Western Digital Gets Sued For Sneaking SMR Disks Into Its NAS Channel (arstechnica.com) 79
An anonymous reader quotes a report from Ars Technica: All three of the surviving conventional hard drive vendors -- Toshiba, Western Digital, and Seagate -- have gotten caught sneaking disks featuring Shingled Magnetic Recording technology into unexpected places recently. But Western Digital has been the most brazen of the three, and it's been singled out for a class action lawsuit in response. Although all three major manufacturers quietly added SMR disks to their desktop hard drive line-up, Western Digital is the only one so far to slip them into its NAS (Network Attached Storage) stack. NAS drives are expected to perform well in RAID and other multiple disk arrays, whether ZFS pools or consumer devices like Synology or Netgear NAS appliances.
Hattis Law has initiated a class action lawsuit against Western Digital, accordingly. The lawsuit alleges both that the SMR technology in the newer Western Digital Red drives is inappropriate for the marketed purpose of the drives and that Western Digital deliberately "deceived and harm[ed] consumers" in the course of doing so. Hattis' position is strengthened by a series of tests that website ServeTheHome released yesterday. The results demonstrate that although Western Digital's new 4TB Red "NAS" disk performed adequately as a desktop drive, it was unfit for purpose in a ZFS storage array (zpool).
Hattis Law has initiated a class action lawsuit against Western Digital, accordingly. The lawsuit alleges both that the SMR technology in the newer Western Digital Red drives is inappropriate for the marketed purpose of the drives and that Western Digital deliberately "deceived and harm[ed] consumers" in the course of doing so. Hattis' position is strengthened by a series of tests that website ServeTheHome released yesterday. The results demonstrate that although Western Digital's new 4TB Red "NAS" disk performed adequately as a desktop drive, it was unfit for purpose in a ZFS storage array (zpool).
proof of corruption (Score:5, Interesting)
the fact that a business is willing to even try this is proof of how secure they feel with the laws of the land.
the board and executives should be criminally charged of grand larceny in the amount of sales associated with these products.
that would stop stupid shit like this cold!
Re: (Score:3)
Re: (Score:1)
Re: proof of corruption (Score:2)
Re: (Score:2)
Still, not requiring an engineering sign-off on marketing promises is a sign that the truth doesn't matter to them. In-turn, that shows that they don't think the law will call them on it.
Re: (Score:2)
Re: (Score:2)
I would be interested what VW has to say about that.
Re: (Score:2)
the fact that a business is willing to even try this is proof of how secure they feel with the laws of the land.
Actually it's more a proof of the fact that they didn't test everything or think things through and released a shitty product as a result. HDDs have *always* changed over their model life. Typically with every change in RRP the models have changed significantly internally. WD or anyone else never explicitly promised what technology would be given, what platter density, what number of heads or any of those details.
The difference is quite often the change has resulted in improved performance, usually as aeria
Re: (Score:2)
Not an edge case. Very much standard operation being in conflict with how RAID controllers manage drive health. Even a cursory look would conclude that this is utterly unworkable. Unless you go out of your way to decide that a giant NAS is only needed for tiny files. Why would anyone be such high capacity if that were true? And yet the drives were specifically marketed for NAS use.
Re: (Score:2)
Not an edge case. Very much standard operation being in conflict with how RAID controllers manage drive health.
Not at all. The only RAID systems which have problems are poorly configured controllers that mistake a drive that responds as busy as dead. The edge case is performance during resilvering a mirror. Otherwise the drives (as evidenced in this test) work just fine for the vast majority of workloads.
Unless you go out of your way to decide that a giant NAS is only needed for tiny files. Why would anyone be such high capacity if that were true?
Is that a serious question? The size of a NAS has nothing to do with its workloads. Some NASes will see large continuously blocks of data, e.g. media files, some will see lots of very small writes, e.g. databases, s
Re: (Score:3)
Please don't. I actually like drives which get cheaper over time.
I like knowing what I buy. The drive could be SMR or actually a SSD with 10TBW endurance inside, it does not matter - as long as I know about it before I buy it by reading the datasheet.
With these particular drives, they were not even cheaper, at least in my local shop.
