back to article Commodity flash just as good as enterprise drives, Google finds

If you're loading up a heap of flash drives for your data centre, don't bother with “enterprise-class” SLC (single level cell) technology, because cheaper MLC (multi-level cell) drives will do the job just as well. However, the data centre biz needs new techniques to predict drive failures, because the unrecoverable bit error …

  1. Nate Amsden

    seems just about everyone uses MLC

    Usually enterprise grade MLC but still MLC. SLC seems quite rare these days. With several enterprise storage systems putting full support behind MLC on any workload (not just read intensive).

    It would be interesting to know what brands of flash are considered good enough. For me personally on my own gear I only use samsung pro or the Intel 7xx series(have one in my PS4). Consumer flash has had too much history of dodgy quality it's hard to tell who to trust most any testing focuses purely on performance. There was one site that wrote petabytes of data onto flash drives to see which would fail that was interesting though didn't seem like a very realistic test. From what I recall reading people commented on how the flash drives would fail during power cycling and I believe that particular test didn't do any of that though could be mis remembering.

    Happy to pay a premium for something that is higher quality. Costs are low enough now that it makes sense to me. The 10 year warranty on the samsungs give me confidence that they have more confidence in their product. Even though i can't see myself using the particular drives past 5 years(probably be too small at that point).

    I bought one piece of shit corsair SSD many years ago. Lost data on power cycle. Stuck to seagate hybrid drives for a while until the samsung 850 pro came out. Also have a samsung 840 evo to store an image collection. I remember reading about firmware issues on the 840 evo though I haven't bothered to try to upgrade mine it seems to work fine under a really light workload.

    On the same note always wanted to see a reliability test of USB flash drives. Never have seen one though sites only test performance last I saw.

    1. Sandtitz Silver badge

      Re: seems just about everyone uses MLC

      "There was one site that wrote petabytes of data onto flash drives to see which would fail that was interesting though didn't seem like a very realistic test. From what I recall reading people commented on how the flash drives would fail during power cycling and I believe that particular test didn't do any of that though could be mis remembering."

      It was The Tech Report and their"SSD Endurance Test". W

      The drives were left after unpowered for a week after 300TB and for several weeks after 600TB. They all manager from .7 PB (Intel) to 2.4PB (Samsung 840 Pro). Curiously each and every one of them bricked after power cycle which is a terrible way to handle the problems. The drives should at least be available as read-only to scavenge data!

      The article itself states that the test was somewhat academical. At least I'm pretty sure that whatever SSD I'm using personally won't die because of over-use. Also, the test ran for over a year, which is a long time in computing. Some models may be obsolete and unavailable then.

      "Happy to pay a premium for something that is higher quality2.

      I fully agree. It's just hard to discern the difference. Samsung and Intel - both gigantic corporations - have had catastrophic firmware problems, including bricking. Then again so have all the other players. The same is true with HDDs - Seagate and WD have both published plenty of corrective updates for their drives. I just don't know whether Samsung premium models are more premium than others in the long run. Past performance doesn't guarantee future results. Samsung's software engineering QC has been outstandingly poor in some of their products I've had the misfortune to install and use. (digital signage, thin clients, printers)

      "On the same note always wanted to see a reliability test of USB flash drives. Never have seen one though sites only test performance last I saw."

      That may be even more academic than the SATA test since there are way more manufacturers (or at least brands) than in the SATA SSD business, and even more models available. Writing through USB is slower process and the endurance limits are thus even further away in the future. Also, USB drive usage is completely different from internal drives. I reckon for most people the sequential speeds are the most important factor.

      What matters to me with personal computer SSDs are just random and sequential performance, and even the poorest performers are pretty awesome for regular home/office use. Most people can't notice the difference between Samsung 850 Pro and older Samsung 840's, since SATA has been saturated for years. PCIe is a different beast, but still - how many do need something like 2 GB/s transfer rates in personal computers today? Some certainly do, but most don't.

      1. Alan Brown Silver badge

        Re: seems just about everyone uses MLC

        "Samsung and Intel - both gigantic corporations - have had catastrophic firmware problems, including bricking. "

        The Intel drive switched to read-only and clearly did it based on the total specified writes. All the others kept on trucking well past their specified "use by" number until they stopped (which admittedly isn't great, but it means I got a replacement drive under warranty because samsung couldn't verify how much had been written to the drive as a ZFS L2ARC)

        In a sensible environment once you get to the total published endurance you'll be treating any storage media gingerly, if not premptively replacing it. Even if "only a cache", downtime can be disruptive.

    2. Alan Brown Silver badge

      Re: seems just about everyone uses MLC

      "I remember reading about firmware issues on the 840 evo"

      840s have very slow garbage collection routines. If you trim or don't fill them, it's not so much of an issue.

  2. Anonymous Coward
    Childcatcher

    Dodgy journo?

