WD Red CMR vs DM SMR - Mixed RAID 10TB RAID ReBuild Test

[@augurseer]

I must say. This series on CMR SMR and its impacts on NAS is fantastic. I love the real world approach and thinking.

[@SirHackaL0t.]

You spent a few videos going through different scenarios but when you get to the one that most people are probably interested in you change the setup. Can you check how long a rebuild on the non shingled drives would take so we can compare the time for shingled/non shingled drives.

[@cataria3903]

all those raids are conventional raid setups, correct? could u do a ZFS storage setup rebuild at 95% full please? i heard lots of reports of SMR garbage drives creating issues in ZFS storage setups, so it would be great, if u could do rebuilds on full ZFS to test this.

[@TechyBen]

Yeah. Again. "Reddit are all lying, you customers are all stupid, nothing breaks in RAID with SMR, oh, by the way, buy CMR for raid as it's better performance". Is totally missing the point of 1) Reddit only ever claimed these drives failed with ZFS full raid systems due to read/modify/write delays and 2) they could not buy CMR *because the companies submarined the drives into the stock listings!!!*

[@azuravian]

The 59.15% slower is incorrect. This is closer to 70%. One took a bit over 10.5 hours, the other took 18 hours. (1082 minutes - 640 minutes)/640 minutes = 69.06% worse.

[@TrevorJackman]

It would be interesting to see how the Synology NAS would behave if you were using an SHR raid of AX drives and second Synology NAS with RX drives and then trying to expand the array by introducing a new AX drive when they are >95% populated. Some people choose the Synology NAS for the Features of SHR over others. Also for the tests that you performed can you explain in the final overview if there was any issues or notable performance problems while trying to populate the NAS drives to nearly 10 TB?

[@tangofan4u]

This comparison isn't all that useful, since in both cases you pull and reinsert an AX (SMR) drive. The interesting part would be to have a setup with 4RX drives comparing with a setup with 3RX and one AX drive. Then compare the rebuild when an AX drive is pulled vs. when an RX drive is pulled. This simulates a user having the choice between an SMR and a CMR drive, when replacing a failed drive. (The other 3 drives doing necessarily have to be RX drives as long as they are identical in both setups.)

[@panoshountis1516]

Great comparison, Rob. Thank you! I think it would make sense if you do the same with SHR instead. Also, kindly post the results data in a table someplace so we can view and digest it in peace.

[@rickhulbert5175]

Why do you pull only the single AS drive in both scenarios?

[@PingJerry]

Pulling two drives out of a raid5 destroys the data. Raid5 can only sustain a single drive failure/loss.

[@kaisa85ks]

please test this setup: 4x RX 95% full, then pull one of the RX and replace with an AX and make a rebuild. (this is my setup, 4x RX and i am worried what happens when one disk will die and must be replace with an AX because RX are out of market)

[@bigphaddy]

I think this is the most important scenario that most people are going to encounter for sure, as most people will be running the RX drives at the moment and will at some point "have to" replace them with the AX drives as RX drives will cease to be available 👍 It would also be useful to maybe do some quick/simple read/write tests with say 4 x RX drives populated then replace one RX drive with an AX drive and rerun the tests (4 x RX vs 3 x RX + 1 x AX). That would show if the system would suffer by swapping out 1 failed drive for a newer AX variant 👌🤔

[@srfurley]

I recently, 12 days ago, installed a home NAS, having previously had experience with multiple 16 bay rack mount systems until I retired from working in I.T. seven years ago. I decided I didn’t need the Red Pro drives, and didn’t really want them for cost and noise reasons. Had difficulty seeing what the difference was between Red and Red Plus drives, and chose the latter. To make matters more confusing I had noticed Red drives advertised with the same capacity but different model numbers and different cache size. Didn’t understand the difference at the time. Have I understood this correctly and the RX Red drives are now discontinued? Are the Red Plus drives basically the same as the older RX Red drives; if not, what is the difference? Would a Red Plus drive be a good choice for somebody who has a RX Red array, and needs to replace a failed drive? There doesn’t seem to be much mention of the Red Plus range.

[@radumurzea6112]

In that text file at the end, all the results are marked as being for RAID5 arrays, but sooner in the video (e.g. at 07:35 ), when starting the rebuild in the case of Synology, you can clearly see it's mentioned to be SHR (Synology's proprietary RAID implementation). What's up with that? Can you clarify this please? Thanks.

[@tangofan4u]

SHR-1 is essentially RAID1/5 with drives of the same size.

[@jinx0192]

Question: Ok so I bought one of the ax 4tb wd drives before all the news came out about them being SMR drives. I am going to use this in a home environment. How screwed am I?