Right now, I have a choice of 4TB drives:
WD40PURZ - CMR, TLER, 5400RPM, "designed for surveillance" - 113EUR
WD40EFAX - SMR, TLER, 5400RPM, "designed for NAS" - 121EUR
ST4000VN008 - CMR, TLER, 5900RPM, "designed for NAS" - 132EUR
So, the WD "NAS
Re: (Score:2)
I like knowing what I buy. The drive could be SMR or actually a SSD with 10TBW endurance inside, it does not matter - as long as I know about it before I buy it by reading the datasheet.
I agree, but don't go applying new requirements to an industry that has been not telling people anything other than performance figures for the best part of 30 years and then declaring it a sue-able offense.
Re: (Score:2)
Please don't. I actually like drives which get cheaper over time. I also would appreciate it if you didn't suddenly get outraged over a standard business practice of the past 20 years.
So you don't care if it's slightly slower than chiseling on a stone slab as long as it's cheaper? Do you take a similar view on reliability?
Of course nobody minds if the speed and reliability are better than promised. This was no edge case, they have known for a long time that SMR offers slow writes. There are applications whe
Re: (Score:2)
So you don't care if it's slightly slower than chiseling on a stone slab as long as it's cheaper? Do you take a similar view on reliability?
Maybe you should re-read my post. Specifically about the fact that changing drives without telling people has not only resulted in cheaper drives, but actually resulted in *faster* drives. And as evidenced in this test, for many workloads even the SMR drive can be faster.
So no, don't stop "stupid shit" like changing drive internals that has seen us get every *better* performance and cheaper prices. But by all means hold drive makers to account for selling a drive that no longer meets the spec.
Re: (Score:2)
So no, don't stop "stupid shit" like changing drive internals that has seen us get every *better* performance and cheaper prices. But by all means hold drive makers to account for selling a drive that no longer meets the spec.
YEAH! You tell ... um ....them .... and .... um ... uh ...... uhhhhh .....
Okay, I think I know what happened here, either my Adderall wore off already, or your Adderall wore off already.
Re: (Score:2)
Please enjoy your new slower than chiseling the data in stone drives. Again, of course nobody minds if the drive is at least as fast as the older tech, just as reliable, and a bit cheaper (that certainly won't create any problems), but that isn't what happened this time. If it's cheaper but slower AND they truthfully reveal that, it's a trade-off that I can make rational decisions about for my application. If they don't disclose the slower part, it's a rip-off.
Re: (Score:2)
Please don't. I actually like drives which get cheaper over time.
So what if they get slower too, they’re cheaper! Cheaper! I don’t care if they sell a pencil and paper inside a disk drive box and make you hand write all your data, it’s three dollars less so BOOOYAH BITCHES!
I also would appreciate it if you didn't suddenly get outraged over a standard business practice of the past 20 years.
Those rat bastard sons of a motherless whore!
You tell them! It’s STANDARD people! Sheeesh!
Re: (Score:1)
False advertising at most... Not entirely unlike Twitter pretending to the advertisers to foster "safe friendly environment" [twitter.com], while allowing their platform to be used for — nay, dominated by — encouragement of vicious violence [twitter.com] and doxxing [twitter.com].
Re: (Score:2)
Speaking as a lifelong citizen, the United States has become quite the disgrace. It’s such a shame, the potential for continued greatness was there. In the end, we succumbed to divide and conquer. I think it was easier than anyone anticipated.
Anyway, I’ll go ahead and be our spokesperson for this article:
“Consumer protection? Why that’s communist! And socialist! And scoobydoowhereareyouist! I don’t want to live in a country where the shrewd can’t prey on the unsophisti
Re: (Score:2)
Get real, people have been doing crap like this since biblical times. As memory serves some of the very earliest texts from Mesopotamia (Code of Hammurabi) dealt with fraud by commercial interactions such as construction of housing.
https://www.history.com/topics... [history.com]
What the OP lacks... (Score:4, Informative)
Don't you know what "Shingled Magnetic Recording" is?
https://en.wikipedia.org/wiki/... [wikipedia.org]
Also, raid drops drives that timeout (Score:4, Interesting)
Also, in the context of this story it's helpful to know that SMR drives aren't just slower in a raid context, they fail. A raid controller treats a unresponsive drive as a dead drive and drops it from the array. Unresponsive is defined as "didn't respond in time", so a too-slow drive is dead as far as the raid controller sees it. Two SMR drives in an array with standard drives can lead to total data loss.
