasheet, published in December 2019, lists both WDx0EFRX and WDxEFAX models for Reds. Interesting differences in Cache and speed listed between the two without explanation.

3. Amazon and others still have WDx0EFRX and WDx0EFAX drives listed separately, I purchased a "spare" WD Red over the weekend - it arrived today and is a WDe0EFRX model.

4. Qnap has a hardware compatibility list - My NAS, QNAP TS451, does not list WDx0EFAX as a compatible drive. It does have WDx0ERX spelled out.

5. On the Synology compatiblity list - the WD60EFAX and the WD20EFAX are listed as SMR Drives


The following is not verified - but was mentioned in the QNAP and Synology Forums. The WDx0EFAX drives may have been modified thru cache to give SMR drives better compatibility with RAID.

 

[msg7086]

Torrent on SSD is the easy answer for not much more $$ than those small drives.

Cool. But what's the benefit of using a SSD when your home internet is still merely 100Mbps or maybe 500Mbps if you pay a premium? How does that justify the not much more $$ at all?

I mean, c'mon, even a $10/pc 5 year old used 500GB drive could easily crush those brand new REDs on torrenting.

 

[Evan]

And that’s always and option right , buy used.
8tb HGST is good value for money as a used enterprise drive and great performance

 

[i386]

But for SMR drives, if you try to torrent on those drives, you get merely 10MB/s after a few minutes, and your system becomes lagging due to super high latency I/O on those drives. I'm perfectly fine with even an IDE drives that runs at 60MB/s, but 10, no. The record I saw from seagate 8TB SMR was 750KB/s writing speed at torrenting, max out.

I'm pretty sure that any hdd will have shitty performance for torrents as torrents are >99% random writes/reads.
Eg: the current archlinux iso file is split in 1298 pieces of 512KByte large chunks and your client writes the piece in the order it gets the pieces from other peers.

 

[brewmonkey]

Yeah, SMR drives have a very niche use. I ran them no problems in the past, in the form of the Seagate Archive drives, but my use case was ideal for them - running in an unRAID array that slowly filled with videos and the likes, and only read once in a while. I did try them in a Synology, and even in RAID1 they weren't terribly compatible.

Again, I have to recommend Toshiba drives, I've had zero issues with them, and they're basically a HGST drive with a different label. Heck, the 4TB Toshiba N300 I got a few weeks ago is physically identical, and even has similar firmware IDs compared to a HGST Deskstar NAS drive I've had for nearly 6 years.

Thanks for N300 recommendation. I sure do wish they'd sell their 5400 drives to retail. :/

 

[brewmonkey]

Just doing random writes is not the problem. The "DM" (device managed) part of DM-SMR deals with this as long as your rate of writes does not go faster than their caching/management of actual writes to the drive can handle.

The problem comes up during recovery/resilver - especially when recovering data to a replacement drive in a Raid-5/6 (or RaidZ 5/6) pool. In that case you get a massive number of writes to the replacement drive that are often not done as one big sequential stream. That is where this was discovered - drives dropping out of pools during recovery/resilver activities. That is also when the problem would be most insidious because you've believed you had data protection but when you need it most it fails you in ways that risk damaging the entire pool.

For all the walls of text I've seen on this subject, this paragraph is about the best, most concise explanation I've seen. Thanks!

 

[brewmonkey]

They tried to hide it because they know that anyone working with arrays would avoid them like the plague.

This is perhaps one of the things that most baffles me. Did they really think they would get away with this deception? Revealing the crippling performance (or outright failure) is easy to reproduce with very basic, low-cost hardware and a free OS. It's not like it's some super obscure issue that could only be revealed by a huge research lab or something.

It's really a head-scratcher. There's just no good answer to any of this.

 

[urbanracer34]

They just put out a press release:

On WD Red NAS Drives - Western Digital Corporate Blog

 

[cesmith9999]

https://arstechnica.com/information-technology/2020/04/seagate-says-network-attached-storage-and-smr-dont-mix

I agree that the PMR/CMR/SMR/HMR all need to be spelled out for us power users...

