Buying second hand server grade SSD's, should I be concerned?

[cammj]

Hey STH!

I am looking to move away from my Samsung 870 EVO's and in to the world of server grade 2.5" SATA SSDs. I noticed that eBay is full of these second hand but often they have little detail about how they were used/how old they are, so I am a little concerned about how degraded the memory cells are.

When looking at server grade SSDs, such as the Intel DC S3520, should I be concerned about the tbw, or are these -that- well made that it's unlikely they'll die on me in the days/weeks after installation?

 

[BlueFox]

Pretty unlikely they'll die. You can always ask for SMART details, but endurance on them is so high that you'll seldom see anything below 90% life remaining.

 

[cammj]

You can always ask for SMART details

Yeh, the issue is that the sellers for these types of drives seem to be selling so many of them that providing SMART details for one drive appears to be some kind of hassle that they're not wanting to go down for the price.

I can't seem to find very good information on endurance for the drives, because intel's official sites seem to get pulled down now they're no longer sold. However, the limited information I can seem to find seems to put for instance the DC S3520 TBW in a very similar ballpark as the 870 EVO consumer drive, which doesn't make sense and causes concern for me - considering my 870 EVO's are already showing signs of wear and I haven't used them in a datacenter..

 

[mmo]

for DC S3520, found the following document from google search.

https://wfcache.advantech.com/www/certified-peripherals/documents/96fd25-s240-inb2_datasheet.pdf

 

[SnJ9MX]

S36xx and especially S37xx have quite a bit higher endurance. S35xx is lowest endurance.

 

[Stephan]

My usual recommendation for years has been: If you can, get two SSDs but from different manufacturers, same size, use RAID1 or RAIDZ1 (ZFS) for redundancy. This avoids two similarly worn SSDs from giving out at around the same time. Or hitting the same firmware bug in both. Only pick SSDs with power-loss-protection (PLP, all server-grade have it). Only pick models where you can find firmware updates, so you aren't stuck with a 32768-hours-boom-dead bug. Anything beyond 1 PB TBW is plenty, unless you are into Chia then nothing will help you. Try benchmarks with and without write cache set to ON to see which performs best. If possible, stay with MLC.

For firmware availability Intel, Micron, Dell is easy, plain HPE is pretty easy, anything they bought (3Par etc.) is cancer, IBM/Lenovo almost easy (avoid their Javascript and wrappers, unpack any ISO manually), Samsung, Netapp is cancer but some firmware has been "liberated" (try Yandex), Oracle and Cisco is stage 4 cancer, there is a HGST collection for the MLC/SLC stuff.

 

[cammj]

Thanks all this has been super helpful..

Nothing on the drives is super critical to me that I would be super upset if the drive died. I take 4hrly proxmox backups on the drives anyway.

I would mostly be annoyed that I spent $50-$100 only for it to just die on me. It sounds to be that's unlikely with server grade hardware (intel S*) in the first few months of getting it, unless to @Stephan 's point they're doing Chia.. :/

 

[cammj]

for DC S3520, found the following document from google search.

https://wfcache.advantech.com/www/certified-peripherals/documents/96fd25-s240-inb2_datasheet.pdf

Thanks, I saw this - but for some reason I misread the 870's TBW and I mistakenly put the Samsung 870 evo on par with the S3520 - I read it again, and saw that the S3520 has almost double the TBW. Effectively, the S3520 would need to be written once over every day for it's TBW to be exceeded.. that puts me semi at ease..

 

[Stephan]

Intel had at least one bad run with SSDs but I forgot which series. Might have been S3510 S3610 S3710. Make sure to yandex or search here and on level1 and reddit for "intel yourmodel failure" or similar.

 

[Whaaat]

Cisco is stage 4 cancer

I've never ever seen a Cisco drive with hard-locked to 520 bytes firmware or with limited functionality. Firmware packs available online, come in unencrypted .iso and cover hundreds of drive models. One of the home-lab friendliest vendor. Some of my HGST drives with cisco firmware I successfully cross-flashed to a generic one (if available at all), Sandisk with a timebomb was fixed with an update, Samsungs were updated to the actual version from an iso pack.