[@leexgx]

The Synology results seem to be the more expected results for a smr disk (the qnap one seems but odd how fast it did it, but as they are AX drives results can not really be consistent as they perform more like a QLC ssd and qnap might rebuild the drives very differently from Synology) Other problems can happen after 2-8 weeks with smr drive use (it really depends how much write data activity there is Drobo's really dislike smr disks

[@LimbaZero]

Simple answer could be that QNAP detects SMR driver and using bigger cache for one or multiple pages (Sorry for my naming from flash technology) for writing in rebuild scenario so no penalty over CMR. Also drives cache should be enough if rebuild is done with linear read/write from disk and not from filesystem by files

[@PitboyHarmony1]

That 59% alone is enough of a reason for concern to wave my yellow flag. Begs the question; how many more actual physical revolutions, and shingle read/writes the AX drives had to endure, to complete the rebuild in this test, compared to the RX drives? ie; how much closer to normal drive, end of life, failure did we get just from this test? Why would I want a set of disks that appear guaranteed to fail sooner, even if admitting this only impacts drives during raid rebuilds on near full drives, which is not regular daily use of a NAS. Still ... why would I buy these? Although AX are cheaper to buy, in the longest run, they are not as you will need to replace them more often and accept a potentially higher risk of multi drive failures therefore possible raid failures, and loss of data. To test next? Using SHR, 4 near full AX disks, replace one with an empty RX . People leapt at these cheaper AX, and with results like this, may plan to slowly convert to RX, as they fail. Also curious about the errors experienced in the QNAP machine during this rebuild. Were they operator related or related to AX shingle read / write issues and why did errors show up in the QNAP and not Synology, if the data were mirrors of each other all round.?

[@pgotze]

I have just extended Synology DS918+ from 2 *FRX disks SHR to 4 *FRX disks SHR - 92% full volume - total time 61h10m (almost 2 and half days). Can imagine, that if i would extend original 2 *FRX by adding another 2 *FAX, i would have problems. Shame there was not such a test. All disks are 4TB disks.

[@awesomearizona-dino]

This is exciting... Just Before the SMR news came out , i bought a 4tb AX drive. Getting my chips(aka crisps) and soda first

[@LimbaZero]

Single drive may not be problem if you don't do lots of random writes.

[@Rob_III]

Weird; I'm currently in my last rebuild of my SHR2 array with 5x WD60EFAX being replaced by WD6003FFBX drives and only about 6.9TB of the available 15.7TB is used and my rebuilds up until now all take *at least* 44 hours. I'm keeping track of the entire thing here: blog.robiii.nl/2020/04/wd-red-nas-drives-use-smr-and-im-not.html I have to admit that, during this entire rebuilding (and rebuilding and rebuilding and...) my Surveilance Station has continued to keep writing 6 camera streams (about 8MB/s) and about 4 (pretty much unused) Linux VM's being mounted as iSCSI in VMWare but their I/O is negligible. I have 5x6TB and SHR2 whereas you have 4x4TB and SHR; I guess the biggest difference would be SHR vs SHR2 then?

[@LimbaZero]

Multiple active write streams is problematic in SMR. With SMR you start to have requirement for SSD caching without 10G interface or at least some kind bigger write caching. you should write amount of "page" data at one time or multiple of it to SMR drive.

[@d3xbot]

And that, Western Digital, is why you listen to your community when they call you out on SMR (and other things)

[@jeffm2787]

Bar graphs .... nuff said.

[@vaguedirector_7342]

Interesting results. SMR drives seem to work fine in a RAID environment . Even if they rebuilt at half the speed it doesn't really matter IMO as rebuilding is a rare event, all that matters is that they do it successfully.

[@radumurzea6112]

If it takes longer, that suggests the drives need to go a more intense load/stress in order to complete the operation. Which increases the risk of drives failing while the rebuild is going on, especially if the other drives (from which the read is done) are also kind of old.

[@TechyBen]

@@radumurzea6112 THIS!!!

[@mrfrenzy.]

Please try to Secure Erase the drive before rebuilding so you get "a brand new drive". It is not certain formatting is enough.

[@SirHackaL0t.]

The NAS wouldn’t trust the data so a secure erase shouldn’t make any difference.

[@mrfrenzy.]

@@SirHackaL0t. If you don't secure erase the harddrive firmware will still think the drive is full of important data so it has to move tracks in and out of cache in order to do the SMR write process. The same problem happens if you use an SSD on a raid controller without Trim.

[@leexgx]

Even formatting it wasn't needed as the whole drive would have been overwritten in the rebuild process any way, if anything it actually made the smr AX drives faster to rebuild as they was empty

[@perkunast9680]

I think i can beat you all of you clowns, I have two, 4 bay nas boxes , both in raid 0, one is a back up. Super fast almost SSD and in theory 1 nas can catch fire, and i wont lose my data.

[@AllanSavolainen]

One is not a backup, you need to copies of the data before you have a backup.

[@timothydavies5337]

you dont anything about computers FOOOOOOOOOOOOOOOOOOL