Hacker News Reader

Comment by guardiangod | Hacker News Reader

At work, I've been using hundreds of the top of the line SSD drives from a certain S company. The drives are used 100% from the start in 1U servers. The failure rate for them is higher than HD imo. The more worrying thing is that none of the drivess' SMART hinted their imminent failure. The drive would just die one morning, taking all the data with it.

Thank god for their 10 years warranty. I guess people like me is the reason why the S company lowered the warranty to 5 years for their next generation product.

View on HN

>At work, I've been using hundreds of the top of the line SSD drives from a certain S company

that really doesn't help narrowing it down. we have:

1. samsung

2. seagate

3. sandisk

4. sabrent

5. sk hynix

6. silicon power

I interpreted it as Samsung, and I think that's consistent with the statement about 10 and 5 year warranties.

taking all the data with it

Man, I thought the failure mode of SSDs was (mostly) "can't write anymore, but can at least read, so backup NOW and get a new one". Or am I overanalyzing parent's statement, and taking it too literally?

I don't have a lot of experience with SSDs, but my impression is that their failure mode tends to be abrupt complete failure.

The "soft" failure mode is based on the assumption that the flash chip wear-out is the failure mode and thus the worn out pages will be (slowly) accumulating and being (gracefully) replaced with non-worn-out pages. If that happens, the failure would indeed be gradual and result in the SSD going into a read-only mode.

My (limited) experience is that is not the dominant failure mode. It would be really useful if BackBlaze could weigh in on their failure mode experiences.

----

It is like LED lights where the manufacturer advertises essentially infinite LED life and they are not lying; the LEDs last a very long time. The problem is, the dominant failure mode is the electronics that power the LEDs and have a MUCH shorter life that the LEDs themselves. <cough> cheap caps </cough>

This is marketing and wishful thinking, most times when SSD dies critical system nand gets corrupted. Intel was boasting about safe read only failover, and then when it turned out their shit drives just die like the rest they invented "we shutdown read access to protect our clients" :---)

https://techreport.com/review/26523/the-ssd-endurance-experi...

"After a reboot, the SSD disappeared completely from the Intel software. It was still detected by the storage driver, but only as an inaccessible, 0GB SATA device.

According to Intel, this end-of-life behavior generally matches what’s supposed to happen. The write errors suggest the 335 Series had entered read-only mode. When the power is cycled in this state, a sort of self-destruct mechanism is triggered, rendering the drive unresponsive. "

Inmy experienc SSDs fail hard and without much warning though they are often designed to try and go into a read only mode to prevent further writes. That being said, all the SSDs I've had fail on me were nearly new like just imaged/deployed, had power issues that maybe killed the controller vs the storage itself, or just died. One drive I was using for VMs at home just stopped running and when connected prevented the entire PC from posting.

That being said, I'll happily take the speed increase of solid state over spinning disks given I'm not storing media, at least for consumer sized drives. 5 years is much more realistic for general tech lifespan before a newer/faster/cheaper/more efficient upgrade in the enterprise space and definitely the consume space.

We're sticking with Intel at work (now SK hynix for capacity I guess) and we've had great success, but we replace drives quite early in their life, we don't even let them wear past 90% reallocs still available. We've had a couple Samsungs, albeit a bit cheaper they've failed horribly many times.

> The drives are used 100% from the start

Looks horrible usage for SSD wear leveling and other functionality.

If you’re using them in 1U servers, then those machines should be data less, and treated like cattle, not pets.

Keep the data on one or more file servers on the network, which has/have lots of built-in redundancy and other backup and DR facilities.

Reply on news.ycombinator.com