Consumer SSD's in production? Value and managing risk

Background:
Long story short my (I’m not the owner, but make the IT calls) small business of about 30 has a small Proxmox cluster of 5 servers. Three R620’s and two R630’s VM’s were some windows desktops for product testing and support, and some linux VM’s for internal DB, Web, misc. It’s all working pretty well. I went down the road of trying to use a NAS for remote storage, but the latency was a killer. All server storage is local now.

Current:
With covid we went remote and what was an office is now a pile of desktops in a closet that people remote into. We started hiring remote so boss says we are staying remote and he wants me to virtualizes the remaining desktops. So I need more servers. When I purchased my existing serves I did not pay much attention to the SSD’s in them and all of them have Crutial BX500, which turns out are some pretty old, crappy SSD’s.

Plan:
So I am going to buy 2 more r630’s but put in brand new Pro-consumer grade SSD’s and replace all the Crutial ssd’s in the other R630’s. I know in the enterprise world consumer grade SSD’ in enterprise gear is a no no, but is there a happy medium? IS there a high enough quality SSD out thre that mitigates much of the risk (not all)? Or more to the point I need the storage to NOT cost 3 times the server.

The R630s I have in mind will be low core high clock v4 zeons for windows responsiveness. And 256gb ram, and about 6-8tb of SSd storage.

Also part of my plans is for the new servers to use ZFS so I can have proxmox replicate between pairs.

Also would it be insane to put in some consumer grade NVME pcie adapters and go consumer NVME? And maybe some spinning rust for local backup?

How bad of a plan is all this? Wow I typed a lot for a simple question… sorry for the book.

May I suggest passthrough SSD for each box then backup the computer to some rust nightly. Would allow you to get cheap SSDs and when they die just dump the backup image on the new drive. Might be a good option.

Does the user workload merit NVME drives?

Yes, the perc h730 allows passthrough, that’s my plan at the moment. The Windows clients don’t necessarily need the NVME speed, but I have experienced some pretty bad IO bottlenecks when you do something like move or restore a VM on a server with windows VM’s.

How many bays do you have maybe mirror the client sata SSDs? If you down time is that big of an issue

I know that there are consumer grade SSDs with >2.5PB TBW. Those should be able to sustain some load to them for quite some time. Whatever strategy you choose, please make sure you have backups and spare SSDs to quickly swap them out if one fails.

Example: Transcend MTE220S SSD 2TB, 4.4PB TBW, 200EUR

If you have 4 of those Transcends and you distribute the writes you have a total of 17.6PB TBW. 17.6PB = 18022.4TB = 18454937.6 GB. 18454937.6GB / 36 month / 30 days / 30 people = 569 GB. Meaning in an ideal scenario every employee could write over 500GB of data a day and you would still be withing spec and most important within warranty.

I was under the impression that enterprise class SSDs had not only better TBW endurance, but they also had better hardware level protections for data recovery, power loss scenarios, etc…

I’d weight that all against how critical you consider up-time and data retention vs a critical worse case scenario.

Yea I am a backup fanatic. These will be in a zraid with spares, so the worse case is I have multiple failures and I have to recover the VM’s from backup. And these are just desktops so no real data loss, just time loss. I will look into those higher TBW drives, that may give me more margin.

Since you mentioned using these drives for VMs replacing desktops, I would guess the drives I suggested work fine. Most people read and write only a couple of gigabytes a day. But you have to evaluate your use case.

Consumer junk with redundancy has worked out for BackBlaze.

They work around expecting failures at a higher rate that enterprise.

They are the reason I still like Seagate. Worth the cheaper price, as the reliability is not as bad as the discount for them, In Rust.

In SSD, both WD and Sea are well over priced for the meagre offerings they present.

1 Like

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.