The three major HDD manufacturers are selling slower drives, without telling us

[Kilrah]

18 minutes ago, mr moose said:

If it's not important then why do you need to know so badly?

It IS important to some, and those want to know. 

 

18 minutes ago, mr moose said:

There are lots of specs we don't know and will never know. 

People don't initially care about knowing whether it's SMR, they want to know the performance. Since manufacturers don't want to give read/write perfromance numbers, then at least tell us if it's SMR since then we can know performance will be massively lower if it's inappropriate for the use case.

 

The other stuff you mention will probably have some small impacts on pertormance. SMR can have HUGE impact, bringing down a drive to having write performance lower than any CMR HDD that's existed in 15 years.

 

Do you expect this read/write difference out of a 2-drive RAID0?

 

 

(Yes is not a valid answer.)

[mr moose]

3 minutes ago, Kilrah said:

It IS important to some, and those want to know. 

He said it wasn't important to him, but he needs to know.  So the only questions is why? 

 

3 minutes ago, Kilrah said:

People don't initially care about knowing whether it's SMR, they want to know the performance. Since manufacturers don't want to give read/write perfromance numbers, then at least tell us if it's SMR since then we can know performance will be massively lower.

 

The other stuff you mention will probably have some small impacts on pertormance. SMR can have HUGE impact, bringing down a drive to having write performance lower than any CMR HDD that's existed in 15 years.

Got some testing to show this difference?  I hear a lot of people make claims saying: "huge", "massive" and"half the performance from", but I have yet to see one person show me a test with the results.  What is the difference?  under what conditions does this massive impact occur?  If you can't tell me and I can't find it using a basic search then on what grounds is everyone making these claims?

 

 

 

3 minutes ago, Kilrah said:

Do you expect this out of a 2-drive RAID0?

 

 

(Yes is not a valid answer.)

I have no idea what type of array, the type of drives, the age of the system or the way it was setup. I know nothing about it and thus cannot answer.  By the way don't ask a question if you aren't going to accept the answer.   It's a pointless effort that serves no purpose.

[Kilrah]

2 minutes ago, mr moose said:

By the way don't ask a question if you aren't going to accept the answer. It's a pointless effort that serves no purpose.

That's what you've been doing since the start of this thread though? 

 

5 minutes ago, mr moose said:

I have no idea what type of array, the type of drives

It's an array that would have had close to the same read and write performance if the drives weren't SMR, but since they are there's a difference factor of 3. 

[mr moose]

14 minutes ago, Kilrah said:

That's what you've been doing since the start of this thread though? 

Which question did I ask that I am not listening to the answer?  The only question I have really asked is give me some numbers. I see you even snipped that question out of the last two posts of mine you quoted.  Cite a test or benchmark,  point me in the direction of testing,. don't just say there's a massive difference, actually back that up with something.

Quote

It's an array that would have had close to the same read and write performance if the drives weren't SMR, but since they are there's a difference factor of 3. 

That means nothing. that is literally what a politician would answer when asked a specific.   All you did was post a picture of a single crystal disk mark score and then blame SMR. 

 

So it appears no one can point me to independent testing or benchmark or anything at all that would indicate but they all want to believe their opinions.

 

I don't care how gullible the rest of the world is, I won't accept anything as being a fact until there is evidence to support it. 

 

 

[rcmaehl]

Not a fan of SMR drives if they die after 2 years

[mr moose]

7 minutes ago, rcmaehl said:

Not a fan of SMR drives if they die after 2 years

So that's a basic consumer drive that's been in operation almost constantly for 2 years, Whats the IO work load its seen?

 

EDIT: in fact that means you have basically restarted once every 3.6 days (average) and it is constantly running the rest of the time.

[rcmaehl]

8 minutes ago, mr moose said:

So that's a basic consumer drive that's been in operation almost constantly for 2 years, Whats the IO work load its seen?

 

EDIT: in fact that means you have basically restarted once every 3.6 days (average) and it is constantly running the rest of the time.

