Enterprise SSDs, Powered Off, Potentially Lose Data In a Week

New submitter Mal-2 writes with a selection from IB Times of special interest for anyone replacing hard disks with solid state drives: The standards body for the microelectronics industry has found that Solid State Drives (SSD) can start to lose their data and become corrupted if they are left without power for as little as a week. ... According to a recent presentation (PDF) by Seagate's Alvin Cox, who is also chairman of the Joint Electron Device Engineering Council (JEDEC), the period of time that data will be retained on an SSD is halved for every 5 degrees Celsius (9 degrees Fahrenheit) rise in temperature in the area where the SSD is stored. If you have switched to SSD for either personal or business use, do you follow the recommendation here that spinning-disk media be used as backup as well?

Comments

I call BS

[Anonymous Coward]

If the contents are lost in a week, we're probably talking about capacitor-backed SSDs that use some other technology than flash memory. Yes, it would be insane to use flash memory for archival purposes as well, but it still should easily retain its contents for at least a decade. When powered on, this problem does not exist as normally the controller slowly walks through the flash refreshing it.

Re:I call BS

[luther349]

probably and no data center is going to not have power for a week and always have non flash based memory backups even if they did.

Re:I call BS

[gweihir]

The statements are actually completely accurate, but a bit misleading. First, this is about what JEDEC requires, not what actual SSDs deliver. Second, this is when SSDs are stored in idle at 55C. And third the JEDEC requirements for minimum off-time data-retention are only 3 months @40C for enterprise-grade SSDs and only 12 months for consumer SSDs at 30C. These are kind of on the low side, although I have lost some OCZ drives that were off for just about a year. (Never buying their trash again...)

That said, anybody conversant with SSD technology knows that SSDs are unsuitable for offline data storage as data obviously has potentially far shorter lifetimes than on magnetic disks, which in turn again have far shorter data lifetime than archival-grade tape. These is absolutely no surprise here for anybody that bothered to find out what the facts are. Of course, there are always those that expect every storage tech to keep data forever, and those dumb enough to have no or unverified backups and those often on media not suitable for long-term storage. Ignoring reality comes at a price.

My personal solution is mixed HDD/SSD Raid1 and, of course, regular backups.

Re:I call BS

[Kjella]

My personal solution is mixed HDD/SSD Raid1

Uhh... doesn't that mean that the RAID controller has to wait for the HDD on every read/write to verify it's the same as on the SSD, so effectively you get HDD performance?

Re:I call BS

[Spazmania]

Every write, not every read. Reads are satisfied as soon as either drive returns the data. And if the raid controller has a battery or supercap so it can cache writes, you'll almost never notice the difference.

Re:I call BS

[DarkTempes]

I might be wrong but isn't it also when the SSD is stored at 55C AFTER having been stress tested at 55C to their endurance rating in terabytes written (page 39) under a given workload?
And even then the cherry picked value was in example data submitted by Intel for unknown hardware and very likely extrapolated and quite possibly meaningless because it wasn't part of the chart targeted for the standard.

The article seems to have totally misrepresented the presentation's purpose: which is to lay out endurance testing methodology/standards.

The only important values were on page 26 where they set the minimum requirements of 40C 8hr/day load/30C 1 year retention for consumer (with a higher error ratio) and 55C 24hr/day load/40C 3 months retention for enterprise (with a lower error ratio.)
And it looks like they haven't actually worked out the consumer workload for testing yet.

Re: I call BS

[bill_mcgonigle]

use ZFS l2arc/zil or flashcache/dm-cache to get a happy medium.

Re:I call BS

[dfghjk]

"Reads are satisfied as soon as either drive returns the data. And if the raid controller has a battery or supercap so it can cache writes, you'll almost never notice the difference."

RAID controllers do not launch reads on all involved drives. That would be stupid.

Implementing battery backed write back cache on an array that uses SSD would be similarly stupid.

RAID 1 with mixed SSD/HDD is the worst of both worlds further complicated by people who don't understand it.

Re:I call BS

[Kjella]

Every write, not every read. Reads are satisfied as soon as either drive returns the data. And if the raid controller has a battery or supercap so it can cache writes, you'll almost never notice the difference.

Ah, I thought RAID1 would warn you somehow of bit flips which I assume would be the way heat-deteriorated storage would show up. Guess it won't, you'll need ZFS or something like that.

Re:I call BS

[drsmithy]

RAID controllers do not launch reads on all involved drives. That would be stupid.

?

For a RAID1, most RAID controllers (and software RAID implementations) will absolutely read from all devices so as to service the read ASAP.

For distributed parity forms of RAID, you inherently have to read from all devices.

For dedicated parity disk forms of RAID, you have to read from all devices except the parity device.

I've never tried a mixed RAID1 of SSD and magnetic disk, but with a large enough write cache the theory seems reasonable. Most controllers [with a BBU] acknowledge the write as soon as it hits cache.

Re:I call BS

[sribe]

Yes, it would be insane to use flash memory for archival purposes as well, but it still should easily retain its contents for at least a decade.

Nope. It's nowhere near that long--more like 1 year, not 10. And that time reduces as the flash wears through being written.

Re:I call BS

[silas_moeckel]

Thats very dependant on whose implementation of raid 1. I've seen everything from read from one drive, stripe reads, and read from both and compare. Linux will actually let you choose from among some of those options.

ZFS and btrfs add a crc for a group of blocks and and detect which drive has the bad data, correct that and tract that it happened.

Re: I call BS

[thogard]

I've been thinking about getting a bunch of cheap usb sticks and building a zfs pool out of them with some redundancy and then using that for a zfs pool for a usenet spool just to see what can go wrong. If anything can abuse a disk its usenet.

Re:I call BS

[jarfil]

the RAID controller has to wait for the HDD on every read/write to verify it's the same as on the SSD

Drives use checksums, no need to read from a second drive if it succeeds from the first one.

Also, "write mostly" configurations for the HDD allow for most reads to be made from the SSD, with the HDD acting as a backup. Writes still need to be duplicated, but this can be done in the background.

Re:I call BS

[Anonymous Coward]

You must be joking. We would be in deep trouble if flash memory held its information for only about a year.

