Comments on: Using NVMe Command Line Tools to Check NVMe Flash Health

By: anp

anp — Thu, 02 Jan 2020 08:42:13 +0000

Where can I find documents related to discovery of NVMe drives connected to a Serial Attached SCSI controller.?

By: AnjanaDhanvi

AnjanaDhanvi — Mon, 05 Aug 2019 09:03:03 +0000

Thanks for this post.I need more details for Thermal Throttle Status?

By: AnjanaDhanvi

AnjanaDhanvi — Fri, 26 Apr 2019 10:18:18 +0000

Thanks for this post.I want more details for Checking NVMe Flash Health

By: Samthedev

Samthedev — Mon, 07 Aug 2017 15:48:18 +0000

Hi Peter,

Thank you for the great article, I have question about the write amplification pls.

My Intel nvme reports a bigger value on the host_bytes_written attribute compared to the nand_bytes_written which doesn’t make sense, even when i try to write the disk, the host_bytes_written attribute seems to increase more than the nand_bytes_written! do you have any explanation to that ? given the fact that the nand_bytes_written should be bigger since it includes also the data written by the host and the overhead of the garbage collector/Wear leveler !

By: Peter Zaitsev

Peter Zaitsev — Thu, 16 Feb 2017 19:39:12 +0000

Hi,

Oh yes. I would not count on SSDs to become read only after they wear out. I think the idea is when you’re operating SSD beyond its operational parameters you’re using it on your own risk.

By: Jörg Brühe

Jörg Brühe — Thu, 16 Feb 2017 19:30:06 +0000

Hi Peter!
Sorry about the late reply …
The article I refer to appeared in the 1/2017 issue of c’t, a German IT paper. It seems the full text is not available online, just the start; this is the URL:
https://www.heise.de/ct/ausgabe/2017-1-Flash-Speicher-im-Langzeittest-3573503.html
To get the full article, you have to buy it (online, from that link), and it is German only.

You are right about the SSDs operating much longer than specified by the manufacturer, c’t also reported that. So my “disappointing” does not refer to the overall quality, it refers to the lack of clear diagnostics that would give a meaningful warning sign to the admin. I’m sorry my comment didn’t make that obvious, my fault.

I followed your link, that text sounds more positive on the helpfulness of SMART data than c’t is. I suspect that is an area where manufacturers differ (or: where quality differs). Or might it have changed from that 2015 report to the c’t 2017 one?
c’t had tried these SSDs of 240 – 256 GB: Crucial BX 200, Samsung 750 Evo and 850 Pro, SanDisk Extreme Pro, SanDisk Ultra II, and Toshiba OCZ TR. It might be significant that in both reports, the Samsung Pro could stand the largest amount of data written.

Also, both reports agree that SSDs fail completely, do not fall back into a read-only mode – and IMO, that is a really sad fact.
Regards,
Jörg

By: Peter Zaitsev

Peter Zaitsev — Fri, 10 Feb 2017 15:41:31 +0000

In reply to Jörg Brühe. Hi Jorg, What article are you referring too ? The article I read on this found what SSDs tend to operate well beyond their file time which SSD would operate http://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead

By: Jörg Brühe

Jörg Brühe — Fri, 10 Feb 2017 15:21:35 +0000

Hi Peter!
Sadly, you cannot rely on the values reported by smartctl giving any good indication of the SSD’s wearout.
German magazine c’t had an article in its issue 1/2017 where they had done a long-term test of SSDs, continuously writing data to them and checking the smartctl values – the results were very disappointing.
And even worse, when a SSD was worn out, it would not only fail to write anything but also deny read access.
All this was about SATA SSDs, but I wouldn’t be surprised if it were no better for NVMe devices.
So: Beware!
Regards,
Jörg