When HDD dies - SSD?

[Yorick]

Looking at https://blocksandfiles.com/2023/05/09/pure-no-more-hard-drives-2028/?td=rt-3a, while I do not expect HDDs to go away anytime soon, I am wondering whether it makes sense for my personal use case to consider replacing failing HDD with SSD.

I have an 8x8TB raidz2 with just about no writes and very light read use: Backups and media files. The drives are going on 5 years old and should last me a while longer, though expecting one or more of them to fail in the next 5 years is reasonable.

For SSDs, “cheap crap” could be sufficient. As in, not bad quality, but not data center either: Consumer QLC. Sure it drops the TBW, but I hardly write - and it craters write IOPS but, oh right, I hardly write.

Prices are still a little mad: 460 bucks for a sata drive, double that for NVMe. Even in the budget space. And, that’ll come down.

Whether sufficiently to make it reasonable to replace a failing hdd with ssd - that’ll be interesting to see.

 

[sretalla]

Sounds like a reasonable plan if your budget will stretch to it.

I guess the later the HDDs fail, the better in that regard.

 

[Apollo]

Looking at https://blocksandfiles.com/2023/05/09/pure-no-more-hard-drives-2028/?td=rt-3a, while I do not expect HDDs to go away anytime soon, I am wondering whether it makes sense for my personal use case to consider replacing failing HDD with SSD.

I have an 8x8TB raidz2 with just about no writes and very light read use: Backups and media files. The drives are going on 5 years old and should last me a while longer, though expecting one or more of them to fail in the next 5 years is reasonable.

For SSDs, “cheap crap” could be sufficient. As in, not bad quality, but not data center either: Consumer QLC. Sure it drops the TBW, but I hardly write - and it craters write IOPS but, oh right, I hardly write.

Prices are still a little mad: 460 bucks for a sata drive, double that for NVMe. Even in the budget space. And, that’ll come down.

Whether sufficiently to make it reasonable to replace a failing hdd with ssd - that’ll be interesting to see.

Don't forget snapshots. They will cause writes to your pool when they are created and destroyed.

 

[Patrick M. Hausen]

Don't forget snapshots. They will cause writes to your pool when they are created and destroyed.

Not really. Snapshots are not copies in ZFS.

 

[sretalla]

Well, snapshots are writes, but they are merely pennies on the dollar in terms of size (or not even that in most cases). It's only writing a list of pointers to the consumed blocks, not the blocks themselves.

 

[Apollo]

Another thing I haven't found clearly advertised is the "Physical block size" of SSD drives. Some seems to support 4k the others I don't know.

 

[sretalla]

Almost all would be 4K native.

Some older ones perhaps 512K, but shouldn't really be the ones you would find for sale now as new.

I note that older USB drives were often 512K (and I still have a relic there on a boot pool that I'm too lazy to reinstall after replacing USB with SSD boot).

 

[Patrick M. Hausen]

PCIe 3 NVMe SSDs reaching 40€ per TB:

SSDs landen bei 40 Euro für 1 TByte

Die Speicherpreise purzeln munter weiter und auch in den nächsten Monaten ist noch kein Ende in Sicht. NAND-Herstellern drohen dadurch Pleiten.

www.heise.de

 

[WI_Hedgehog]

Electricity costs should factor into the Total Cost of Ownership. Another thread was discussing spin-down costs (spin-down/up affects lifespan) vs idle cost; SSD drives idle efficiency as they don't spin.

A quality UPS with power filter should be used, though I say that about any system (based on experience).

 

[cobrakiller58]

I have a pool of 2TB drives that I am planning to do the same with, I installed my last spare constellation drive and all following replacements will be with the cheapest available 4TB SSD from a respectable brand. My pool has no snapshots all data is stored in my main pool or off-site and is primarily random access of small files.

 

[abufrejoval]

Looking at https://blocksandfiles.com/2023/05/09/pure-no-more-hard-drives-2028/?td=rt-3a, while I do not expect HDDs to go away anytime soon, I am wondering whether it makes sense for my personal use case to consider replacing failing HDD with SSD.

I have an 8x8TB raidz2 with just about no writes and very light read use: Backups and media files. The drives are going on 5 years old and should last me a while longer, though expecting one or more of them to fail in the next 5 years is reasonable.

For SSDs, “cheap crap” could be sufficient. As in, not bad quality, but not data center either: Consumer QLC. Sure it drops the TBW, but I hardly write - and it craters write IOPS but, oh right, I hardly write.

Prices are still a little mad: 460 bucks for a sata drive, double that for NVMe. Even in the budget space. And, that’ll come down.

Whether sufficiently to make it reasonable to replace a failing hdd with ssd - that’ll be interesting to see.

QLC SSDs probably like the write amplification of raidz even less than shingled HDDs. Since ZFS is supposed to be lazy on writes perhaps that effect isn't as bad as it could be, but when it comes to faith in storage, HDDs with me still have a bit of an edge for long term stuff: they've never gone bad on me from lying on a shelf.

SSDs did. They actually require power to do maintain those trapped electrons and aren't designed to hold data without it.

I consider cheap crap SSDs even in a RAID0 good enough for things like Steam caches, but I wouldn't trust them to hold data that I'd miss.

So the first thing is to ensure you still have backup, preferably two. And I make those single big HDDs in a caddy-less hot-swap bay for easy handling.

Then you really need to be monitoring your write stats, without a RAID controller in the middle they should at least be easy to look at. If you see drive life dropping fast, the write amplification may have been the issue.

But then I've even seen HDDs these days no longer certified to take the normal patrol read loads that RAID controllers have as a default.

At 8x8TB I don't see SSDs attractive yet. Both SATA and NVMe are still prohibitive beyond 4TB and I'm not sure QLC will ever reach SATA at these capacities, as nobody is investing in new SATA SSDs anyway.

