Hacker News Reader

Comment by gnosis | Hacker News Reader

How does the longevity of SSDs compare to HDs these days?

Usually, an SSD will by far outlast an HDD. The risk of drive failiures is always quite large with mechanical drives. This of course depends on useage, but with an estimated workload of 10 GiB/day a 256GiB will last for about 70 years[1]. (These numbers are for MLC NAND)

[1] : http://www.anandtech.com/show/6459/samsung-ssd-840-testing-t...

Also, most of the time, when SSD's fail, they become read-only, which is a far better outcome than a spinner starting to scratch its plates. So even if SSD's had short lifespans it wouldn't be as destructive.

most of the time

All the crashes I've read about on the web were fatal failures with total data loss. I hope it'll get better.

Well, here's one story on the web for you about SSD failure without total data loss: I've had an SSD that just refused to write anymore, but all the data was still readable. All I had to do was buy another SSD, dd the old SSD's data over to it, and go on my merry way.

Same thing happened to me, and at the time I had no idea that that was how SSDs failed - it was my OS disk, and suddenly I couldn't sudo.

Popped in a bootable USB, and had the data copied off and a replacement SSD in within a couple of hours.

I wouldn't depend on that failure mode, but it's a nice feature.

My first, only, and last-until-hell-freezes-over SSD went crazy last week, after a series of bizarre kernel panics that now appear to relate entirely to the SSD (machine is fine after a reinstall). Following a final reboot the Macbook's firmware displayed a "missing startup folder" icon. On flying home I discover that several GB from the start of the drive has gone missing entirely (check out http://i.imgur.com/riK4CBt.gif -- grey is good, black is zero blocks). Never again relying on a disk technology that doesn't respond to a hammer when it's dying.

One data point is a poor basis for decision making.

Right. So I could drop $2-3k on gaining experience from a meaningful sample size, all the while risking more lost business days, or I could just believe all those anecdotes it took me close to half a decade to convince myself to ignore. I've got bigger problems than seek latency, will retry in a couple more years.

Maybe you could try making backups next time?

Would certainly be smarter than banking on the reliability of a single HDD.

Unfortunately you can't back up billable hours, and carrying an entirely separate laptop or disk on the expectation that your main one will fail seems a bit too weird, especially when robust alternatives exist

As an hourly contractor as well, I back up everything to the cloud and have at least two laptops in a 'ready' state. Trusting hardware to work is... brave.

meh. That's what backups are for? Have you ever restored from time machine? In the 5 years I've lost 3 or 4 mechanical drives in my MBP's and only lost the amount of time it took to buy a new HD and restore.

I'm much happier with the day-to-day SSD speeds and I figure if the drive lasts a year or so I'm still on par with my usual HD burn out rate.

If you have a laptop with a optical disk drive, you can remove the drive and replace it with a second hard drive for backups alone. You can even make it a 1.5TB HDD and use something like Time Machine for multiple version backups that backup all the time. That way you don't have to carry anything.

Most of your billable hours would be saved that way. I also have an external HDD connected to my laptop dock that does automatic backups whenever I connect the laptop to it, and I have crashplan for 'offsite' backups. I don't use my second HDD in my laptop as a backup hard drive although.

Isn't total data loss the default for a mechanical drive failure as well?

I don't think this should be counted as a major weakness of SSDs.

First, no. HDD recovery is expensive but possible in many cases.

Secondly, no one say it should be counted as a weakness, the guy above was just pointing out that it's probably not a strength, as someone above claimed.

Expensive SSD recovery is also an option.

Really?

I've had 4 SSD drives fail across 20 desktops and in all cases the data loss has been total. I've never seen, heck or even heard of, an SSD drive go int read only mode.

Is there some special tinkering you need to do to get one to go into read only mode?

they typically eat your data by corrupting their internal state. I hear about this kind of ssd crashes way more often than conventional HDD crashes.

flash wear is a red herring and is dwarfed by these FTL corruptions.

One thing you should care with consumer SSDs - is always to have a backup battery for power loss. Due to large write cache it's easy to lose a large chunk of file system.

Enterprise SSDs usually have supercapacitors to take care of cache flushing.

http://www.xtremesystems.org/forums/showthread.php?278816-SS...

What power and time are we talking about? Could one high draw cell or AAA battery do it?

The main problem is the failure mode (complete). Either you get a bad one or, like most, you don't... but if it breaks, it's completely gone.

I'd like to know how many have experienced a failure after 90 days or whatever the cut-off is so that I can stop worrying so much beyond a certain point.

My personal failure rate with SSD drives has been close to 50%. I like them, I use them, but I don't trust them!

50% sounds pretty bad. How long did they last? Were you able to recover your data from them?

Must be an OCZ user ;)

I had an Intel 32G X25-V value drive that failed and lost all its data. And then a Corsair drive that failed and I sent it back and they sent a replacement.

I have a bunch of OCZ drives but so far I don't think any of them have failed. Just bought a 'refurb' SATA 2 OCZ for $62/120G.

Never lost any data because I used the SSDs as boot drives, not for data storage.

Details? I've only had one fail about 4 years ago but that was almost straight out of the box so it was bad from the factory.

If you're not doing anything crazy with them, you probably won't be able to kill your average MLC/TLC SSD.

Let's say you write 20 GB to your SSD every day. The SSDs mentioned in the article have a lifetime of 72 TB, which means you can write 20 GB daily for more than 10 years. Also, 72 TB is probably a pretty pessimistic estimate.

That being said, if you do anything I/O-intensive (write-heavy databases, anything with lots of small write operations, ...), YMMV. That's why "enterprise" SSDs are still unbelievably expensive - quite often they use SLC technology (though there are more and more server MLCs) and they usually have way more "spare" cells which are used to replace other cells killed by too many writes.

Enterprise quality seems to mean a different quality improvement each time. Is there any realistic point in average heavy desktop users sticking to SLC?

Can't think of a reason. SLC is a bit faster at writes, but a lot of new MLC SSDs are able to fully saturate SATA 6Gb/s, so there's really no point in going any faster.

If you're really worried about write endurance, stick to MLCs with a lot of spare cells (I think Intel is a good choice there).

...and I'm interested to know what the state of forensic-style software is for SSDs. (Not for recovery; people should have back-ups. But for investigations.)

Look at the warranty, expecting any device to outlast it's warranty is naive.

Longevity has no bearing on the need for backups either, so SSD vs. HDD longevity is irrelevant, in my opinion.

Completely anecdotal and kinda meaningless but I've never had one fail and I've bought 6 drives (ocz and samsung) over last 2 years. Nor have any of my friends/family who own them ever experienced a failure.

No, but I have full faith in SSDs for laptops and desktops now. I've never found the need for them a server.

Does anyone know of any large scale operations using them in servers?

Netflix appears to be using SSD instances for their Cassandra deployment on AWS.[1]

Also, more and more hosting providers start offering SSD-based products (e.g. DigitalOcean)

[1]: http://techblog.netflix.com/2012/07/benchmarking-high-perfor...

Facebook, Apple, Google...

DigitalOcean[1] is a ssd-only VM provider, by their numbers they should have a few TB in SSD storage. Not that large, but would be interesting to hear about their experiences.

[1] https://www.digitalocean.com/

Yes. It would be interesting to read about their experiences as I presume most large SSD deployments would be for databases only.

SSDs in your database server is a wonderful thing.

I just got a SMART error on my gaming machine SSD over the weekend. I'm starting the backup process to external USB drive tonight... just in case.

They're more predictable on lifespan, but hardware errors can always occur.

Reply on news.ycombinator.com