Re: I call BS

[X0563511]

The modern day floppy-raid.

Re:I call BS

[SharpFang]

Cherry-picked for "a week" yes, but still disturbing. It's not an issue for datacenters, but for offices.

Imagine an office PC set next to the radiator - oh, the employees are free to set up their desks as they like, and they really don't care about stuff like that. Given employee going for a holiday break for a month, taking the family for a skiing trip. The PC experiencing 50C on regular basis. That's quite enough to cause the data loss.

Yes, in a responsible company there will be backups - or the data will be held on network drives. Visit your average office and see if they use these responsible practices; roughly half don't. Yes, once that occurs any admin who isn't a total idiot will spot the reason for the problem and recommend moving the PC away from the radiator. Still, the problem is not an impossible one - just needing a bad set of circumstances; iterate over enough PCs and offices and it *will* happen.

Re:I call BS

[Phs2501]

You must be joking. We would be in deep trouble if flash memory held its information for only about a year.

Nope, not joking. It also depends on the process size. Nice big (i.e. low capacity in a large die size) SLC flash cells hold data for quite a long time. The higher-density they get (and the less electrons per bit used to store data) the worse it gets.

So a lot of device firmwares and BIOSes, which generally use nice big chunky flash cells, will last 10-20 years. High-capacity flash storage, not so much.

That's just the way it is. Flash is currently the least worst solid state storage solution we have, but it still sucks.

(Note that if your device is powered on occasionally the flash can error-check and rewrite itself, which at least partially "resets" the time for data loss. This is only an issue for devices that are constantly powered off or devices that will not rewrite themselves as required.)

Re: I call BS

[Anonymous Coward]

Even with crcs on drives, single bit errors happen, on the order of 1 bit flip per TB or so of data read/written. This can be an error uncorrected/undetected error by the crc on disk, bit flip on the bus when reading or writing, or bad RAM. FS checksums protect against everything but bad RAM.

Re:I call BS

[Anonymous Coward]

Belly laugh. Oops, an ant walked over my SSD, oh dear, did it get destroyed!! Once again, no modern off-the-shelf SSD will lose its contents in 1 year if not refreshed.

Re: I call BS

[hr raattgift]

I've done this, experimentally, using not-super-cheap 128GiB patriot and hyper-x usb3 sticks.

For a USENET load, performance will depend on whether your incoming feed is predominantly batched, streaming, or effectively random -- small writes bother these devices individually, and aggregating them into a pool works best if you can maximize write size. One way to do that is to use a wide stripe (e.g. zpool create foo raidz stick0 stick1 stick2 stick3 stick4 ...), which works well if your load is mainly batched or streaming, such that there is lots of stuff for zfs's various write aggregation strategies to cut down on the number of small writes.

You'll also want a reasonably large ashift, for similar reasons.

If your workload is mostly reading (e.g. you have lots of downstream feeds, especially synchronous or synchronously-batched ones, or many client readers), then a sufficiently large ARC should make your choice of pool layout somewhat less important. If you are VERY read-dominant, you'll likely want to take the hit on write performance and use n-way mirroring vdevs. (e.g. zpool create foo mirror stick0 stick1 mirror stick2 stick3 ...).

If you're loss-intolerant, raidzN where N>1, or 3+-way mirroring.

What can go wrong? Mostly that you'll run out of space. :-) For that reason, a mirroring strategy, even though it's space-intensive and slows down writes, provides useful flexibility, as you can always widen or narrow mirror vdevs, or add more mirror vdevs to a pool, and it's easier to swap out vdevs in a mirror with something else than it is to swap out vdevs in a wider individual vdev (a wide raidz stripe for instance). (But you can't remove mirror vdevs (or raidzN vdevs) from a pool...).

Device names are likely be important considerations. Make sure the pool devices use device names that persist across reboots, USB3 hub failures, physical or logical disconnects and reconnects, and so forth. How to do that depends on the specific zfs port and operating system.

What can go right? Negligible seek times! That makes *all* the difference compared to spinny disks. Even for super-cheap USB 3 sticks. Really.

Final thing: no zfs port (none of the openzfs ones, and not opensolaris either) is good with USB 2 in general, and USB 2 sticks are often flaky. Avoid them.

Re: I call BS

[hr raattgift]

Ah, cannot ETA, so: decent USB 3 sticks make *excellent* l2arc devices in front of spinny disk pools. They can often deliver a thousand or two IOPS each with a warm l2arc, and that would otherwise mean disk seeks. I use them regularly.

(The downside is that in almost all cases l2arc is non-persistent across reboots, crashes and pool import/export, and can take a long time to heat up so that a good chunk of seeks are being absorbed by them, so you're limited by the IOPS of the storage vdevs and the size of ARC as it heats up. That's true for *any* l2arc device, though. Building a pool out of things you would use as l2arc devices gives you persistence across import/export, reboot, and ideally panic. :-))

(Some of these downsides are likely to fade with the eventual integration of persistent l2arc in all ports, ongoing changes to arc in opensolaris, being able to assign specific DMU data to specific pool VDEVs and/or further specialization of vdevs (i.e., in addtion to log and cache), and so forth in openzfs).

Re: I call BS

[Anonymous Coward]

No. This is more about the intrinsic capability of the NAND. Any SSD maker worth their salt has some very sophisticated error handlingto avoid this though.

Re:I call BS

[tepples]

For dedicated parity disk forms of RAID, you have to read from all devices except the parity device.

I think the idea is to make a dedicated parity disk RAID with one data SSD and one parity HDD.

Re:I call BS

[Mashiki]

I don't know about that. Even at every office that I've ever done work at even the women refuse to sit near/next to the radiators. That includes in super cold areas where it hits -40C to -50C in the late fall/winter. And in the cases where the 'rad' is pumping out 55C temps, there is already 30-100CM of space around them to simply stop possible burns from the rad which gives you plenty of space to normalize the air temperature. Most places are now on forced air from the ceiling.

Re:I call BS

[Anonymous Coward]

Linux md raid supports tagging drives as write_mostly. In ssd+hdd raid1 writes go to both, reads only to ssd.