Speaking of total data loss from failed arrays, often an expert can resurrect a "totally failed and lost" array through data surgery. I've had to learn how to do that after saying "oops" a couple times with 16-drive arrays.
Re: (Score:3)
Re: (Score:3)
Btw for the same reason, in a raid you want a drive (really a drive firmware) that doesn't try really hard to recover from errors. Consumer desktop drives can spend a long time trying to re-read a bad sector. In a raid, that means the drive is unresponsive and the whole drive is considered dead. Better to just return an error and let the raid handle one bad sector.
Re: (Score:2)
Shingled Magnetic Recording is, in short, just like stacking sectors like shingles rather than like a grid (side by side). It allows for salesmen to say things that should be false.
To my way of thinking, this whole "technology" is false. Each sector overlaps the one next to it, so if you write data to that sector, and the sector next to it has valid data, that valid data has to be written to another sector. Basically, your roof measures some dimensions, let's say 100' x 45'. Now, if you took your shingl
Re: (Score:2)
You get the full capacity. There's just a high cost to writes that cause a cascade of additional writes. It's a terrible design for most use cases, but the one thing you do get is the full capacity.
Moore proof (Score:3, Informative)
Re: Moore proof (Score:2)
Re: (Score:3)
Re: (Score:1)
Of course it could be. Furthermore, you can't ignore that the process ran to successful completion, something that for many years the industry has insisted is impossible, a bogus claim often repeated here by easily duped "/. experts". The idea that a drive can't write its entire surface without a hard failure is absurd (since they would never be able to exit manufacturing). This has been the basis for upselling RAID levels for a long time now and it is a lie.
SMR devices do NOT perform poorly but they can
Re: (Score:2)
Furthermore, you can't ignore that the process ran to successful completion, something that for many years the industry has insisted is impossible, a bogus claim often repeated here by easily duped "/. experts".
How do you know? ZFS doesn't re-verify write checksums until the next read. Also the system gave no indication if the drive remapped data transparently at any point.
This has been the basis for upselling RAID levels for a long time now and it is a lie.
Erm no it hasn't. In fact this argument has never been used in relation to RAID levels. It has been used in relation to checksumming on modern CoW filesystems, but never for RAID. RAID in any form wouldn't even prevent the problem if a write error did occur.
I agree a lot of people here don't qualify for the title "expert" but at least they are p
Re: (Score:2)
"How do you know?"
LOL. You think drives aren't capable of determining their own failures? Do you know what RAID 2 is?
I know because I know how this stuff works. All RAID relies on drives detecting their own failures. Drive have built-in ECC, they don't rely on "ZFS". You think that the file system must verify the integrity of data. You are wrong.
"Erm no it hasn't. In fact this argument has never been used in relation to RAID levels. It has been used in relation to checksumming on modern CoW filesystem
Re: (Score:2)
LOL. You think drives aren't capable of determining their own failures?
Of course they are. Your assumption is that the data corruption is being handled upstream rather than a silent kernel read error.
I know because I know how this stuff works. All RAID relies on drives detecting their own failures. Drive have built-in ECC, they don't rely on "ZFS". You think that the file system must verify the integrity of data. You are wrong.
Listen kid, the reality is that checksumming actually catches errors that are silently passed upstream and that not the entire data path between the disk and the userland has error checking. Honestly I'm not going to bother reading further. "You are wrong" Implies that no one should ever have seen a checksum error caught by zfs or brtfs, since they have I'm going to file your "I k
Re: Moore proof (Score:5, Interesting)
I don't think it can be fixed. SMR drives write data in CMR format and then later when idle re-write it in SMR format. With a RAID rebuild there is no idle time.
Originally this caused the drives to lock up for minutes at a time while re-writing data. These newer ones "fixed" that by continuing to respond but mixing host writes with SMR writes. The result is that instead of a smooth sequential write you get a massive amount of head thrashing.
I'm surprised the drive survived 14 days of constant thrashing.
Re: (Score:2)
Re: (Score:2)
I guess the problem is doing it in a way that works reliably. With these SMR drives there is no way to ask them about things like buffer sizes or tell them to flush or even ask when they have finished flushing.