My main storage is an 6x6TB RAID6 on a trusty LSI 9261i and I keep thinking similar thoughts, but so far nothing is quite close enough in speed and economy, while that machine also runs a 4x1TB SATA SSD RAID0 cache using the onboard ports.

 

[cobrakiller58]

they've never gone bad on my from lying in a shelf

This is a different use case than what I believe is being planned here, if for some reason I was putting my server into a storage unit for whatever reason I would be concerned about this as well.

 

[WI_Hedgehog]

https://www.extremetech.com/computing/205382-ssds-can-lose-data-in-as-little-as-7-days-without-power

 

[cobrakiller58]

That article is from 2015 and I'm sure it still applies to no name SSD's stored at high temperatures. Do you happen to know of an article in the last 3 years testing WD, Samsung, Seagate or other respectable brands?

 

[WI_Hedgehog]

That article is from 2015 and I'm sure it still applies to no name SSD's stored at high temperatures. Do you happen to know of an article in the last 3 years testing WD, Samsung, Seagate or other respectable brands?

It depends on the drive. The Enterprise Solid-State Drives I use have a Product Specification (for that exact drive) stating:

Data Retention*: 3 months power-off retention once SSD reaches rated write endurance at 40 °C
*The time period for retaining data in the NAND at maximum rated endurance.

 

[Apollo]

It depends on the drive. The Enterprise Solid-State Drives I use have a Product Specification (for that exact drive) stating:

Data Retention*: 3 months power-off retention once SSD reaches rated write endurance at 40 °C
*The time period for retaining data in the NAND at maximum rated endurance.

If I read the JEDEC standards correctly, the retention time must be met after the SSD has reached its endurance rating limit.

 

[abufrejoval]

I guess my main point is that SSDs and HDDs do not just differ in €/TB and IOPS, it's important to understand their strenghts and weaknesses with regards to write amplification, long term data retention and how they fail and why.

My backup RAID is composed of 2TB drives which are way over ten years old, but show no signs of failing, perhaps because after some years of online use, they went to cold standby. Mostly I keep them because it makes no sense to throw them away, but I complement them with alternating backups on removable 18TB Helium drives on the same box. As an 8 spindle RAID5 the 2TBs do offer quite a bit more bandwidth and a better match for the 10GBit network than a single Helium spindle, too.

If those ancient 2TB drives should fail, I expect them to do that individually and I still have plenty of spares and a RAID5 to cover that.

Should a Helium drive fail, I still have the other, but the backup will be older.

I had replaced the primary 8x2TB 3.5" RAID5 with a 6x4TB RAID6 using 2.5" HDDs for better power effiency, but then got hit by all 2.5" HDDs using shingled media. Seagate replaced those free-of-charge for 3.5" 6TB CMR drives when they failed on rebuilds and I was glad for the 10Gbit network then. It's a real pity they don't sell the 5TB SMR drive as a 4TB CMR option!

Even after all these years I can't see SSDs quite closing in on HDDs in terms of €/TB, Pure will proclaim differently but their tech doesn't work at home-lab capacity points. And while the natural progression from online RAID drives to offline RAID drives no longer really works for me at current helium capacity points (my storage doesn't just keep exploding), the transition from RAID speeds to mirrored or alternating single spindle drives isn't quite what I had in mind when I put a 10Gbit network in as my home-lab backbone.

 

[joeschmuck]

At this time I do not see replacing 8TB HDD's with 8TB SSD's as a reasonable option, unless you have big money to just throw at it and you have some other requirement such as physical space, noise, or power draw, or heat, that would push you into this path. Total cost of ownership might be a stretch as sort of good quality SSD's are going to be over 5 times the cost, good quality will be even more. Maybe in 5 years the prices will drop but will it drop that much? Then if you are spending this kind of money you need to ask yourself if the cost or the NAS is worth it, since you rarely write to it. Maybe you read from it a lot.

I personally would love to build a fully SSD/NVMe NAS, but my wife would wonder why the credit card bill jumped up several thousand dollars. Then she would kill me when I told her it was to backup all the computer files, photos, and music. She knows all the important data would fit on a 1TB Flash drive with room to spare.

If you were to replace a HDD with a SSD as the drives fail, while it would still be very expensive, spreading out the cost is a good thing. Not sure I'd buy QVO drives. I'd also look at the warranty, you want a long one. 5 years is not long enough for me when spending this kind of money for a home project. A 10 year warranty would work for me.

Another option, if you generally read from the NAS a lot, and write very little, then why not change things up and create a SSD RAIDZ1 or just a Stripe, then routinely backup say weekly to an external drive (could be on any other computer in your home or even the cloud) so you have a copy safe somewhere. if you have important data then back it up to two different locations. The goal here, minimum the number of SSD's you need. I could easily get away with a pair of mirrored 4TB SSD's for my real wants. But my needs would drop to a pair of 2TB SSD's mirrored.

So you may need to reevaluate your pool configuration is all I'm saying if you go SSD all the way.

 

[WI_Hedgehog]

At this time I do not see replacing 8TB HDD's with 8TB SSD's as a reasonable option, unless you have big money to just throw at it and you have some other requirement such as physical space, noise, or power draw, or heat, that would push you into this path. ....

Looking at https://blocksandfiles.com/2023/05/09/pure-no-more-hard-drives-2028/?td=rt-3a, ....

I remember when 2.5" HDDs were supposed to replace 3.5" HDDs in datacenters due to lower power/heat. Somehow that never happened and 3.5" drives generally rule the large-storage world. Granted, 2.5" drives are used for smaller parallel installations, but 3.5" drives seem to have regained their standing in cost vs. performance. That really speaks for @joeschmuck's point of value and TCO.