Re:I call BS

[Spazmania]

Implementing battery backed write back cache on an array that uses SSD would be similarly stupid.

How do you figure? Write to ram is a whole lot faster than write to flash, especially if the flash block has to erase first.

Re:I call BS

[EndlessNameless]

For a RAID1, most RAID controllers (and software RAID implementations) will absolutely read from all devices so as to service the read ASAP.

For distributed parity forms of RAID, you inherently have to read from all devices.

The problem is guaranteed with distributed parity raid; the controller will have to wait for the slowest disk to complete the read. Both reads and writes will be limited to mechanical disk performance levels.

With a RAID1 mirror set, you can get a performance improvement on reads since the SSD would presumably service all of them. Writes will still be delayed by the mechanical drive(s).

In addition, most RAID controllers do not support mixing drive types. Most of them don't even recommend mixing drive speeds (e.g., don't even mix 10K and 15K RPM drives). So you are proposing a whole new product essentially, and the expected gains are quite minimal.

Rather than investing in new tech with questionable benefits, implement an existing solution. You can choose to setup a hot sync, cold sync, or backup device for a standard SSD array that is well-tested and performant.

Re:I call BS

[KingMotley]

For a RAID1, most RAID controllers (and software RAID implementations) will absolutely read from all devices so as to service the read ASAP.

No, almost every RAID1 controller I've ever encountered does not do that at all. It balances the reads across the drives so that the it maximizes throughput and IOPS. Only when one drive attempts to read a sector and it detects an error through it's internal CRC checks and is unable to rectify the error (short period for raid drives, long period for desktop class drives), THEN it will request the data from the alternate drive and have the original drive correct itself.

Re:I call BS

[KingMotley]

Sorry, the same applies to parity drives and dedicated parity drives as well during reads.

During writes, all the data on a particular stripe need to be read so that the correct parity can be calculated.

Re:I call BS

[KingMotley]

Ah, I thought RAID1 would warn you somehow of bit flips which I assume would be the way heat-deteriorated storage would show up.

It does. The description of how RAID1 works was incorrect. No raid controller that I am aware of implements RAID1 that way. That would include DELL's persec raid controllers, INTEL's ICH raid controllers, Adaptec raid controllers, LSI's raid controllers, rocket raid controllers, and window's implementation.

Re:I call BS

[Anonymous Coward]

Don't forget that archival grade tapes have a far shorter data life than printed paper, which has a shorter life than clay tablets....

Re:I call BS

[gweihir]

Indeed. Laser print (or pigmented ink) on good paper will get > 100 years. The question is what you need. "Archival" media usually start at > 10 years ensured and for that archival tape is the best option these days. It is also not that expensive. About 1500 EUR/USD should get you a suitable drive and about 500EUR/USD more should get you a starting supply of tapes. That is not expensive on any professional scale.

Re:I call BS

[unrtst]

RAID controllers do not launch reads on all involved drives. That would be stupid.

I think you mean that they do not launch a read request for the same chunk of data on all drives in a raid mirror. That would be accurate. However, they usually will read from both drives (read chunk 1 from drive A, read chunk 2 from drive B... doing so in parrallel can significantly increase read performance using a mirror).

RAID 1 with mixed SSD/HDD is the worst of both worlds further complicated by people who don't understand it.

Do you mean people like you?
Look up "md raid write-mostly", or try this page (one of many found): http://tansi.info/hybrid/
That setup is for a linux software RAID 1 mirror with one side being SSD and one side being HDD. The HDD has "writemostly" set. This causes the kernel to only do (slow) reads from the HDD if they are really needed, with all other reads going to the SSD.
This is a very very simple setup that can easily give one a huge read performance boost with very safe HDD mirror backing the data. For much higher performance and tons of other data integrity options, see advanced ZFS setups (ex. with a dedicated SSD for the zil, another for L2ARC (a level 2 cache), and also in various hybrid style allocations for the pools).

Re: I call BS

[Wolfrider]

--Is there a decent USB3 thumbdrive brand you can recommend that is available from say, Amazon Prime? TIA

Re:I call BS

[Anonymous Coward]

The RAID controller doesn't have a clue about the "drive type". It sees an SAS or SATA drive, not a spinning platter or SSD.

I image my SSD regularly

[Anonymous Coward]

8 hours of deduplication and compressing aside, it's not being kept on Seagate products, I can assure you.

Re:I image my SSD regularly

[Anonymous Coward]

Seagate makes SSDs now?

Re:I image my SSD regularly

[Dunbal]

Haha! I'm old enough to remember when Seagate was the best.

Re:I image my SSD regularly

[thogard]

And I remember when Al Shugart took that claim from Shugart Corporation with his newer company.

How powered off is "powered off"?

[dwywit]

There's full shutdown, there's power left on at the wall, there's hibernate with wall power left on, there's sleep. Lots of laptops come with (toshiba, for example) "sleep-and-charge" where they will supply current to USB ports while asleep. I rarely do a full shutdown on laptops or desktops, would this be enough to avoid the problem?

Or would there need to be a BIOS feature to ensure current supply to SSDs as well as USB ports?

Re:How powered off is "powered off"?

[luther349]

this isnt even the same kind of ssd that's in your pc probably cap based.flash memory based can stay powered off forever and not lose data.

Re:How powered off is "powered off"?

[Carewolf]

this isnt even the same kind of ssd that's in your pc probably cap based.flash memory based can stay powered off forever and not lose data.

Non-enterprise are in the presention rated to retain data for a year.

Re:How powered off is "powered off"?

[TheRaven64]

Data loss in flash cells is probabilisitic, but the standard for wear levelling software is to mark a cell as dead if it can't guarantee holding its contents for a year. There have been some interesting proposals to expose this to the OS - there's a lot of data where even holding it for one day would be fine (e.g. browser caches, swap, other temporary files), so the more worn cells could be productively used for this data.

Re:How powered off is "powered off"?

[fuzzyfuzzyfungus]

The floating gates are isolated; but not perfectly, and on a modern high density device(especially an MLC one) it doesn't take much leakage to result in the wrong result.