There is another type of SMR drive for enterprise where the SMR is handled on the host, mitigating these issues. There are few consumer grade HBAs that support it though and things like ZFS don't anyway.
The solution is for them to clearly mark SMR drives so we can avoid buying them. They don't want to
Re: (Score:2)
If your point is that successfully integrating consumer SMR that you know nothing about is not possible, almost certainly true. To integrate SMR meaningfully into RAID, the drives must be fully understood by the controller as a system. What use is a RAID system that has no write throughput even if it doesn't fail?
Re: (Score:2)
Yes, exactly. Anyone really curious could do a patent search on SMR. All this stuff has been thought through plenty.
SMR can write fast if it is written properly. The key is for the controller to understand how to write properly. The claim that SMR writes data in CMR is, as a broad generalization, false. Some drives may do that always, but it is not a requirement that SMR always write that way.
Re: (Score:2)
It can be fixed. First, background reconstructions CAN be structured to allow idle time. Second, SMR drives do not always "have to" incur a write penalty provided they are always written such that they don't require rewrites of data. It may be the case that RAID implementors must work with drive vendors to implement SMR effectively but it is not beyond the imagination. People may have done it already, I wouldn't know.
A traditional rebuild will not understand SMR zones and will likely cause some degrada
Re: (Score:2)
Any pause means long rebuild times. Can't avoid the re-writes without some new ATA commands to let the host communicate what it is doing.
Re: (Score:2)
I'm surprised the drive survived 14 days of constant thrashing.
You've never seen a database before have you.
Re: (Score:2)
I wonder if at least some of it could be "fixed" in new RAID controllers going forward to use slightly different algorithms for these kinds of drives
There are no RAID controllers here. RAIDZ is a software function of the ZFS filesystem. ZFS treats drives as just dumb I/O and relies on the kernel to support typical performance functionality of the drives such as TLER.
Re: (Score:2)
Software "controller," then. No practical difference. Either way, I don't think AHCI even exposes any commands that will help manage the write efficiency. The drive basically lies and then tried to fix the dates later. It doesn't give true raw access.
Re: (Score:2)
Actually massive raw and fundamental differences. There is no "controller" in this setup. For zfs the drive may as well be a single desktop drive. Zfs makes no assumptions on the underlying disk, it doesn't even make an assumption that what is underlying even is a disk, it handles everything on a per file basis. That is especially significant for the resilver example where zfs performed disastrously due to resilvering on a per file basis. On a traditional RAID system you would see far better resilvering per
Re: (Score:2)
So this is your level of "RAID expertise". Explains a lot. Where is "here"? Your parent's basement?
BTW, even "dumb I/O" implements its own ECC and correct its own errors. What's attached to a RAID controller, you know that's "not here"? Dumb I/O.
Re: (Score:2)
If you don't even understand that zfs works on a file level whereas tradition RAID works on a block level there's really no helping you. But then you were a lost cause based on that other brain dead comment of yours I already replied to.
BTW, even "dumb I/O" implements its own ECC and correct its own errors.
I know man, I taught you that when I replied to your last post about being able to write to a drive without errors. Not that it's at all relevant to anything being discussed in this thread.
What's attached to a RAID controller, you know that's "not here"? Dumb I/O.
At no point was there every a RAID controller. Whatever argument you think you were ha
Re: (Score:2)
I wonder if at least some of it could be "fixed" in new RAID controllers going forward to use slightly different algorithms for these kinds of drives, but I don't know enough about the underlying technologies to know if that's a ridiculous statement, something possible but not worth doing, or something practical that can be worked on.
"Normal" RAID (md-raid or hardware) would probably work better with SMR drives because the rebuild operation is more linear compared to zfs. Assuming a brand new drive would be able to do linear writes directly to the shingles it would work fine (as long as the array did not get a lot of writes during rebuild).
Re: (Score:2)
Re: (Score:2)
Describe "normal" RAID.
As I said, md-raid or hardware RAID.
The drive responds to linear writes well. It also has a CMR write cache, which means that as long as your write load on the array is not too high and the drive has time to rewrite the shingles it would work fine.
Of course, if you are constantly issuing random writes to the array SMR drives would be unsuitable.
Re: (Score:2)
Of course, if you are constantly issuing random writes to the array SMR drives would be unsuitable.