Was used for cold storage, including Windows File History. In fact, I'm pretty sure Windows File History is what killed it.

[thedude4bides]

On 4/19/2020 at 6:58 PM, mr moose said:

 

 

Because I don't see there being a problem. 

 

 

Because *you* don't see a problem means there isn't a problem?  Or does it mean you can't see all that well?  hmmmm....

[mr moose]

Just now, rcmaehl said:

Was used for cold storage, including Windows File History. In fact, I'm pretty sure Windows File History is what killed it.

yes, but it was running for almost 2 years straight,  that's not cold storage, that's Luke warm storage at best.

 

 

[mr moose]

1 minute ago, thedude4bides said:

Because *you* don't see a problem means there isn't a problem?  Or does it mean you can't see all that well?  hmmmm....

Maybe if you read all my other posts you'll work out what I mean.

[leadeater]

9 minutes ago, rcmaehl said:

In fact, I'm pretty sure Windows File History is what killed it.

Probably VSS, MS is really not clear how how this works, shuffling data around in the background constantly. Not sure how it compares to the old Backup and Restore but that was fairly ok, at least I know how that one works.

[TechyBen]

2 hours ago, mr moose said:

You can decide that, but you can't pretend there is a legitimate reason for needing that information other than you want it.

 

Link or citation please. 

 

If it's not important then why do you need to know so badly?

 

SMR is only a problem if there is evidence to show it is.  You may as well try to claim the oxide coatings they use is a problem because they haven't told you what they use. Some oxides are better than others so they should have to list it right?   Because it must be important to someone.  No, they don't have to list it and unless you have a reason to demand they list it then trying to argue that it is important to someone isn't a good enough reason.

 

 

I am not making that decision.  I am pointing out that claiming SMR is a spec that is necessary to know without being able to provide evidence to show why (again, that is evidence that there is an issue that can only be overcome/mitigated by labeling the drive as being an SMR drive) does not make it an important or necessary spec to list. 

 

Citation? Linus when he tried shingled drives. Like I think it's impossible to have a discussion with you here. There is a cache. There is a requirement for it to remap the clustered read/write/read areas. It performs differently. Thus some may wish to buy different models with different performance. Some may even wish to get SMR disks with higher cache, but if it's not specced/noted, only buying and testing internally/personally will show the differences (between brands or in a brand).

 

Why do I waste breath on this?

 

Quote

Replace SMR in any one of your arguments with any of these components and you will see how pointless the demand is without evidence:

 

-oxidde coating

-size of coil

-magnet core material

-IC manufacturer

-specific code in the bios

-static mitigation architecture

-PS ripple rejection topology in the signal amplifier.

"I'm not buying brand X because the oxide always dies". Yeah. I know some brands and some devices. I know some processes and some products. I'm done, I'm not even reading the rest and I'm out. Not in protest. I've just got better things to do then argue over trivialities and insults.

[RejZoR]

Does oxide coating, size of coil, magnet core, IC vendor, specific BIOS code or whatever affects things notably? No. Does SMR? Yes, yes it does, very. What a stupid argument...

[wanderingfool2]

2 hours ago, Kilrah said:

It IS important to some, and those want to know. 

 

People don't initially care about knowing whether it's SMR, they want to know the performance. Since manufacturers don't want to give read/write perfromance numbers, then at least tell us if it's SMR since then we can know performance will be massively lower if it's inappropriate for the use case.

 

The other stuff you mention will probably have some small impacts on pertormance. SMR can have HUGE impact, bringing down a drive to having write performance lower than any CMR HDD that's existed in 15 years.

 

Do you expect this read/write difference out of a 2-drive RAID0?

 

 

(Yes is not a valid answer.)

Well -YES- is an valid answer.  That reports you only have 1TB of space...so if that is a RAID 0, those are older drives or very cheap drives...as you would have a 500GB HDD.  So if you are intending to use that as evidence, then please mention the models and such so people can verify your claim.  SMR slowness wouldn't show up like that anyways.