Decay within a week is pretty aggressive for anything you'd have the nerve to sell; but all flash memory can be expected to lose its contents over time.

Re:How powered off is "powered off"?

[Anonymous Coward]

My PC, and therefore its SSDs were shipped across the Atlantic, took seven weeks, no power, no problems. Thought the salt air might be an issue, but it was mainly furniture that suffered, not electronics.

Re:How powered off is "powered off"?

[gweihir]

Untrue. Very, very untrue. Even industrial-grade long-term storage FLASH has often only 10 year data retention. And these are selected SLC cells and you pay about 1000x per capacity of what a regular SSD costs. "Cap based" storage is called "DRAM" and loses data after seconds to minutes.

Seriously, do a bit of research before claiming complete nonsense.

Re:How powered off is "powered off"?

[jellomizer]

When ever something is "Enterprise" class it means it is vastly inferior to other solutions on the market but cost 3x as much.
For some reason this isn't a buzzword that sends shivers down every IT workers spine yet.

Re:How powered off is "powered off"?

[dshk]

You may be right in case of other equipment, but enterprise grade drives are really better. For example I do not know any consumer SSD which has power loss capacitors (Intel 320 is not produced anymore). Most consumer drives don't contain even those capacitors which would be necessary to prevent the loss of - not the freshly written but the - old(!) data in case of a power loss. Consumer HDDs lied (or lying?) about sync, they confirm sync before they actually save the data to disk. And I am sure that consumer SSDs do something similar, because consumer SSD are usually faster (although their speed frequently fluctuates to extreme extents) than their corresponding enterprise variants, which is impossible in a safe way without power loss protection capacitors.

Re:How powered off is "powered off"?

[drinkypoo]

Or would there need to be a BIOS feature to ensure current supply to SSDs as well as USB ports?

there would need to be a hardware feature. the power supply doesn't send power to the drives when the computer is not turned on. they are unpowered even in some sleep modes.

Re:How powered off is "powered off"?

[plover]

Your stuff's cargo container was not heated during shipping. If it was stacked below the deck line, where it was not exposed to the sun, it didn't reach 55C during the journey.

Re:How powered off is "powered off"?

[ihtoit]

reason being that you shouldn't be buying Enterprise grade through a brick retailer. You should be leasing it via a support contract: the premium is with a tech on the other end of the phone who's out in a couple hours to replace a dud drive and have your RAID rebuilt before the day ends, rather than you running to the nearest PC World for a TB Seagate pocket spinny. Like I've just had to. If I'd had a support contract (hence Enterprise grade drives as they generally insist on anyway since they're easier for ICTs to RMA) then I'd've been at my mum's now sealing her windows rather than babysitting my new drive still.

Re:How powered off is "powered off"?

[fnj]

Yes, conceptually all flash data has to decay when powered off. But implementation tradeoffs vary widely. A dirt-cheap Microchip PIC18F2580 microcontroller has flash data retention without refresh "conservatively estimated" at 40 YEARS MINIMUM and 100 years typical (page 3, 10, 435). The number of previous erase/program cycles that retention is predicated on is not given, but is probably a single one, or a few, out of a typical endurance of 100,000 cycles and a minimum of 10,000 from -40 to +85 C. AFAIK there is no wear leveling or block sparing in microcontroller flash memories.

I have NEVER HEARD of an embedded guy ever running into a case of either cycle exhaustion or data decay in the program memory such microcontrollers (if data is written to flash during operation, specs do have to be considered).

SSDs have flash design tradeoffs remarkably different from this. For example, MLC individual cell endurance is around 1000-10,000, and retention is far less as seen in the article and comments here. In return, the access time is vastly faster and the density vastly higher.

Re:How powered off is "powered off"?

[lgw]

There's confusion here between various storage devices that happen to have a capacitor. There's DRAM, of course, and there are RAID controllers that keep enough power to "finish a write" during power loss, to greatly speed write caching. But there are also SSDs, and SSDs have the same problem with "finishing a write" during power loss. Most consumer SSDs risk losing data in a power hit, while the better ones have a capacitor just to avoid that problem. I think that's what they're talking about here: the grade of SSDs where it's safe to believe the drive when it tells you a write has completed.

It amazes me the number of devs I've worked with over the years who just could not understand that write order isn't necessarily preserved across crashes (or even restores from backups), and being clever with "I wait for the write of A to finish before I write B, so there can never be a B without its A" is wishful thinking.

Re:How powered off is "powered off"?

[mister_playboy]

For example I do not know any consumer SSD which has power loss capacitors (Intel 320 is not produced anymore).

The Intel 730 series have these capacitors.

Re:How powered off is "powered off"?

[gringer]

You may be right in case of other equipment, but enterprise grade drives are really better.

BackBlaze disagrees with you:

https://www.backblaze.com/blog...

Overall, I argue that the enterprise drives we have are treated as well as the consumer drives. And the enterprise drives are failing more.... Enterprise drives do have one advantage: longer warranties. That’s a benefit only if the higher price you pay for the longer warranty is less than what you expect to spend on replacing the drive. This leads to an obvious conclusion: If you’re OK with buying the replacements yourself after the warranty is up, then buy the cheaper consumer drives.

Re:How powered off is "powered off"?

[SharpFang]

You're way optimistic with that "forever". Flash memory is based on electromagnetic charge which has the nasty property of leaking.

I tried some MicroSD cards that's been sitting in my drawer for past 4 years. All needed to be reformatted. SSDs may retain data longer, but it's by no means 'eternal'.

Re:How powered off is "powered off"?

[dgatwood]

It really is unsurprising. IIRC, enterprise drives and consumer drives are usually the same basic drive mechanism, with the main differences being the lack of a park ramp in enterprise drives and firmware differences, such as parking the heads less frequently, spinning down less frequently, etc. And while those firmware differences could make a difference if they happen to tickle some serious design flaw, for the most part, I'd expect the differences to be lost in the noise.