You mean like in a NAS when you have multiple users, a purpose that WD advertises is the purpose of their Red models as indicated by the name: WD Red NAS [westerndigital.com] If WD put these on their Blue, Green, or Purple line, it's less of an issue. But they put it in a line where all it's guaranteed to cause issues: multiple writes simultaneously in a RAID array
Re: (Score:2)
There's nothing about a NAS that says continuously sustained workload. NAS is not some magical limiting use case. You can have NAS systems that do large continuous rights, small random writes, you can sustain either of the two cases, or they can be intermittent.
Re: (Score:2)
There's nothing about a NAS that says continuously sustained workload.
When did I say that? The problem again with SMR is as the drive fills up, it has to spend time to redo shingles. Not a problem if it is a single drive for home use. So what it takes a little longer in this workload. If it's in a NAS with a RAID, the drive doesn't respond to the controller within a reasonable amount of time and appears to fail. With a NAS there is a high chance you will have multiple writes happening simultaneously.
NAS is not some magical limiting use case.
We are talking about WD Red drives. WD Red "NAS" drives as Western Digital la
Re: Moore proof (Score:4, Funny)
It took 19 days to write 4TB?
To be fair, the data went through an intermediary paper tape punch and reader, so about what you'd expect.
Re: (Score:2)
More
Re: (Score:2)
SMR drives are divided into "shingles" or "bands" that are, say, 256MB in size. If you want to modify the data in the middle of the band, you have to rewrite the part of the band after the place you modified (though maybe drives just rewrite the whole band).
This is obviously very slow, so the drive has a cache which operates like a normal hard drive and is used to write the data. Then, when the drive is idle, it rewrites the shingles with the new data. This works acceptably as long as the writes come in bur
Re: (Score:2)
This is all true, but it does not preclude SMR-aware software from mitigating these characteristics. Your desciption is well done, so the following is not a criticism.
Consider SSD. They present as block devices even though they are a different technology, and they "emulate" a block device using what you could call an underlying "filesystem". Any RAID controller could do the same with their underlying storage and plenty have.
There would be every reason for RAID architects to consider what to do with SMR a
Re: (Score:2)
SSDs are much faster than hard drives and do not have a problem resilvering a zfs pool. Also, leaving some unused space at the end of a SSD (overprovision) makes it lose less performance when being written to.
Reason for the disconnect between WD and buyers... (Score:4, Interesting)
.. is that WD sees two NAS markets
1) Real Enterprise users who would never (except for niches like cold storage) want SMR NAS drives -or NAS drives as small as 4TB (generally at least 8TB),who expect to pay a premium for Enterprise-grade NAS drives.
2) Home and SMB users who are using plug and play appliances that either have drives provided by NAS device vendor, or drives purchased as a quick add-in to what is essentially a consumer product. For these users, who have no idea what SMR is, TB/$ is biggest metric and WD thinks if they don't provide the cheapest option there someone else will.
The PROBLEM is there is a
3) DYI market for tech hobbyists / SMB / integrators who want to build, using free NAS software particularly FreeNAS (now "TrueNAS Core") that lets you take commodity NAS drives (which WD has traditionally marketed for 2) ) to build superb almost enterprise grade server equivalents.
The disconnect is that WD thinks those kind of boxes should be using much more expensive enterprise grade NAS drives but the DYI crowd wants to minimize their hardware spend (or reallocate to other key items like more ECC RAM) by using commodity hardware whenever possible and drives are by far biggest cost part of a NAS system.
Drives like the WD Red have been a godsend to the DYI crowd but the product positioning teams in marketing at all the major drive manufacturers think if you're making enterprise grade servers, you're cheating by using consumer drives. But DYI segment is never going to pay enterprise drive prices that big corporate OEMs and customers are willing to pay, so a showdown is coming. (SSDs may wind up making this whole fight in DYI redundant once 10 to 20 TB of SSDs is not astronomically priced.)
Re: (Score:3)
Despite anti-consumer changes like this, the WD fanbois remain faithfu
Re: (Score:2)
WD created the Red drives by removing TLER from their regular drives but keeping it on the Red drives. That is, the Red drives aren't "better," they just retain a feature useful for RAID which used to be present on the regular consumer-grade WD drives, but which WD intentionally crippled to force you to buy a more expensive Red drive. Seagate and Toshiba didn't do this - you can still use their regular drives in RAID without issues.
