Thread 107078661

They would claim they aren't but its not true.
SMR status is hidden.

They will destroy your data unlike CMR. But its cheap. And dont think its only about Seagate.

https://www.heise.de/en/news/WD-to-investigate-problems-with-older-SMR-hard-drives-10872057.html

2TB to 6TB WD Reds were SMR and they broke down.

Drives:
WD RED:
> WD*0EZAZ, WD*0EDAZ, WD*0EFAX

They are all released since 2020, older than year 2020 might be "relatively fine" as far as SMR goes.

Even if SMR doesnt break, its performance is worse than in CMR. There really is nothing good about the tech.

what about my WD60EFPX

>>107078666
I'm afraid you got the case of "hidden SMR"

its kind of a hybrid between CMR and SMR drive, not true CMR

yeah none of this means fuck all to me.
>dude SMR
>dude CMR
maybe explain what that means BITCH
I don't work for BIG HDD as an engineer

>google it
no!

HDDs deprecated by SSD and considered harmful

>>107078661 (OP)
I have these drives:
WD6003FRYZ
ST6000NM019B-2TG
How bad are those?

>>107078704
my efpx is faster than my seagate cmr despite spinning slower too

>>107078743
ask grok then

>>107078759
I would blame seagate itself for this, not the technologies involved

>>107078743
SMR writes some data "in between" two data tracks, CMR doesnt, but works in more exact manner.

What this means is SMR allows more stuff written in same physical area (like a magnetic disk made in 3.5" format, the real life size) but SMR works slower in both read and write operations when the cache (256MB?) has been filled. Once cache is full it needs to take several readings of the SMR region. In CMR on the other hand one read operation works reliably instantly and there is no need to do a second one to make sure it read correctly.

>>107078764
no!

>>107078782
Thanks

>>107078661 (OP)
Read about Blue/Red WD drives failing the other day/week...

My bet is that it's the old 'head parking' issue. Unless WD has changed within the last decade or two, the old Blue/Green drives were essentially re-branded Red drives that had super aggressive head parking--8 seconds, along with regular head parking that nearly every modern hard drive has. The trick was to edit the firmware to completely disabled head parking, effectively making Blue and Green drives psuedo-Red drives. I currently have three white-label Red drives in my NAS that I've had zero issues with, along with the litany of modified Green/Blue drives I've had over the years that have never failed.

As far as I know, once you get past a certain capacity, most hard drives are SMR since CMR is incompatible with higher capacities. I'm sure there's the odd model out there, but that's like sourcing SLC or MLC SSD's when they're exponentially more expensive and lower capacity than TLC.

Mind, I learned about this years ago--I don't think I have the time, patience, or money anymore chasing my OCD whims.

How about you just tell us which drives are good instead of which ones are bad

>>107078842
are you me?

i think the WD enterprise gold drives should be good enough for anything vital though, its what ive been using for several years since i dont have time to shuck drives and shit anymore

>>107078855
Not that poster, but I feel like a more important question is, "how large is your wallet."

Instead of spending a small fortune on a single drive, only for it to prematurely fail, it's better to get multiple smaller drives and RAID them for redundancy.

>>107078863
I've been shucking Easystore drives myself for the white labels.

I acknowledge that's a gamble though since they tend to have the 3.3v issue that you have to work around/tape off. Everything is so expensive nowadays, so I need to keep my data hoarding to a minimum, for the sake of my wallet.

I'll probably just buy NAS drives outright, assuming I don't need to make a purchase any time soon.

>>107078842
CMR in fact can be done in capacity of 20TB
it just that it has then 10x drive platters instead of the usual 2-4

its impossible to write more than 2TB onto a single 3.5" platter on CMR

shouldnt be a surprise that manufacturing 20TB CMR drive which has 10 drive platters is a lot more expensive than 8-16TB SMR drive which has 2 to 4 platters with 2020s technology

>>107078661 (OP)
Oh look the schizo is back
Can't wait for another thread with zero evidence

>>107078661 (OP)
do they satill silver plate their boards and have the head pogo contacts tarnish and the drive apears to have "failed" but in reality the contacts just need to be cleaned?