There are some exceptions, of course. Some of the newer enterprise drives have extra bearings so that the motor shaft is suspended in two places to reduce vibration. This mainly results in a performance win (fewer and faster head position calibrations), but in theory could slightly reduce the rate of head crashes in an enclosure with a large number of drives. (In practice, I'd imagine you'd have a hard time measuring an effect that small, though.) And in theory, it could slightly reduce the risk of writing a track slightly out of place, forcing the disk to retry a couple of times while reading it back. Neither of these constitutes a significantly increased risk of hardware failure, though, I don't think.

Of course, hard drive manufacturers tell us that consumer drives aren't rated for continuous use. That seems pretty unlikely to me. After all, a motor is under the least stress when it is either spinning at a constant speed or stopped. It takes very little power and very little force to keep something spinning. Based on that, I'd expect a drive to last longer when spinning constantly than when spun up and down constantly. That was certainly true historically; I have many drives that spun continuously for more than a decade. The hard drives I've owned that have failed (ignoring purely acoustical failures of ball bearings) have invariably been the ones that were spun up and down routinely. (Well, okay, I did see bad blocks in one swap partition of one drive that had been spinning for seven or eight years, but after disabling that swap partition, I kept using the drive for about another half a decade or so before retiring the server, so I'm not sure I'd call that a failure, per se.)

And it isn't as though we're still using ball bearings, where one could reasonably be concerned about a continuously spinning drive not having full lubrication because the lubricant is spun out towards the outside of the bearing or something (hence "acoustical failures"). So I can't really think of any plausible reason why a drive operating 24x7 would ever have a higher failure rate than a drive spinning for a fraction of that time, hence I have a hard time believing that there is any truth to the manufacturers' claims. I suspect it is mostly a scare tactic to make more money off of those customers who can afford to spend more... or perhaps an excuse to renege on their warranties if they suffer another round of capacitor plague. Either way....

Re:How powered off is "powered off"?

[dgatwood]

By definition, all drives lose data when power disappears suddenly. You can't guarantee that a filesystem metadata change won't be halfway complete unless the OS is in control over when the drive powers down.

Consumer drives shouldn't be at any higher risk of data loss than enterprise drives in this regard. Either data is on disk or it isn't. If a file gets halfway written, it will still be halfway written whether the drive stays up for an extra few hundred milliseconds to flush its buffer or not. More to the point, assuming the buffer is flushed to disk at the same speed, the average risk should be almost precisely the same.

Say you're writing data for ten seconds, and it takes an extra 200 milliseconds at the end to flush the buffer to flash. (These numbers are arbitrary.) During the first 200 milliseconds, the first block is effectively not yet on disk, because the drive has not written out the log entry indicating that the block is now on a different flash page than before). If power fails during those 200 milliseconds, the consumer drive will not lose data, whereas the enterprise drive will, because that initial write will get written to disk.

Thus, on average flushing the buffer (outside of the operating system's direct control) should cause data loss by allowing a write to start that wouldn't have started otherwise almost exactly as often as it will prevent data loss by allowing a write to complete. So IMO, the whole "extra capacitors" thing is a cute idea to sell more drives, but in practice, it is basically worthless.

BTW, although write order isn't necessarily preserved, if such behavior is really important, that's where you need to take advantage of a sync system call, which guarantees that all data previously sent towards the disk is complete before the call returns. (Note that in some operating systems, the first sync system call returns immediately, but a second sync system call blocks until the first one completes. For this reason, you should always make that system call in pairs.)

If a sync system call fails to enforce write ordering, then we have a word for drives that report that a sync operation is complete before it actually is. Defective. Such a disk is fundamentally untrustworthy for any purpose. Fortunately, with the exception of a few badly broken USB drives from many years ago, that should never happen in practice.

Re:How powered off is "powered off"?

[dbIII]

Backblaze pack their stuff in tight with very poor airflow so the only conclusions we can draw from them is which drives behave better in hot environments. I'd go more with the google tests from a few years back which give the same sort of answer in general terms, unless of course you plan to pack things in very tight and reduce life by running them hot.

Re:How powered off is "powered off"?

[dbIII]

And it isn't as though we're still using ball bearings, where one could reasonably be concerned about a continuously spinning drive not having full lubrication because the lubricant is spun out towards the outside of the bearing or something

Still happens, just over a longer lifetime than it used to. Lubricant is also lost over time whether the drive is spinning or not, so at some point (probably well over ten years now) that old drive in storage is not going to spin up.

Re:How powered off is "powered off"?

[Anonymous Coward]

The hard drives I've owned that have failed (ignoring purely acoustical failures of ball bearings) have invariably been the ones that were spun up and down routinely.

How often is that, just to give me an idea?

Re:How powered off is "powered off"?

[lgw]

You can't guarantee that a filesystem metadata change won't be halfway complete unless the OS is in control over when the drive powers down.

The filesystem needs to be able to trust that a drive has completed its write when the I/O completes. This is a difference between SSD and spinning rust - most SSDs lie, as they're often still shuffling things around when they report I/O completion on the bus. What's more, the SSD tend to have far more internal metadata, and you can actually lose the whole volume on some of these SSD due to sudden power loss. Finishing up the write internally fixes that.

It's the same old problem with RAID controllers - NetAp built a huge business on simply ensuring trust in write buffering, with battery backup of cache and whatnot, so that you could trust the controller when it reported the I/O complete (even though the data hadn't made it to disk).

BTW, although write order isn't necessarily preserved, if such behavior is really important, that's where you need to take advantage of a sync system call, which guarantees that all data previously sent towards the disk is complete before the call returns.

This is exactly what I'm on about. This doesn't always work. All you can do with a "sync" is get the data through filesystem buffering. Not RAID controller buffering. Not disk internal buffering. Everybody lies.

And none of that matters anyhow, if you're writing to two volumes (as is often the case with such schemes), as it's possible you'll lose the whole system, and get restored from backups taken at different times.

If a sync system call fails to enforce write ordering, then we have a word for drives that report that a sync operation is complete before it actually is. Defective.

Nope. "Normal" is the word. That's normal for consumer SSDs. Normal for RAID controllers. Normal for some old-school drives in some error conditions. And the whole thing is irrelevant when it comes to restoring from backup.

Such a disk is fundamentally untrustworthy for any purpose.

