My SMR drive - opinions please

[Shrek]

Seriously, i feel like we're a decade overdue for more filesystem options in windows, even if it's just on non-OS partitions

I thought one could now boot from a ZFS drive

 

[Chrispy_]

[SMR drives] have a CMR cache area kind of like how TLC/QLC SSDs have a SLC cache, so if the writes are not high enough to not go in that cache there should be little to no performance difference in theory.

That's another annoyance of the one SMR drive I accidentally ended up with, though it depends if you have host-managed SMR or drive-managed SMR.

Mechanical drives aren't quiet; There's the 7200rpm whine, and depending on what case they're in there's the constant chatter of head moves. With an SMR drive that writes to the CMR cache area, there's several minutes of background chatter long after the OS says that the copy/move/write is finished and the drive claims to be idle.

It's not really a big deal but I found it to be yet another annoyance of SMR and since mechanical drives are now the realm of large capacity datasets that I can't justify on SSDs, chances are good that we're talking about hundreds of gigabytes which means that a typical write to the drive completely saturates its cache area and takes the maximum possible time to calm down after it's "finished".

 

[Shrek]

...there's several minutes of background chatter long after the OS says that the copy/move/write is finished and the drive claims to be idle.

But does that not increase the chances of corruption should there be a power cut?

 

[lexluthermiester]

SMR fine for WORM-style usage.
Keep it unless the lack of write performance annoys you and then buy either an Ironwolf/Ironwolf Pro or Red+/Red Pro.

As shown earlier in the CrystalDiskMark run, performance is solid 165MB to 170MB per second. Qubit's drive and the one I have perform on par with each other, so they shouldn't have any issues.

But does that not increase the chances of corruption should there be a power cut?

That wouldn't be any different than any other drive in a power-cut.

 

[Shrek]

That wouldn't be any different than any other drive in a power-cut.

Except, there is more chance catching it when busy if it spends minutes tidying up the cache.

Unless it is clever and does not validate a move till it is complete, that way it can recover from being disturbed.

 

[Chrispy_]

But does that not increase the chances of corruption should there be a power cut?

No, the data is moved from CMR area to PMR area. I'd be surprised if the drive wipes the source sectors in the CMR area before it's finished writing to the SMR area. Presumably it'll just continue where it left off after power is restored.

As shown earlier in the CrystalDiskMark run, performance is solid 165MB to 170MB per second. Qubit's drive and the one I have perform on par with each other, so they shouldn't have any issues.

SMR's downside isn't sequential writes to blank areas of disk, it's IOPS - and since even decent mechanical drives are such a terrible choice for IOPS compared to the alternative options these days, that downside is becoming less relevant by the day.

I still personally avoid them, but SMR isn't bad when used the way they're intended to be used - as single (non-array) secondary storage drives. Don't use them in RAID, don't install your OS on them, and you'll be fine.

 

[lexluthermiester]

Except, there is more chance catching it when busy if it spends minutes tidying up the cache.

Um, not really and because...

Unless it is clever and does not validate a move till it is complete, that way it can recover from being disturbed.

...all modern drives do. That's been a thing since the inception of the SATA/SAS protocols(AFAIK)..

SMR's downside isn't sequential writes to blank areas of disk, it's IOPS

Given the benchmarks Qubit and I both ran earlier in the thread, that doesn't seem to be the case.

Don't use them in RAID

It would be better said, don't mix CMR and SMR drives in an array. Mixed drives is a bad idea anyway.

don't install your OS on them

Again, given the performance shown, I can't agree with this. It's not optimal, sure, but one could do fine with such a drive.

 

[Chrispy_]

Given the benchmarks Qubit and I both ran earlier in the thread, that doesn't seem to be the case.

Again, given the performance shown, I can't agree with this. It's not optimal, sure, but one could do fine with such a drive.

You're using a tiny 2GiB test area in CDM and that's barely even stressing the DRAM cache, let alone the CMR staging cache. Performance is fine (for mechanical) in that CDM test simply because it's too small and too short.

As for using CMR for your OS, that was the use-case that caused the majority of the hate for SMR in the first place. I don't have a strong opinion on it because IMO mechanical drives are a poor choice for your OS regardless of whether they're SMR or CMR. It's definitely possible to run your OS on an SMR drive if you have no better alternative, but it's sometimes going to crawl along, especially if you are short on RAM and the pagefile starts getting hammered.

There are literally hundreds of threads with tens or hundreds of thousands of complaints about laptops and desktops with SMR OS drives by now. They're just all old, dead threads by now and nobody cares any more since you can pick up a new 128GB SSD on Amazon for under $20 or possibly just find one being handed out for free (I gave away 150+ 128GB drives to Techreport forum users FOC because they weren't worth enough to bother selling).

 

[ThrashZone]

Hi,
Yeah you might as well throw out a as ssd test as some sort of proof, both are just playware

 

[swerist]

My choice is HGST or Hatachi since last 10 years
Now with 6 HC320 drives.
Also, using air drive not Helium drive

 

[lexluthermiester]

You're using a tiny 2GiB test area in CDM and that's barely even stressing the DRAM cache

Oh? One moment...

Doesn't make much of a difference does it? But hey let's try the maximum size CDM can test...

Hmm. Margin of error kind of thing.

but it's sometimes going to crawl along

175MB per second is hardly a crawl unless you compare to SSD's, which you clearly want to. In reality most people could run an OS with those speeds and be perfectly happy. So having an SMR based drive as mass storage is simply not a problem.