    The way this article reads - throwing terms like "flash" "SSD" "MLC" and "SLC" around and quoting the article as concluding:

    "By comparison only 3.5 per cent of hard disk drives (HDDs) will develop bad sectors in 32 months"

    I got the impression that SAS/SATA is the way to go (maybe MFM/IDE will do!)

    *sigh* I'll read the source and make my own mind up

    1. Anonymous Coward
      Anonymous Coward

      Re: Dodgy journo?

      "8 Comparison with hard disk drives

      An obvious question is how flash reliability compares to that of hard disk drives (HDDs), their main competitor. We find that when it comes to replacement rates, flash drives win."

      From: "7 Comparison of MLC, eMLC, and SLC drives"

      "Based on our observations above, we conclude that SLC drives are not generally more reliable than MLC drives"

      I was taking the piss in my original post. The above are direct quotes from the study. So our journo has got got most of the message across. You can spin this however you like but in general:

      * MLC is as reliable as SLC - according to Google and the UoT

      * HDD and flash are different - according to me and lots of other people

      1. Gigabob

        gigabob@comcast.net

        Think of the difference between Flash and HDD like carbon fibre and wood. Wood alerts you to a failure while carbon fibre can endure far beyond expectations - until it fails catastrophically. This is why you should never buy a used carbon fibre bike frame - you don't know how it has been stressed or when it might fail. Check out http://www.cnn.com/2015/11/20/politics/air-force-gunship-lost/

        As new materials enter the marketplace and move from technical to consumer grade users many mistakes are being made and new usage models are being learned. Besides not putting your e-cig Lithium batteries in your pocket for them to overheat we should not do barrel-rolls in gunships.

        However, I maintain that the flash community needs something akin to the S.M.A.R.T. alerting for SSD.

  3. YARR

    More than 20 per cent of flash drives develop uncorrectable errors in a four year period

    If they have been using commodity flash for four years, then the research must be based on four year old flash drives and controllers. What I'd like to know is if the error rate for flash increases with flash capacity, e.g. Does a 1Tb flash drive have 4x the probability of developing errors than a 256Gb flash drive? If it does, then today's high capacity flash drives will have a much higher error rate.

    Error rates for flash v hard drives may scale differently with capacity. Flash controller failure may be less likely than physical hard drive failure, but individual flash cells may be more likely to fail permanently than the magnetic region representing a bit on the surface of a disc.

    1. Anonymous Coward
      Anonymous Coward

      Re: More than 20 per cent of flash drives develop uncorrectable errors in a four year period

      Still, the notable risk of a sudden catastrophic failure (a bricking) certainly raises some concerns since this is the toughest type of failure to cover: one day it's all fine, the next BOOM! Especially since a failure like this raises the specter of a failsafe failure if your backups also fail similarly. At least with gradual failure (something I'm personally dealing with now—a portable drive I've been using is starting to crash) you still have time and can employ mitigations.

      1. Anonymous Coward
        Anonymous Coward

        Re: More than 20 per cent of flash drives develop uncorrectable errors in a four year period

        Sudden catastrophic failure (and hard unrecoverable bad block) is the easiest to recover from if you mirror. I think laptops ought to include one M.2 and one 2.5" slot so you can either include cheap bulk storage in the form of a 2.5" HDD while still getting the benefit of SSD speed, or two SSDs if you prefer data protection. The laptops that want to dump the 2.5" slot to go thinner/lighter could at least include a second M.2 slot - or someone needs to invent a mini M.2 format so you can plug two such drives into a single M.2 carrier for mirroring.

        1. Charles 9

          Re: More than 20 per cent of flash drives develop uncorrectable errors in a four year period

          "Sudden catastrophic failure (and hard unrecoverable bad block) is the easiest to recover from if you mirror."

          But that takes time, and as long as you're trying to reconstruct, you run the risk of the mirror failing, too, thus the mention of failsafe failure. At least with a gradual failure you can still extract useful data from the failing drive to reduce the load on the mirror and reduce the chance of failsafe failure.

          1. Anonymous Coward
            Anonymous Coward

            Re: More than 20 per cent of flash drives develop uncorrectable errors in a four year period

            If you are that paranoid that you can't run for even a short time unprotected, you need to triple mirror or use dual parity RAID6.

      2. Alan Brown Silver badge

        Re: More than 20 per cent of flash drives develop uncorrectable errors in a four year period

        "one day it's all fine, the next BOOM!"

        Apart from one cheap and nasty ATA-interface flash drive, the few I've had that completely failed gave months of warning.

        One of the more interesting failure modes in some is a lockup at 49 days (yes, that infamous tick timer).

        A couple of other commodity drives lock up after several weeks of being powered up and then need to sit powered down for 24-96 hours before working again.

        Both of these issues are something that would never be noticed in a laptop or the average windows box.

        A few uncorrectable errors aren't a big deal in an enterprise setup. That's what you have RAID for. (Experience here is that mechanical drives fail at a rate 5-10 times higher than SSDs.)

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like