Most people never notice. Most software will never care. And caring used to mean you either pay 10x, or take a 10x performance hit. Now with SSDs, it's the bit extra enterprise SSDs cost.

Before Microsoft's elaborate snapshotting technology, it used to be the realm of 6-7 figure backup products to solve the cross-disk backup problem (taking coherent multi-volume snapshots requires applications cooperate in quiescing). That's now a solved problem with mainstream software and mid-tier storage devices, thanks to MS demanding it for the lower-end stuff Windows servers tend to run with.

Re:How powered off is "powered off"?

[gweihir]

There is also a grace-period with a good-quality PSU: The SSD detects when the power is going down (like all disks do) and if the gradient is flat enough, has quite some time do get into a safe state. The OS also has at least 20ms with a good PSU, as it will keep voltages stable for that time or longer after power_good gets de-asserted. With a sane firmware design, that should be enough for the SSD to make sure no data that already was written gets corrupted (due to large sectors). Of course, with a catastrophic power-loss (crowbar over the power-rails), anything can happen, but it is not cost-efficient to prepare for that scenario.

As usual things are more complicated. The biggest problem is however that drives lie about having flushed data tor surface/FLASH, as that makes it impossible for FS designers to do a reliable 2-phase commit. Usually things are not that bad though and a power-loss during a write will always result in loss of some data. In a sane design, it will only affect files that were being written at power-down or shortly before and that is usually enough. For an actual reliable 2-phase commit, you either need a journal on a device with reliable flush or make sure the power does not go down without that being announced to the software minutes before (hence UPS) though.

Re:How powered off is "powered off"?

[gweihir]

Actually, as the Linux FS folks found out some time ago, most HDDs do lie too. Makes them look better in benchmarks.

If you really need to have a reliable 2-phase commit, use an UPS that forces an orderly shutdown way before it runs out of power or put the journal on a device with reliable flush. You may also want a redundant PSU. This is a solved problem. It is just that for consumer-grade and cheaper server-grade hardware it does not make sense to implement the solutions and, in addition, people have unrealistic expectations and incomplete understanding of what happens. In most cases things work well even without hard flushes.

Re:How powered off is "powered off"?

[Agripa]

The USB Flash drives I have tested only scrub on reads and writes. Whether the drives were powered or not did not affect retention which was months when new (at room temperature).

I assume that SSDs perform scrubbing while powered. I would like to know if Compact Flash and SD cards do but my guess is that they do not.

Re:How powered off is "powered off"?

[Agripa]

Flash memory retention has nothing to do with whether it has capacitor backup and being powered only matters if the drive scrubs itself.

Re:How powered off is "powered off"?

[Agripa]

The Crucial M500 has them as well. I wish there was a list of drives which have been verified to not suffer corruption because of power loss during writes.

Re:How powered off is "powered off"?

[Agripa]

I have exhausted UV erasable EPROMs and UV erasable EPROM based microcontrollers but never a Flash based one.

While cell size and the number of bits per cell matters for the number of program and erase cycles and retention, the type of Flash memory also matters. NAND Flash suffers from significant read (and write) disturbance which is much less of a problem with EPROM, EEPROM, or NOR Flash.

Re:How powered off is "powered off"?

[Agripa]

The power loss problem with SSDs is actually much worse than with rotating storage extending past just losing recently written data. Every write to a SSD including garbage collection requires manipulation of the sector translation table and if *that* becomes corrupted, then the drive will usually become inaccessible. The power loss protection is needed more to protect the drive's metadata than pending written data. If the SSD *only* lost the most recently written data, then there would not be a problem.

It amazes me the number of devs I've worked with over the years who just could not understand that write order isn't necessarily preserved across crashes (or even restores from backups), and being clever with "I wait for the write of A to finish before I write B, so there can never be a B without its A" is wishful thinking.

Drives support some form of write barrier which is suppose to allow this but of course many drives, rotating and solid state, lie about completed writes in the quest for better benchmark scores.

Re:How powered off is "powered off"?

[Agripa]

There is more going on with Flash based drives which makes power loss during writes more dangerous than with rotating media. Besides the risk of corruption to the Flash drive's metadata including the sector translation table which can render the drive completely inoperable, the Flash chips themselves can operate unpredictably if power is lost during a write.

If the risk was only to pending writes, then Flash drives would be no worse off than rotating media.

Re:How powered off is "powered off"?

[Agripa]

There is also a grace-period with a good-quality PSU: The SSD detects when the power is going down (like all disks do) and if the gradient is flat enough, has quite some time do get into a safe state. The OS also has at least 20ms with a good PSU, as it will keep voltages stable for that time or longer after power_good gets de-asserted. With a sane firmware design, that should be enough for the SSD to make sure no data that already was written gets corrupted (due to large sectors). Of course, with a catastrophic power-loss (crowbar over the power-rails), anything can happen, but it is not cost-efficient to prepare for that scenario.

The hold-up time after the power good signal is deasserted should be long enough however the drives do not have access to this signal. Even the operating system does not have access to this signal on a normal desktop computer because it results in the equivalent of a hardware reset.

Re:How powered off is "powered off"?

[Agripa]

My own experience with USB Flash drives is that even when new, they had a powered and unpowered retention of less than a year.

Re:How powered off is "powered off"?

[SharpFang]

Probably crappy controllers that don't refresh the written data. I guess the retention is 1 year since last write of given block.

Re:How powered off is "powered off"?

[Agripa]

I suspect that the difference is that SSDs always or usually support scrubbing when powered during idle periods while various Flash memory cards and drives (USB, CF, SD, whatever) only scrub on reads and writes.

Re:How powered off is "powered off"?

[dgatwood]

The sector translation able is what I was talking about. That table is in the form of a log journal. If a drive is writing to that journal, thus recording the fact that the block moved from point A to point B, prior to actually finishing writing the data at point B, the drive is defective by design. Period. And if the drive did not write that journal block, then the data did not move to a new flash page, and the write didn't happen. It should not be possible for background updates to cause data loss, because the old copy will always still be there on the disk until such time as the new copy has been fully written and the translation table/log journal has been updated.

