Live data from Hacker News

SSD Write Endurance Considered Sufficient

ef.gy

21–30 of 34 posts

Re: SSD Write Endurance Considered Sufficient

#22
post #8
post #4

This is a pretty poor article. Consumers haven't bought SLC in years and MLC is much less reliable (although the actual reliability of flash is still higher than the rating). It doesn't take write amplification or any kind of real workload into account. And it may contribute to the alarmism that it claims to dispel; at worst your SSD will last 172 days? Yikes!

Not like magnetic is necessarily immune. WD greens used to have astoundingly bad spin down logic. I killed a bunch of them with default Linux settings and near idle servers in about that time frame.

I have a Scorpio Blue that parked its heads incessantly until I deployed a utility to force better ACPI settings.

Not write endurance, per se. But you only get to park the heads so many times, on average...

Re: SSD Write Endurance Considered Sufficient

#23
I remember when SSD's first started hitting the market years ago, everyone's utmost concern was about wear leveling.

As it turns out, if you ship a drive with buggy firmware and it refuses to be recognized after the 50th cold boot, write leveling ends up not being so important.

Re: SSD Write Endurance Considered Sufficient

#24
SMART says the years-old, low-capacity MLC SSD in this computer is 5% worn out. SSDs in RAID arrays at work are getting a much heavier workout and handling it fine.

You can use smartctl or the equivalent to check the media wearout indicator to see how much the actual workload has worn out a drive. If your computer isn't constantly writing, you're probably fine--that is, your writes will last until you'd be upgrading anyway.

But, keep backups. Besides the many practical problems you can have (laptop lost! fire!), SSDs have firmware bugs every so often.

Re: SSD Write Endurance Considered Sufficient

#26
post #8
post #4

This is a pretty poor article. Consumers haven't bought SLC in years and MLC is much less reliable (although the actual reliability of flash is still higher than the rating). It doesn't take write amplification or any kind of real workload into account. And it may contribute to the alarmism that it claims to dispel; at worst your SSD will last 172 days? Yikes!

Not like magnetic is necessarily immune. WD greens used to have astoundingly bad spin down logic. I killed a bunch of them with default Linux settings and near idle servers in about that time frame.

Greens are not enterprise drives and shouldn't be used in RAID arrays.

Their logic makes perfect sense for single disk and "green" (power saving.) Its just incredible what people will do to save a little cash on servers. Your company's data depends on them and you put in the cheapest Best Buy inventory you can find?

Re: SSD Write Endurance Considered Sufficient

#27
post #6

tl;dr I'm guilty of the SSD fear. My first Intel (320) drive was meticulously maintained in fear of shortened life span and performance losses. I went so far as to question every file copied to the drive vs. mounting an external HDD. The thing bricked itself due to a firmware bug in less than 12 months. Using a Samsung drive now. I don't think about read/writes anymore. I just use the darn thing.

The 320 had a brickable fw bug? I have 5-10 or so in production and have no issues. Maybe mine have a newer FW. I always update them when I give them out.

Re: SSD Write Endurance Considered Sufficient

#28
post #8

Earlier quoted context omitted.

Not like magnetic is necessarily immune. WD greens used to have astoundingly bad spin down logic. I killed a bunch of them with default Linux settings and near idle servers in about that time frame.

Greens are not enterprise drives and shouldn't be used in RAID arrays. Their logic makes perfect sense for single disk and "green" (power saving.) Its just incredible what people will do to save a little cash on servers. Your company's data depends on them and you put in the cheapest Best Buy inventory you can find?

AFAIK the RE-GP series have the same issue and are advertised as server/enterprise grade.

Re: SSD Write Endurance Considered Sufficient

#30
post #8

Earlier quoted context omitted.

Not like magnetic is necessarily immune. WD greens used to have astoundingly bad spin down logic. I killed a bunch of them with default Linux settings and near idle servers in about that time frame.

Greens are not enterprise drives and shouldn't be used in RAID arrays. Their logic makes perfect sense for single disk and "green" (power saving.) Its just incredible what people will do to save a little cash on servers. Your company's data depends on them and you put in the cheapest Best Buy inventory you can find?

I'm not sure who you are replying to. Your fictitious account does not match my reality.

A drive that is accessed a couple times a day seems like the perfect use case for a green. Sadly, it head parks after 8 seconds of idle time, and Linux was doing one IO every 30 seconds, I believe as part of the S.M.A.R.T. monitoring (its been a while), the drive is rated for 300000 head parks in its lifetime, so in 105 days the drive has thrashed away its specified lifetime because you were monitoring the drives' health, and head parks wasn't one of the parameters tracked, so it didn't even help to monitor!

This is not because the Green is not Black. It is not the grade of bearing, or the balance tolerance of the platters. This is because WD wrote ridiculously silly firmware for their drives and they self destructed on one of the most common computing platforms.

Back to the article topic: The similar workload machines I deployed 14 months ago have an SSD for their root drive and newer greens (tweaked to not load cycle) for their bulk. None of the SSDs have even reach 1% of their media lifetime count in their S.M.A.R.T. data. I'm not worried about wearing out SSDs.

Post reply on HN