My belief is that the consumer SSDs actually have very similar performance and reliability requirements to the enterprise because they are generally deployed without redundancy, and so any failures can be catastrophic for consumers. And then there are lots of consumers that will go on Amazon or Newegg etc. and leave very bad reviews.

So I personally believe the 'enterprise' distinction is mostly a scam, and developers who are looking for inexpensive hosting have been mainly waiting for A) datacenter providers to realize that or B) a slow market perfusion that mitigates it.

I think the market is going through a shift as more redundancy is used at the server level and the SSDs are considered more as a commodity.

Define perform. But short answer, lots.

To be fair, consumer SSDs these days are pretty solid, and will perform great for most workloads - even many workloads which enterprise drives would be used for.

The big difference though, is with extremely heavy write workloads. The endurance of enterprise SSDs in terms of how much data you can write before the drive will fail, is a couple orders of magnitude greater than consumer drives. For example, I have an 800gb Intel S3710 SSD. It's rated to handle 10 DWPD (drive writes per day) - meaning that one could write 8,000 gigabytes of data a day, 365 days a year, for 5 years (~17 petabytes), and the drive will still be chugging away just fine.

A 1TB Samsung 850 Pro (which has some of best endurance of consumer ssds), on the other hand is only rated to 300 terabytes of data written. So the enterprise drive has a 60x improvement in endurance.

It actually means something more like "we don't feel confident that the drive will last more than 5 years at 8TB/day." --- and if you do the maths, assuming perfectly even wear, this means they're rating the flash at ~20K cycles/cell, which sounds a bit ambitious for 20nm MLC considering that ~10K cycles was the norm for 65nm MLC several years ago.

1TB Samsung 850 Pro (which has some of best endurance of consumer ssds), on the other hand is only rated to 300 terabytes of data written.

That totals around 300 cycles/cell, which seems like 3-bit/cell ("TLC") flash to me, or cheap small-process MLC.

What is the real average durability requirement?

Have there not been some relatively large providers using consumer SSDs in the datacenter successfully for some time?

They reflect what the manufacturers will offer the warranty until. A quick google search suggests about 2 petabytes of writes is the max you can get on a samsung 850 before it completely fails (and after 1 petabyte you will have many failed sectors and data corruption starts creeping in)

At the same time, I bought my S3710 used and when I got it, it had almost 8 petabytes written and it's still just like new.

> Have there not been some relatively large providers using consumer SSDs in the datacenter successfully for some time?

Yes, that wouldn't surprise me at all. Like I said, for many workloads, consumer SSDs are acceptable. But that doesn't mean enterprise drives are a scam.

From the technology side, there is at least the amount of extra flash that the drive contains and that it can use for better performance and longer life times (TBW rating).

Our anecdotal experience is that there is quite a noticable difference between drives: - 99.5 percentile latency during normal operations. Some enterprise drives guarantee response times, others are all over the place. Note that consumer drives often claim a much higher maximum IOPS rating whereas you can get enterprise drives with a sustained rating. - drives hanging for >60s, presumably during internal cleanup. - firmware stability, drive getting unresponsive when maxing it out over longer periods.

> How many studies have been done verifying ...

With due respect, what verifies the allegations in the comment?

But they don't want to assume all devices will behave that way. Better to specify the server scenario at realistic server temperatures.

So I wonder if there's a tradeoff between longer data retention when powered off (by using higher write and erase voltages) and longer MTBF (by using lower voltages). I could be completely wrong here, but if there is such a tradeoff it would make sense for consumer and enterprise drives to make different choices.

The same mechanism that makes SSD's not retain data (that is, the electrons escaping) and the same thing which makes them not retain the data that you do want to keep would presumably also make the electron buildup that you get that kills the cells abilities to continue their write/erase cycle.