 

For those wondering; at 5TB in size, it takes 90 minutes of constant file transfer before SMR kicks in, so they are perfectly okay for general consumer use (and the reason I don't think it really doesn't matter as much on the consumer level)

https://www.anandtech.com/show/14539/seagate-backup-plus-portable-5tb-backup-plus-slim-2tb-review-portable-smr/3

 

1 hour ago, rcmaehl said:

Not a fan of SMR drives if they die after 2 years

That is a 2 year old drive, that has been on for 2 years...the warning could very well be due to the drive age (2 years on, it is likely the case has been bumped or just general wear and tear for a consumer drive).  Consumer drives were not meant for that level of punishment.

 

Again, it is my opinion that for things such as NAS drives, SMR should be list as it negatively impacts the intended usecase of the drive.

It would be great if it was listed on consumer drives; but I would not hold a company responsible for not listing it (as use of that technology falls into the general purpose of the drive)....again though, this is not giving WD a pass though, as using it in the NAS drives without labeling is inexcusable.

 

In regards to the Seagate consumer drives, can someone actually point me to a proper article or anything that says they were not always SMR drives...it is hard to claim bait and switch when it might have always been using SMR.  The issue I have with this thread is that people blame SMR when it likely isn't the issue.  SMR at worse case causes delay in write (which is a terrible for NAS rebuilds, but shouldn't cause it to fail) [again SMR shouldn't be used in NAS, but as Moose said it shouldn't effect rebuilds].  This is likely a buggy implementation of the firmware with SMR or perhaps the person who initially reported the problem actually has a messed up drive.

[Kilrah]

1 hour ago, wanderingfool2 said:

Well -YES- is an valid answer.  That reports you only have 1TB of space...so if that is a RAID 0, those are older drives or very cheap drives...as you would have a 500GB HDD.  So if you are intending to use that as evidence, then please mention the models and such so people can verify your claim.  SMR slowness wouldn't show up like that anyways.

You dropped a 0. They're ST5000LM000 and from my experience must have no PMR cache. But there's no way to actually know, which is the whole point.

[wanderingfool2]

38 minutes ago, Kilrah said:

You dropped a 0. They're ST5000LM000 and from my experience must have no PMR cache. But there's no way to actually know, which is the whole point.

Yea, fair enough, I goofed on the 1TB part; I think I missed seeing the 1 in between the numbers.  Either way though, I looked at the model of drive you have...I still don't think it would be SMR that is causing that issue.  A glance at other disk marks that were ran with the model.

https://www.nikktech.com/main/articles/peripherals/external-storage/portable-hard-drives/7284-seagate-backup-plus-5tb-usb-3-0-portable-hard-drive-review?showall=1

(The model was mentioned was in the 5TB seagate enclosure, and exhibits a higher sustained rate than what is observed in yours).  Actually it can also depend on how the RAID is being run...I know I had trouble with my combination of IronWolf's in software RAID0 before (had to switch sata ports).

 

My point though is that there is so much negativity around SMR, that it doesn't make sense adding in a perceived negative to a product in the specs when it doesn't effect the use-case it was designed for.  It is why I think for NAS grade drives it is something that does need to be advertised as but consumer level it doesn't.  The articles that I have posted seems to show that the effects of SMR on performance are greatly oversold; and just looking around on the internet there is a lot of miss-information about SMR as is (with some posts claiming read speeds are adversely affected)

[Kilrah]

  

59 minutes ago, wanderingfool2 said:

and exhibits a higher sustained rate than what is observed in yours

I just ran a test again and on an actual large file copy I get about 130MB/s average, but highly inconsistent, oscillates between 100 and 170 and is still less than half the read speed. Average speed is consistent over a 200GB copy so no PMR cache. 

I also have the same drive in an enclosure and I do get about half the speed with it so I know it's not the RAID that's inefficient. And I have known PMR 2.5" drives and they write >100MB/s without issue, and with similar read and write performance (this one is known to be SMR, there are no >3TB 2.5" drives that aren't).

