Are SSD drives suitable for RAID-1 use these days?

[NeddySeagoon]

ipic,

That's it. You can even add mostly-write to a member of an existing RAID.
The docs say, or used to say, that its a create time option.

[ipic]

Thanks Neddy,
I have set the writemostly status on an existing raid member by doing the following:

Code: Select all

echo writemostly > /sys/block/md1/md/dev-sda1/state

I failed to find a way to do this using mdadm, without failing and then re-adding the drive. For example:

Code: Select all

mdadm /dev/md1 --write-mostly /dev/sda1
mdadm: --write-mostly does not set the mode, and so cannot be the first option.

mdadm --manage /dev/md1 --write-mostly /dev/sda1
mdadm: Must give one of -a/-r/-f for subsequent devices at /dev/sda1

I'll have to wait to the next reboot to see if the 'echo' approach makes a permanent change.
The drives being on SSD's, failing and re-adding will take a significant chunk out of their TBW allowance.

[NeddySeagoon]

ipic,

A readd of an almost fresh raid element need not do a whole resilver.

Updating the chunks in the write intent bitmap MAY be sufficient. It may not be too.
I think the echo approach works.

I can't test any longer. The system with the NVMe/SSD raid1 set has failed big style. It has a short on the +12v on the motherboard.

[ipic]

Just rebooted, so quick confirmation that this:

Code: Select all

echo writemostly > /sys/block/md1/md/dev-sda1/state

does indeed survive the re-boot. So that change is permanent.

Sorry about your failure, @NeddySeagoon, here's hoping the short didn't take the drives with it.

[ipic]

As suggested way back at the start of this thread, here is an occasional update on the state of the SSD drives in my RAID-1 arrays:

Code: Select all

FIKWOT FS810 4TB                             Written: 13.99TB  Power on for: 10.33 months
CT4000MX500SSD1                              Written: 14.50TB TBW used%: 1.4499%  Power on for: 10.46 months
CT2000MX500SSD1                              Written: 3.07TB  TBW used%: 0.4389%  Power on for: 5.32 months
CT2000MX500SSD1                              Written: 3.23TB  TBW used%: 0.4621%  Power on for: 6.45 months
Samsung SSD 870 EVO 2TB                      Written: 4.89TB  TBW used%: 0.4074%  Power on for: 4.84 months
WD_BLACK SN850X 4000GB                       Written: 6.61 TB TBW used%: 0.2754%  Power on for: 3.74 months

The WD SN850X is my first foray into nvme drives - attached via a motherboard PCiE slot. As with anything new, I was nervous, but attracted by the removal of two potential points of failure vs a freestanding SATA SSD: the data cable, and the power cable.

My RAID-1 arrays now all have three mirrors - probably overkill but suites my paranoid style

Interesting observation:
- When I join a SATA SSD drive to a RAID-1 array, it writes the full capacity of the array to the drive while syncing.
- When I joined the nvme drive to a RAID-1 array, it only wrote the equivalent of current data capacity - thus a lot less writing.

That appears to mean that using nvme in a RAID array causes less of a hit to the TBW than a SATA SSD. I wonder if this is common to all nvne drives - or just the Western Digital model I have.

[ipic]

It is close to a year and a half since I posed the question for this thread, and took the plunge by replacing my spinning rust drives with SDD alternatives.
Currently, the position is as follows:

System RAID-1 - a three way mirror, with 2x NVME drives, and the original Crucial SDD SATA drive, capacity 4TB:

Code: Select all

CT4000MX500SSD1          Written: 25.26TB   TBW used%: 2.5261%    Power on for: 1.45 years
WD_BLACK SN850X 4000GB   Written: 17.3TB    TBW used%: 0.7208%    Power on for: 10.34 months
WD_BLACK SN850X 4000GB   Written: 10.7TB    TBW used%: 0.4458%    Power on for: 4.58 months

Secondary RAID-1 - where my VM disk images reside (4 active Gentoo VMs) - a three way mirror, with 2x Crucial and 1x Samsung SDD SATA drives, capacity 2TB:

Code: Select all

CT2000MX500SSD1           Written: 4.33TB  TBW used%: 0.6183%     Power on for: 1.02 years
Samsung SSD 870 EVO 2TB   Written: 6.15TB  TBW used%: 0.5126%     Power on for: 11.70 months
CT2000MX500SSD1           Written: 4.49TB  TBW used%: 0.6414%     Power on for: 1.11 years

One of the original 4TB SATA SDD drives was a Fixwot. This starting showing timeouts on some cache clear kernel actions, which produced a noticeable pause in system activity. I thus relegated the Fixwot to be a standalone scratch drive, rather than just throw it away. Since it is not part of the system, any hiccups are limited to software directly using it.

I think the original question I posed has been answered: yes, you can use solid state drives as part of a main system RAID-1 array, based on 1.5 years of continuous usage, if you use branded drives with published data sheets.

There are a few things I discovered that may help others considering this:
- SATA cables and connectors are a significant source of error. In my case both a faulty cable, and a loose motherboard connector. My guess is that the faster speed at which solid state drives respond compared to mechanical drives contributes to this being amplified.
- Off brand drives (e.g.Fixwot) are probably not worth using. They have no published data sheets, and what is actually inside them is anyone's guess. The Fixwot consitintly runs at 10 degrees C higer than all the other SATA drives, even when idle. I think this is a pointer to 'lower quality'
- The two Crucial 2TB drives have a data sheet that claims TBW of 700TB. However, SMART data percentage used shows 1% with the 4.3 and 4.4TB written so far. All other drives with published TBW figures are tracking percentage used correctly - so just the two Crucial 2TB are not. My guess - these two are using the next cheaper chips, and Crucial hoped no one would notice.

All this just in time for SDD drives to become unafordable again - properly ironic