Chris

 

[cesmith9999]

On WD Red NAS Drives - Western Digital Corporate Blog

This almost makes me want to no longer buy from WD...

Chris

 

[nasi]

Thanks for N300 recommendation. I sure do wish they'd sell their 5400 drives to retail. :/

Oh nice, there do exist other 5400rpm drives? What are they called? Any chance to get them used or something like that? I'm really interested!

 

[marelooke]

They just put out a press release:

On WD Red NAS Drives - Western Digital Corporate Blog

https://arstechnica.com/information-technology/2020/04/seagate-says-network-attached-storage-and-smr-dont-mix

I agree that the PMR/CMR/SMR/HMR all need to be spelled out for us power users...

Chris

Best PR Seagate (and Toshiba) could have possibly hoped for. They should send Western Digital a "Thank you" card.

 

[edge]

That wd blog is the definition of tone deaf.

 

[Klee]

Well that explains the higher than expected prices for the older used 3TB reds on ebay. I repurposed a 3TB red out of a WD MyCloud after the motherboard in the mycloud started to fail. So I though maybe buy a couple of used ones and replace the three old 2TB HGST drives in my file server.

Going for like 58-70 dollars each and some even have higher bids than that.

 

[alfredborges]

right now (4/22/20) on Amazon you can get new WDx0EXRF drives for little or no premium over stock WDx0EFAX drives

In the WD blog post, they basically tell people that are concerned about this to buy WE Red Pro at a 100% premium or WD Gold at about a 30% premium

 

[Klee]

right now (4/22/20) on Amazon you can get new WDx0EXRF drives for little or no premium over stock WDx0EFAX drives

In the WD blog post, they basically tell people that are concerned about this to buy WE Red Pro at a 100% premium or WD Gold at about a 30% premium

Or buy another brand hard drive.

 

[alfredborges]

will consider another brand drive. My current array has 4 yr old Red drives, I wanted a spare on hand that I have confidence in.

Concerned about problems in rebuilding the array if I have a failure

 

[ttabbal]

Apparently, WD doesn't want to be able to sell the Red brand anymore. They might as well shut it down, it's dead now other than known-good model numbers. Nobody is going to buy the premium versions either, when they can get a different brand for a better price and performance. The only thing I can figure is some suit up top that doesn't understand this stuff anyway saw he could increase margins a few percent. I have some reds in my server. I won't be buying more of them as I can't be 100% sure what I'll get.

The other problem they are creating for themselves is that they have demonstrated they are willing to do this sort of thing without telling anyone. So we can't trust them not to do it to other lines. The only possible exception is Enterprise storage. Companies buying millions of dollars in storage have the clout to demand better treatment.

Seagate is half right, SMR has no place in arrays. I also say it has no place in desktops either. Who is building desktops without SSDs? Most people wouldn't even use SATA SSDs these days, opting for nvme instead. Well, perhaps as a second drive for backups and media. Might as well make them USB really, as they can't saturate even that bus, but sure. At least then you can easily backup multiple machines. Since not everyone wants to deal with a NAS.

Not disclosing it is no different than selling a hybrid drive as an SSD saying that its profile is the same for desktop use. No, it isn't. It might well be good enough for some things, even most things, but it's still false advertising.

 

[msg7086]

I'm pretty sure that any hdd will have shitty performance for torrents as torrents are >99% random writes/reads.
Eg: the current archlinux iso file is split in 1298 pieces of 512KByte large chunks and your client writes the piece in the order it gets the pieces from other peers.

My point is, one is 50x shittier than the other one. I don't know if you every tried torrenting some large files on a SMR drive, but I encourage you to not try it. I have a server that's equipped with 10Gbps network and all SMR drives. Never enjoyed the 10Gbps on torrenting, not 1Gbps, I have to limit the speed to 100mbps to prevent system locks.

The lowest speed I measured from these drives are 700KB/s at full load (meaning its 100% busy writing and 0% reading).