Did you really mean the CMR cache or the SMR cache?
Wouldn’t a normal RAID rebuild be almost completely linear? In that case the SMR drive (especially a new one with no data in it) could probably just write the data directly, bypassing the cache.
1 Like
SMR disks contain an amount of CMR storage to use as a write cache. New writes go to the CMR write cache, which are later replayed to the SMR area with spare IOPs. This way, SMR disks appear to have the same performance as CMR disks as long as the write throughput remains under a certain level.
3 Likes
thanks a lot, now that sentence makes sense to me. The CMR cache inside SMR drives.
1 Like
That’s a good point. I’m not sure if SMR disks recognize that and deal with that by skipping the CMR cache. I guess it should be tested. But it would not surprise me if you run into similar issues.
The two SMR drives I have (that I know of) are in use, so I cannot really test that.
In the article they said they “prepared” the drives by writing a lot of data to them, the large linear write probably went as normal, because if it took them days to initially fill the drive it would have been mentioned.
Also, if someone buys a drive, formats it and tries to copy his files to the drive as a backup etc, the speed drop would be really noticeable and the user may even think the drive was defective.
1 Like
They started with an array of only CMR disks. Only the new one they swapped in was SMR. So yes, that went normally, but since it was CMR at that point that doesn’t tell us anything.
Prior to beginning this sequence of tests, the drives were prepped by having 3TB of data written to them, and then 1TB of that data is deleted. Testing commenced immediately after the drive prep was completed.
Looks to me that they created a 1TB “dirty” are on the drives to properly test the performance, because you would probable be able to completely fill a brand new drive sequentially at normal speed. ZFS resilver is not linear.
EDIT: So, basically, SMR drives can work in ZFS for Storj, as long as the node does not get a lot of ingress (since I have two of them in my pool and they work fine), however, once a drive fails, you absolutely have to replace it with a CMR drive or you will wait for weeks for the pool to rebuild even if the pool is otherwise idle.
A class action lawsuit has been filed against WD regarding this. If you bought one of those disks you can contact the law firm on this page.
2 Likes
Just to be clear:
Did WD actually change these drives to SMR while keeping their SKUs, so customers have no way of telling what drive they get?
Or did they new drives get new SKUs as well, in which case you could simply not buy these?
If the latter is the case, I fail to see why you would file a lawsuit…
It’s the latter. You would have to ask the law firm, I don’t really agree with their wording on that page. But my guess would be that the Red brand stands for certain promises that these drives simply don’t meet. I’m not a lawyer though, I’m leaving the lawyering up to the lawyers.
Just providing the link in case others find it useful.
1 Like
It’s in the data sheets, so you get what you buy. If people are unable to read, that’s not WDs fault.
That shady law firm wants to make some good dollar, nothing more 
It didn’t used to be.
But yeah, law firms want to make money. I don’t think that’s breaking news. I’m merely providing the link for those interested in joining the class action. I’m not endorsing or promoting it. Everyone can decide for themselves.
1 Like
What do you mean by that?
a) the data sheet didn’t state the kind of recording technique used in the drive?
b) it was falsely advertised as a CMR drive?
c) there hasn’t been a spec sheet?
The first one. They omitted that information until this whole thing exploded and they publicly published which drives used SMR. They must have added this to the data sheets afterwards.
This, and when the journalist called WD and asked about it the first answer from them was “that information is not disclosed to consumers.”
(this link was in one of the first posts).
Seagate and Toshiba owned up to using SMR when asked. However, only WD advertised a SMR drive as “designed for NAS”, where the Toshiba and Seagate drives were only advertised as “for desktop”.
Only after the backlash WD and Toshiba published the list of SMR drive models and included this information in the datasheet, so there was time period when someone could buy a Red drive having no idea that it is SMR and then have problems resilvering a zfs pool (IIRC this is how this whole issue was discovered in the first place).
Seagate has not published the list of SMR models, but promised that they know that SMR cannot be used in a NAS and thus their IronWolf line is definitely CMR.
6 Likes
Yes, there’s in the datasheet, some red HDDs are CMR and some red HDDs are SMR. We all know it.
But…wouldn’t have been better to clearly distinguish models by product line? i.e.: red=CMR, purple=SMR (or whathever)?
It seems the marketing is trying to confuse customers.
It’s in there right now, it wasn’t before, so people really had no way of knowing that the drive was SMR without buying and testing it.
Purple is probably the only model that WD would never be able to make SMR - it is designed for surveillance and video cameras record and overwrite old data. This results in random writes and the hard drive has to have guaranteed minimum performance (though it can be slower overall).
A funny fact was that in my local shop, a Purple (CMR) and same capacity SMR Red drive had almost the same price. Purple was even a few Euros cheaper. They both have TLER.
1 Like
I just picked a random color 
I wanted to say that it should be better to highlight that important property in the packaging as other specs like size, RPM, interface, etc. rather than hiding it in the specs.
I agree, however, before the backlash it wsa not even in the specs. WD just wanted to sell a cheaper-to-make SMR drive at the same price as a CMR drive hoping that nobody would notice.
2 Likes