A Study of SSD Reliability in Large Scale Enterprise Storage Deployments | Hacker News Reader

A Study of SSD Reliability in Large Scale Enterprise Storage Deployments

usenix.org

29 pointszdw7 comments

I didn't read the paper, just the slides. The study, which looked at enterprise drives, found them to be quite reliable:

§ The average ARR across the entire population is 0.22%, but rates vary widely (0.07 - 1.2%)!

ARR: annual replacement rate.

My anecdotal experience with SSD drives is that they are quite prone to catastrophic failures. But that is for consumer quality, not enterprise.

Predictable catastrophic failures or other?

There are stats you can get from the SSD which will indicate the end of life.

stats? SSD most often simply vanish from the system without any warnings, no promised RO mode, no errors, drop of IO. Almost zero problem in enterprise, hard freeze on personal OSes and immediate loss of data.

I worked as a technician for a while recently.

> There are stats you can get from the SSD which will indicate the end of life.

I've seen too many consumer SSDs which failed completely without warning from one day to the next with no chance of data recovery. At least if an HDD starts to fail, there are signs that you can act on (files not opening/saving, Windows behaving weird) and you usually have plenty of time before it actually happens that you can start data recovery and get a decent amount (or even most) of data out of it. And even once it fails, data recovery is possible. SSDs are just dead black boxes. It's why when I upgraded a customer to an SSD, I always put a big emphasis on making backups onto an external drive or a cloud service or, well, anything.

Out of all the failed SSDs I've seen, only a single one managed to put itself into an RO mode for data to be recovered. And they often give the weirdest stats. Like, some of them will report everything is okay even though you're getting garbage or nothing out of it and can't write to it.

When they work, they're great, but the failure states are just obscene.

I've worked with plenty of HDDs and SSDs (software engineer for enterprise storage products, working on the hw-sw integration), HDDs and SSDs have pretty rapid failures. Sometimes you get several days or weeks of low-frequency errors or high latencies but more often than not you'll get an almost one-instant failure.

Most of the HDDs I worked with were 7200RPM enterprise drives (SATA & SAS) and the SSDs were enterprise quality too (SAS & NVMe).

we from croit.io use and suggest SSDs on a daily basis in our Ceph based Software Defined Storage clusters. So far, SSD is much better than HDD due to good indicators like wear level. Even if a drive fails, any enterprise class storage should be able to heal itself.

I personally don't know a person that would go back to HDD after having SSD in operation. Far superior, much more productivity,..

Anyone got a guess which manufacturer is which?

Reply on news.ycombinator.com