 

So since my observations match the others that can be found of about a 2x reduction in write performance, plus the fact that the theory checks out (for every X data the drive actually has to write 2X to the platter) it's going to be hard to convince me that pure SMR doesn't cause a very significant slowdown.

 

Now caching can improve it, but that will only work on some workload types, and again what's missing is the information. How will it work in my particular application? Since it potentially can have serious impact on some workloads it is important to make it clear when a drive is SMR.

 

 

[mr moose]

7 hours ago, TechyBen said:

Citation? Linus when he tried shingled drives. Like I think it's impossible to have a discussion with you here. There is a cache. There is a requirement for it to remap the clustered read/write/read areas. It performs differently. Thus some may wish to buy different models with different performance. Some may even wish to get SMR disks with higher cache, but if it's not specced/noted, only buying and testing internally/personally will show the differences (between brands or in a brand).

Link to this evidence.  Seriously, you are making the claims you need to provide the evidence, i am not trawlng through hundreds of videos on shingled drives to find the data you need. 

 

 

7 hours ago, TechyBen said:

Why do I waste breath on this?

 

"I'm not buying brand X because the oxide always dies". Yeah. I know some brands and some devices. I know some processes and some products. I'm done, I'm not even reading the rest and I'm out. Not in protest. I've just got better things to do then argue over trivialities and insults.

Because unless you have just as much of a reason to demand knowing the oxide content as you do demand knowing if it is SMR. 

7 hours ago, RejZoR said:

Does oxide coating, size of coil, magnet core, IC vendor, specific BIOS code or whatever affects things notably? No. Does SMR? Yes, yes it does, very. What a stupid argument...

You don't know that and that is my point, you have no idea what the performance is like with different coatings or the difference to the consumer with SMR.  Unless you can show that a specific tech has a specific effect on the drive then you may as well be picking any part of that drive and demanding they list it.

5 hours ago, Kilrah said:

You dropped a 0. They're ST5000LM000 and from my experience must have no PMR cache. But there's no way to actually know, which is the whole point.

So you don't know if they are SMR, but you are assuming they are and you won't accept anyone telling you there is no way to know.   And you won't give me any kinks to data or tests that even remotely support your claims.

 

[Kilrah]

4 minutes ago, mr moose said:

So you don't know if they are SMR

Read the post, I know they are SMR, what I don't know is if they have PMR cache but everything points to "No".

[mr moose]

1 hour ago, Kilrah said:

Read the post, I know they are SMR, what I don't know is if they have PMR cache but everything points to "No".

O.K, enough evidence exists to say it is, I'll accept that.  so now I want to know why you are using laptop and external storage designated drives in a raid array.

 

Remember the storage triangle;  Fast, cheap, high capacity - pick two.   You went for high capacity and cheap and now you are arguing had you known that they were SMR before hand it would have made a difference to your purchase.  What if it turns out that for the designated use SMR make little to no observable difference? I mean had you bought an ironwolf or disk designed for raid and performance and had this issue then I would be very sympathetic to your cause (really I would),  but as it stands it looks very much like you bought a domestic storage disk designed to work on it's own, put it into a raid and now are upset because it doesn't perform like a professional disk.

 

Here's a post form a year ago about that exact drive in a raid array and he claims it was SMR back then, so Did you know it was SMR when you bought it?

 

[Lady Fitzgerald]

[leadeater]

9 hours ago, RejZoR said:

Does oxide coating, size of coil, magnet core, IC vendor, specific BIOS code or whatever affects things notably? No. Does SMR? Yes, yes it does, very. What a stupid argument...

The SMR implementations on disks are different, and I don't mean DM-SMR vs HA-SMR. A disk that is designed for general desktop usage i.e Seagate Barracuda will have a different design implementation to one that is designed for NAS or RAID workloads i.e. WD Red or one designed for archive workloads Seagate Exos 5E8.

 

All DM-SMR disks have a PMR zone btw. What I can tell you is that for the Seagate Archive disks the 8TB model has 25.6GB of Media Cache (PMR zone) and this is located on the outer tracks.