>>107078666
>>107078704
>>107078759

All modern air filled hard drives up to 8TB are SMR with some exceptions (certain old 1-2TB CMR models which somehow are still in production), including Red Plus, Purple, Ironwolf and even Gold/Ultrastar etc. You can easily confirm it yourself by seeing their behavior in raid/zfs or doing long random small block writes.

Here are several examples of them failing during resilvering/scrubbing. It takes 10 times longer to resilver a new Red Plus/Ironwolf compared to old hard drives.

https://forums.freebsd.org/threads/seagate-ironwolf-doesnt-like-zfs.98545/

https://forums.servethehome.com/index.php?threads/latest-seagate-ironwolf-4to-st4000vn006-supposed-to-be-cmr-behaves-like-smr-in-zfs-raidz-rebuild.37585/

https://old.reddit.com/r/zfs/comments/1lg744y/what_is_a_normal_resilver_time/

https://old.reddit.com/r/zfs/comments/1mskfla/zfs_resilvering_6mbs/

They've simply changed the definition of CMR, now CMR means SMR without TRIM. If you want last genuine CMR models then your best bet'd be seeking for used HGST Ultrastars because modern Golds/Ultrastars are also SMR. (serial number starts with WUS* not HUS*). In short - CMR is history and SMR is shit even for cold storage because it fragments data even during reading.

>>107078661 (OP)
Why would you buy reds and not golds?

>>107079121
>schizo is back
this whole board is schizos. they're the only ones with enough time on their hands to maintain the dozens of generals.

>>107079764

New gold models are also SMR with 2TB per platter density unless they're helium because it's physically impossible to create intertrack gaps between such tracks without overlapping and heads cannot read and write on such tiny tracks without a second layer translator.

>>107078661 (OP)
just use an ssd retard. there's literally no downsides

>>107080215
Price is the downside.

>>107079825
Where the 10TB air drives at?

>>107079754
>All modern air filled hard drives up to 8TB are SMR with some exceptions
lies.

I'm patiently waiting for the AI bubble to pop so hard drives become cheap again

>>107082343
prices only go up, they choke production to keep them high

>>107082343
Seagate Exos 28TB for 370€ (includes 19% VAT)
which comes out as 13.21€ per 1TB
what more do you need?

>>107082514
>refurbished
sure thing lol

>>107078661 (OP)
Normal air drives are cheap slop now, so it's not surprising they put SMR in them. Either get a helium drive if you want a lot of space, or an SSD if you want it to be fast.

>>107078842
Red drives have used 60 sec head parking for over 15 years now.

Greens haven't been on the market since fuck knows when. Blues were the bottom of the barrel cheap crap that you should not buy. If you buy either, you deserve your data to die off.

Just write something to the drive every 5 seconds to avoid head parking

>>107079114
>its impossible to write more than 2TB onto a single 3.5" platter on CMR

Not impossible. Toshiba has 24TB 10 platter CMR drives. Seagate has even larger ones.

The only trick is that so dense platters need to be made with more resistant materials, and you need energy assisted write heads for those. And these are more expensive, so you only see them in very large enterprise drives. For smaller drives you get SMR, because they are too small for enterprise to buy those, and too expensive for consumers to pay premium for quality.

>>107079754
more horse shit.

The zfs resilver on those takes long because resilver is heave on the random i/o, which is incidentally why it kills SMR drives. It's normal for it to run slowly if you are running it on a highly fragmented old array. The comments themselves state these. And at least one of those links mention drives that predate SMR existing.

>SMR is shit even for cold storage because it fragments data even during reading.
that's zfs attempting to defragment itself during resilvering, not SMR.

>>107080251
gold 10tb air is cmr WD103KRYZ

>>107083564

Did you even read what he wrote?

"With the ST4000VN008 it takes between 3 and 4 hours to resilver the pool, with the ST4000VN006 that operation took 30 hours with the NAS basically idle !!!"