Yeah you might as well throw out a as ssd test as some sort of proof, both are just playware

Ok, let's try something else..

Anyone want to argue with this?

 

[ThrashZone]

Hi,
Sure did you make a point somewhere ?

 

[lexluthermiester]

Hi,
Sure did you make a point somewhere ?

You tell me. You called CDM "playware", so I answered with a different and much more involved/grueling test.. Results didn't really change much, especially given that write caching was bypassed and the file size was 16GB which would push passed any drive caching schemes, effecting direct to disk writes.
BTW, that test took 48minutes to complete.

 

[ThrashZone]

You tell me. You called CDM "playware", so I answered with a different and much more involved/grueling test.. Results didn't really change much, especially given that write caching was bypassed and the file size was 16GB which would push passed any drive caching schemes.

Hi,
Only thing I noticed was you increased the space tested
Besides that you may call it margin of error but I see inconsistencies between the same playware and even compared to another being atto I didn't mention but was part of a few I was referring to
Throw in as-ssd and all three will be different

So you're kind of proving my my point

Only thing these "utilities" might prove is just basic trouble shooting performance issues.
Otherwise they just hammer small blocks by default.

 

[chrcoluk]

That's another annoyance of the one SMR drive I accidentally ended up with, though it depends if you have host-managed SMR or drive-managed SMR.

Mechanical drives aren't quiet; There's the 7200rpm whine, and depending on what case they're in there's the constant chatter of head moves. With an SMR drive that writes to the CMR cache area, there's several minutes of background chatter long after the OS says that the copy/move/write is finished and the drive claims to be idle.

It's not really a big deal but I found it to be yet another annoyance of SMR and since mechanical drives are now the realm of large capacity datasets that I can't justify on SSDs, chances are good that we're talking about hundreds of gigabytes which means that a typical write to the drive completely saturates its cache area and takes the maximum possible time to calm down after it's "finished".

Yep I had that concern and another reason I stayed away. Whats weird is how closely priced SMR is to CMR, probably why they tried to get away with not labelling it on the spec.

 

[lexluthermiester]

Hi,
Only thing I noticed was you increased the space tested
Besides that you may call it margin of error but I see inconsistencies between the same playware and even compared to another being atto I didn't mention but was part of a few I was referring to
Throw in as-ssd and all three will be different

So you're kind of proving my my point

Only thing these "utilities" might prove is just basic trouble shooting performance issues.
Otherwise they just hammer small blocks by default.

Given that opinion, there's no testing method that would satisfy your position, which has the effect of rendering your point mute. So believe whatever you want and ignore science.

 

[ThrashZone]

Given that opinion, there's no testing method that would satisfy your position, which has the effect of rendering your point mute. So believe whatever you want and ignore science.

Hi,
I've used all three on 2 adata sx8200 pro's testing
Returned both of them from the terribly inaccurate read/ write of the adata spec's sheet of them on two different systems
So they have a purpose but not just one of them.

 

[qubit]

Mechanical drives aren't quiet; There's the 7200rpm whine, and depending on what case they're in there's the constant chatter of head moves. With an SMR drive that writes to the CMR cache area, there's several minutes of background chatter long after the OS says that the copy/move/write is finished and the drive claims to be idle.

In an absolute sense no, they're not quiet. Thankfully, depending on the model, they can be quiet enough, like the WD Blue that's the subject of my thread here.

My PC is in an open case on the table right next to me. All I hear is the quietest swish from the bearing if I listen out for it, no whine and if the head is doing lots of accessing, a slight rattle, nothing too obtrusive. Clearly these drives are engineered for quietness, which includes reduced spindle speeds. Note that there is some fan noise from the PSU and a tiny bit from the CPU cooler, which helps to cover up the drive noise.

 

[ThrashZone]

Hi,
Yeah I've never notice my WD blacks whine at all
Same here on the open bench with 2 1tb right next to me I'd know it if they made any sounds

 

[qubit]

Hi,
Yeah I've never notice my WD blacks whine at all
Same here on the open bench with 2 1tb right next to me I'd know it if they made any sounds

Thinking about it, my WD Black is pretty quiet too. Bit noisier than the Blue.

 

[lexluthermiester]

Bit noisier than the Blue.

That's to be expected.

 

[Mussels]

Except, there is more chance catching it when busy if it spends minutes tidying up the cache.

Unless it is clever and does not validate a move till it is complete, that way it can recover from being disturbed.

NTFS does that, writing into a empty location and only when complete does the file list get updated marked as active for those clusters

As an example, when you defrag it keeps the original location (and original fragments) in place until they're finished moving, then adds the new record before deleting the old one. This is how un-delete programs work, since the files are still present and only the listing in the file table was removed.

Various file systems do it differently and that's probably only 90% correct, but close enough

 

[lexluthermiester]

NTFS does that, writing into a empty location and only when complete does the file list get updated marked as active for those clusters

As an example, when you defrag it keeps the original location (and original fragments) in place until they're finished moving, then adds the new record before deleting the old one. This is how un-delete programs work, since the files are still present and only the listing in the file table was removed.

Various file systems do it differently and that's probably only 90% correct, but close enough

Drive controllers do this natively as well.

 

[Athlonite]

Seeing as WD saw fit to rename their Green line of drives Blue I'd be inclined to burn it with fire then bury it in the back garden
if you're looking for serious long term storage get an HDD meant for servers or a NAS like WD's Red Pro or Seagate's Ironwolf series

 

[qubit]

That's to be expected.

Indeed, it's much higher performance.