Thumb Drives

[Anonymous Coward]

Aren't thumb drives flash? isolated charge type memory - they don't seem to forget sitting in my thumb drive wallet for weeks on end.

Re:Thumb Drives

[fluffernutter]

I looked into this because I once wondered why no one uses thumb drives for backup. Thumb drives are reasonably safe for a couple years but after that many can degrade. I saw many sites indicating that flash is not a safe media. This caused me to wonder what they did different in SSD technology. The only safe media that I know of is tape.

Re:Thumb Drives

[Anne Thwacks]

If the temperature in my pockets exceeds 55C, then data on a flash drive is not likely to be my biggest concern. YMMV.

Re:Thumb Drives

[jarfil]

Tape is some of a myth.

The only safe media, is that which you keep copying before it deteriorates. Not HDDs, not SSDs, not CDs, not thumbdrives, and not tape. Any media you leave untouched past its data retention period, will lose data.

What you need is to check every copy of your data for any sign of degradation, and replace it with a fresh copy as soon as, or before, it begins to fail. Tape may give you the most time between checks, but it doesn't change the fact that data you forget about is data you will lose.

Re:Thumb Drives

[ToasterMonkey]

I looked into this because I once wondered why no one uses thumb drives for backup. Thumb drives are reasonably safe for a couple years but after that many can degrade. I saw many sites indicating that flash is not a safe media. This caused me to wonder what they did different in SSD technology. The only safe media that I know of is tape.

Tape is some of a myth.

The only safe media, is that which you keep copying before it deteriorates. Not HDDs, not SSDs, not CDs, not thumbdrives, and not tape. Any media you leave untouched past its data retention period, will lose data.

What you need is to check every copy of your data for any sign of degradation, and replace it with a fresh copy as soon as, or before, it begins to fail. Tape may give you the most time between checks, but it doesn't change the fact that data you forget about is data you will lose.

You're talking about archives I think, and the previous guy was talking about long term backups.

An archive might be the only copy of your data, and everything you said applies. But backups only have to live as long as your retention period, so once you meet that requirement, you're set.

The biggest myth is that backups or archives are simple :\

Re: Thumb Drives

[Anonymous Coward]

I actually transferred a bunch of mp3s to a cheap USB stick for car use years back (car radio has usb socket). In car 24/7, all weather (full sun on a hot day means at least 55C I'd guess), used once in a blue moon, often not for months on end, still working after all that time.

Actually the whole article reminds me of the olde linux assumption that no-one ever powers down their computer. I often leave my desktop PC physically off for weeks on end at least and don't think it unreasonable to assume the contents of my hdd will still be there when I come back, no matter what the weather.

Re: Thumb Drives

[Agripa]

Older Flash memory is fabricated using larger process sizes leading to longer retention.

Eureka!

[Sertis]

So.. It's better to cook my SSD so that it can retain longer at 30 degrees poweroff? Guess i'll mount it to my cpu's heatpipe next time instead of the drive bay with the fans and good airflow, screw common sense! I guess those thermal pads were actually to help keep our nand toasty instead of cooling them after all!

Re:Eureka!

[hcs_$reboot]

Keep them at 0 K to be sure.

Re:Eureka!

[Anonymous Coward]

Read again - more temperature means less time. So if there's no lower limit, SSD without power will be best stored at 0 Kelvin

Re:Eureka!

[Sertis]

Read again - more temperature means less time. So if there's no lower limit, SSD without power will be best stored at 0 Kelvin

No I read the article. It's related to electron mobility at the time the data is written. If you heat up the SSD after it's turned off, electron mobility increases and you'll get more leakage. But if you wrote data when it was hot, you have a better signal to noise ratio.

no problem

[Black Parrot]

Scotty will have the power back just in time.

Re:no problem

[Paradise Pete]

Scotty will have the power back just in time.

But first he'll tell you it's impossible.

Re: no problem

[Buck Feta]

Oh shit I'm wearing a red shirt today

Re: no problem

[Anonymous Coward]

Be careful, non enterprise drives come without a Scotty.

What it really says...

[dshk]

The relevant table is on 27. page.

In short: if you use the SSD in a cold environment AND store it in hot environment than you may lose data quite quickly. Quicker than two weeks.

Client drives are also affected, but the data loss occurs slighly later. I guess reason of the difference is that enerprise drives assume a higher work temperature.

So the advice is that if you use the SSD in your air conditioned basement in a good case then do not store your SSD on the sun for extended periods.

And no, I do not use spinning media as a backup. I use tapes. Using spinning media for proper backups is almost impossible. See http://www.taobackup.com/

Re:What it really says...

[squiggleslash]

Interesting, because that could affect quite a few businesses in Florida.

1. Business switches to SSDs. Uses them in a 75 degree air conditioned environment.
2. Hurricane.
3. Business unable to get power for a week. Computers down. No power to SSDs. No power for air conditioning. Temperature in office rises to over 100 degrees during the day.

That's a very likely scenario, especially for smaller businesses (but not small) that wouldn't be organized enough to have back-up power or work from home capabilities.

Re:What it really says...

[cfalcon]

If the step 2 of your failure scenario is "hurricane", you...

1)- Have plenty of possible mitigation. Hurricans are trackable and reasonably predictable, and you could load your things into an evac van, or back them up, or have a generator. If you don't have a generator, it's super possible to acquire one- those things are sold by the side of the road after a real storm.
2)- Have a pretty good plan. A hurricane hitting your data center, house, or anything at all is certainly able to destroy it anyway, without a convoluted thing. Additionally, you'll have hurricane insurance, which will cover stuff like this.
3)- Got pretty much smited by a storm god. If that's your failure scenario, that's ok.

Re:What it really says...

[cfalcon]

Also, you'd have to get really unlucky to have 100 degree days right after a hurricane. It can happen, but it's by no means likely.

Re:What it really says...

[camperdave]

The master paused for one minute, then suddenly produced an axe and smashed the novice's disk drive to pieces.

I'll bet the master is also the one who infected the novice's computer with a virus, and set fire to the novice's building.

Re:What it really says...