 

Because these are actually all different you can't just take an old review of say the Seagate Archive v2 and apply that to a different disk designed for a different purpose, neither can you for a Seagate Barracuda. If you want to know how something performs you actually have to test it, not all PMR/CMR disks are the same so why are you applying such a generalization to DM-SMR disks that you would not to PMR/CMR disks. There is a very big difference between a Seagate Barracuda 5400 RPM 64MB cache disk than there is for a Seagate Ironwolf, they are both PMR disks so they perform the same correct?

 

DM-SMR will have a performance impact on NAS disks, what that actually is has not been tested. The current reports about WD Reds are not a performance issues it's actually a fault with the product, the fault is not that it's slow it's that it's actually broken/buggy.

 

Additionally to that outside of resolving the bug and the disk being able to be added to arrays without failing even in the singular situation the rebuild time is slower than a PMR disks are you really going to hinge your entire purchasing decision over a one time event that might be slower as it'll be situational? These WD Red DM-SMR disks will perform better than the old revision using actual usage in NAS devices they were made to go in, the much larger cache will do that however making the cache 4 times larger doesn't come for free. Disks with 256MB and 512MB cache can be upwards of 10 times faster for small I/O writes.

 

So do you care about actually using the disk and the NAS for data storage and how it performs or do you only care about rebuilds? If you care about the actual usage of it do you want better performance under some workloads or all workloads? Do you want to pay more for the next revision of the disk or do you want the cost to stay the same or potentially even be less?

 

WD made a choice about progression of one of their product lines, evaluated the technologies they could utilize and made a balanced choice in the new design, one they believed in. They chose a design that increase cache which increase costs and balanced that with the usage of DM-SMR. This doesn't actually mean the new revision has worse performance, it can be both faster and slower and you need good testing of the actual product to know what these differences are. The design choices made didn't make it unsuitable for NAS usage, releasing a disk with broken firmware and lack of proper testing made it unsuitable for NAS usage.

 

And it may very well be that the changes made to the disk may no longer make it the best choice for your situation, just as it could have no difference at all.

 

Should WD have published the usage of DM-SMR, sure. From a me advising them standpoint doing so would mean if there is an issue, even if one is not expected, at least you said something before hand. If people did see this change they may not have purchased 3 at once and test 1 first. A storage review publication may have also decided to acquire the new disk revision and review it, would have been great for us if that happened, we'd actually have something to talk about properly.

 

Deciding not to publish the change or put it on the spec sheet is looking like it's biting them in the ass. However saying DM-SMR is not suited to NAS usage is incorrect. Also the 2 Seagate models that are suited for NAS usage that are listed in the OP are tailored for archival.

 

P.S. The disk firmware does have a significant performance impact.

[mr moose]

5 minutes ago, leadeater said:

snip

 

So basically what we know is that some disks have problems.   There are many compounding factors in these problems and we have no tests or data that points to a cause,  or even a performance metric for anyone of the outlined cases.  Yet people are adamant they know it is SMR alone and that lynch mob mentality is winning out.

 

So much for critical thinking in the 21st century.  And people think we are getting smarter as a race.

[leadeater]

18 minutes ago, mr moose said:

So basically what we know is that some disks have problems.   There are many compounding factors in these problems and we have no tests or data that points to a cause,  or even a performance metric for anyone of the outlined cases.

The only good test data and research papers I have found have been for the older SMR archival disks.

 

Here's a presentation on DM-SMR, relevant part starts at 4:50

https://www.usenix.org/conference/fast15/technical-sessions/presentation/aghayev

Source paper: https://www.usenix.org/system/files/conference/fast15/fast15-paper-aghayev.pdf

 

And the paper that lead me to this one:

https://www.researchgate.net/profile/Ziqi_Fan/publication/317393826_Performance_Evaluation_of_Host_Aware_Shingled_Magnetic_Recording_HA-SMR_Drives/links/5a28c839a6fdcc8e8671c0be/Performance-Evaluation-of-Host-Aware-Shingled-Magnetic-Recording-HA-SMR-Drives.pdf

[Lady Fitzgerald]