Well, actually, there is more to a consumer/prosumer/SmallMediumBussiness NAS/SAN drive than TLER. NAS/SAN drives have dual attachement of the spindle (consumer drives have just one), improved vibration tolerance, vibration sensors, and a different cache management strategy.
I built many incarnations of my NAS with consumer drives, from many brands, but in this day and age, it makes no sense to do so. Shuking drives is even worse. Between non-TLER drives, SSHDDs, 512e drives, SMR drives and no vibration cont
Re: (Score:3)
.. is that WD sees two NAS markets
Both Seagate and WD see the NAS/SAN market as three segments, while Toshiba sees it as two:
Consumer (highly price sensitive, the company assumes the customer does not know the technology) NAS/SAN:
WD: RED 5400 RPM, some SMR
Seagate: Ironwolf 5400 RPM, no SRM
Toshiba: N300 7200 RPM, no SRM
ProSumer/SmallMediumBusiness (less price sensitive, but the company assumes not necesarily knows the technology i.e Photo/Video graphers, non tech related Small/medium bussinesses like
Re: (Score:2)
How many people in group #2 need 4TB or more of storage that won't be making large sequential writes? No way you're filling that up with Word documents.
I want in (Score:1)
Re: (Score:2)
People made this decision not company (Score:5, Insightful)
Re: (Score:2)
Re: (Score:2)
> If people were charged and had to pay the price this would stop or at least be slowed down.
Not if they're allowed to pass the buck to some scapegoat in a technical department.
Everybody forgets that permanent corporations were illegal in the US before the Civil War because they were known to be too dangerous to have around.
Partnerships are much better, because the owners remain liable. Wall St. investment banks were required to be partnerships until a decade before the Great Financial Crash.
Re: (Score:2)
This x1000 but then they could say that it's part of the company culture and are under constant pressure to reduce cost by any means.
Pssst, racist tweet on Twitter. Not sure how you can think about this stuff at a time like this. Come on, let’s use it to get some attention for ourselves. I’m thinking of taking the “racism is bad” position.
Re: (Score:2)
I'm sick of hearing "Company A" committed fraud. Companies don't commit fraud, people do and they need to be charged, prosecuted, and if found guilty fined or jailed.
Amen, I’m with yohmygod, some B-list celebrity just made a racist tweet! Holy shit am I outraged! I’m so outraged! Who is outraged with me? I’m going to Twitter right now to make sure everyone knows that I’m outraged! I hate racists, because I’m not one. Are you one? Better head to a Twitter now and make sure
Fraud (Score:1)
This is a criminal matter. It seems corporate fraud laws are weak in the US. :\
Re: (Score:2)
In what sense? To claim fraud you have to show they acted fraudulently. In order to do that you need to show you got something other than what was promised. So the key part of your fraud claim rests on the ability to prove that WD claimed their drives would be able to sustain their peak performance numbers indefinitely, and I'm sure you're going to fail to prove that after you read all the asterisks in the marketing material.
The reality is drive makers have been changing drive internals mid product for 30 y
Re: (Score:2)
The reality is drive makers have been changing drive internals mid product for 30 years now.
And we like it like that! I’m still using a computer with 640K of RAM too. 300 baud modem in the hizzouse! Civil Rights Act? What was wrong with Civil Rights the 200 years prior?
Leave things the way they are!!!
It's worth reading the Wikipedia article.... (Score:2)
So in addition to the si
Re: It's worth reading the Wikipedia article.... (Score:2)
Re: It's worth reading the Wikipedia article.... (Score:2)
I'm more upset at lack of a RPM spec. (Score:2)
One of the shenanigans these drive manufacturers pull off is completely providing ZERO information on the RPM of some of their drives because they don't want customers to know they are the crappy, slower 5400 RPM ones.
It's the same bullshit some bicycle manufacturers pull and not list the actual weight of the bike.
BOSE does the same shit and doesn't list THD on some of their speakers.
Hint: If a manufacturer has to resort to sleazy tactics of "omission" about common specs. in the appropriate sector then they
Re: (Score:2)
Apple doesn’t even list the amount of RAM in their iOS devices, but they list all the other specs.