When looking at the write speed during resilvering, you see that the VN006 model resilver at around 11MB/s whereas the VN008 do the same operation at a around 95MB/s"

>It's normal for it to run slowly if you are running it on a highly fragmented old array.

Again I'll quite one of these users again:
"No they weren't. Most ultrastar and deskstar were also available as "coolspin" variants with 5700 rpm, like this one (which is doing well at 119270 hours of operation, with no such resilver problems):"

>>107083496

It's only possible for helium filled drives with a different recording technology. For air filled CMR drives the physical limit would be 1,67TB per platter because otherwise the magnetic head will be too wide to be able to read and write on such tiny and dense track.

>>107078661 (OP)
what's even the reason having huge hdd? what do you store there, anon?

i could understand one needs quite huge ssd for LLM models, as loading 90gb into ram might take some time on hdd.

so what's that space occupied with?

>>107085077
I store my LLMs on huge HDDs

>>107085077
A 4k movie is easily 30GB and more.

>>107085204
>I store my Large Loaded Milkies on huge Hanging DDs

>>107084960
>Did you even read what he wrote?

Did you?
Resilver times depend on the fragmentation of data on the zfs array. If it is a heavily fragmented parity array, it can take quite a lot longer. Parity array = more random IO, Fragmented data = more random IO, more random IO = HDD speeds will tank because HDDs cannot handle random IO.

>For air filled CMR drives the physical limit would be 1,67TB per platter because otherwise the magnetic head will be too wide to be able to read and write on such tiny and dense track.
That's not the issue, the issue is that the magnetic grains will be unable to hold onto their polarity if you only write to a too small amount of them. Which results in random bit flips. The way to get around that is using platter material that's more magnetically stable, which also means it's harder to write on them. That's what energy assisted writing fixes. Toshiba uses MAMR (microwave), seagate uses HAMR (heat assisted), WD uses ePMR (they just put more juice in the heads).

This requires more expensive material because the magnetic heads need to withstand the extra energy without deforming.
And because it is more expensive, they are not putting it in $15 2.5" drives, only bleeding edge enterprise grade stuff.

>>107078743
>Continuous Magnetic Recording
Hard drives have a read head and a write head. The write head records data onto a track with a finite width, but usually, this track is much wider than what the read head actually needs to read data. What this means is that you can overwrite a section of data without touching adjacent tracks.
>Shingled Magnetic Recording
To squeeze more data onto the same physical space, many manufacturers have switched to SMR, where writes to adjacent tracks will overlap somewhat. It's no problem for the tiny read head, but it does mean that if you want to overwrite data, you will inevitably be damaging adjacent data and will have to rebuild it, which, depending on how full the drive is, could vastly extend data write times.

>>107085416

>Resilver times depend on the fragmentation of data on the zfs array

Do you know why the data on ST4000VN006 becomes very fragmented but on ST4000VN008 it's doesn't?

Because ST4000VN006 is an SMR drive that has a second layer translator which automatically fragments data and throws it on random parts of the platter even during linear large block writes.

Here's how it looks like on paper. Logical sectors and zones are not aligned with physical adresses on the platter unlike on CMR. While a CMR drive writes the data very consistently, sector after sector, track after track (unless heavily fragmented but even then it's magnitudes faster than any SMR).

This is why SMR shits itself during torrenting and raid scrubs. Hell, even during just reads 2nd level translator can randomly fragment stored data even further because such technology is nothing but malicious malpractice.

>And because it is more expensive, they are not putting it in $15 2.5" drives, only bleeding edge enterprise grade stuff.

And despite that even air filled WD Gold/Ultrastar Vela-A1 models are also SMR, or such drives should no longer be considered enterprise grade?

>>107085883
I was talking about the last two links specifically which mentioned WD30EFRX drives, those cannot possibly be SMR yet still showed slow resilvering.
Automatically ignored the first two links because
>Seagate

>tfw have SMR drives
How can I avoid running into severe issues?