[squiggleslash]

I suspect you haven't been through many. Essentially, yes, you get some warnings. 9/10 of those warnings end up going nowhere, with the hurricane making landfall 100 miles North or South, or sweeping by and hitting North Carolina.

But be that as it may, most smaller businesses know they're going to be crippled anyway in the event of a major hurricane hitting. They expect to offer little but a skeleton service after the hurricane, if that, and many businesses just close for a week until the power comes back on.

Businesses in that sort of situation would not expect any need to do something special because they've switched to SSDs. SSDs are widely advertized as a drop-in replacement for hard drives based upon similar store technologies to that in your camera or cellphone. Those never lose data, so why would you expect your brand new Dell PC with the super-fast SSD disk to either?

Finally, to answer your other post: 100 degrees indoors in buildings without AC a couple of days after the hurricane is quite normal. We have a home we're refitting as a rental that currently only has the AC active for a couple of hours a night, mostly to maintain humidity. I've been there during the day during the summer and it's been in the high nineties when I entered the home, warmer in than out.

And if you're about to say "But the high 90s isn't 100...", yes, but the house is at least getting some AC.

Re: What it really says...

[Anonymous Coward]

I did tech support for fast food restaurants for a braid time in my life... Even they knew to grab the back up hard drive and keep it with them during a hurricane, any business that is dependant on data and can't manage better than the local Wendy's shouldn't remain in business.

Re:What it really says...

[Anonymous Coward]

But 100F isn't even near 40C. 55C is 130F, and I don't think there are many places where hurricanes are a problem that are likely to reach such temperatures.

Re:What it really says...

[nolife]

Big picture. A business that keeps all it's data in one data center is at risk. That one data center being in a known hurricane zone makes it worse but using SSD is not the tipping point here that suddenly makes it a risky.

 

Re: What it really says...

[Anonymous Coward]

100F is about 38C. What's your standard for "close"?

Re: What it really says...

[Anonymous Coward]

To be fair, the local Wendy's has a multi-million dollar corporation behind it to make suggestions like this.

Re:What it really says...

[Anonymous Coward]

Isn't that JEDEC spec for *accelerated* testing, "This means 1000h stress duration is equivalent to 9 years of use."
The JESD47H.01 (Flash-level spec) test is used to verify that the Flash will indeed retain data for a *minimum* of – 1 year for 100% of max cycle count – 10 years for 10% of max cycle count.

Re: What it really says...

[Anonymous Coward]

less than two Mississippis, what is yours?

Re: What it really says...

[jazzis]

mOD ac UP AS hILARIOUS!

Re:What it really says...

[cfalcon]

Definitely lost a house in a hurricane and lived in hurricane country for years, so yes, I know what I'm talking about.

> Essentially, yes, you get some warnings.

"Some warnings" = Paper talks about it, all over the internet, all over the news, weather station tracks it a huge percent of the time, government issues statements.

You also have a huge window to evacuate in. This is not some sky-is-falling thing- the power of the storms are well known, their trajectory is iffy when they are mid-Atlantic but well known by the time they are going to make land (no, you don't know whether your building will take eyewall or not), etc.

Everyone. Fucking. Knows.

People evacuate. Valuables go with them. Pets go with them. If hurricanes weren't well known and had huge warning, the death tolls would be incredible. If you have enough time to evacuate huge swaths of population, you can grab your fucking SSDs on the way out.

"Businesses in that sort of situation would not expect any need to do something special because they've switched to SSDs."

Sure, and *you wouldn't let your hard drive ride out a storm either*. Certainly not if you give a fuck about the data in it. You can have A ROOF GO AWAY during a storm, and even if you aren't facing that kind of storm, it's reasonable for water to get absolutely EVERYWHERE. So no, you don't assume your HDDs are good to sit, you don't toss them in a little safe and bail, they go in the vehicles. So do the SDDs.

"Finally, to answer your other post: 100 degrees indoors in buildings without AC a couple of days after the hurricane is quite normal. "

It absolutely is not after a hurricane. Most big hurricanes hit in the fall, not the summer. You COULD get an August storm (unlucky), that is powerful enough to wreck the power (unluckIER, as those are normally less powerful), followed, not by cooling rains, but by clear days or other warm weather (even WAY unluckier, as the path of the storm normally will bring in some lesser weather over the next few days). Telling me that it can get past 100 in the tropics, not news. But saying that you that you would be really unlucky to have 100+ degree days after a serious hurricane is absolute fact.

Re:What it really says...

[Lost Race]

Using spinning media for proper backups is almost impossible. See http://www.taobackup.com/

There is nothing in that story to suggest that HDDs are considered inappropriate for backup media. What is your theory? I've used HDDs for deduplicating daily snapshots for the last 15+ years and found them to be every bit as reliable as tapes, and far far easier to use.

Re:What it really says...

[dougmc]

And no, I do not use spinning media as a backup. I use tapes. Using spinning media for proper backups is almost impossible. See http://www.taobackup.com/

Your link doesn't really seem to explain how using "spinning media for proper backups is almost impossible", so you'll either need to point to exactly where it says that, provide some other reference, or expand on that on your own.

Re: What it really says...

[Anonymous Coward]

JEDEC says a day per 5C... so how much better does the situation look with 3-4 more days of time to recover after the storm?

Backup won't help you

[cerberusss]

FTFS:

If you have switched to SSD for either personal or business use, do you follow the recommendation here that spinning-disk media be used as backup as well?

So how do backups help you? Except for ZFS and btrfs (?), no file systems check for data integrity. You're not going to detect the bitrot taking place, and you'll happily send that rotten data to your backup until the corruption is noticed in some other way.

Re:Backup won't help you

[bws111]

The bit rot happens when the drives have been powered off for an extended period. The backups are taken before the power is removed.

Re:Backup won't help you

[Anne Thwacks]

And real backups are taken regularly and retained for a long time. Taking a backup every day and recycling the tape on the 8th day may save money, but won't save your data.

Backup to disk is not much use (TM).

Re:Backup won't help you

[Megol]

Read the parent post again. Repeat until you understand his point.

Re:Backup won't help you

[Anonymous Coward]

So how does just re-reading it help you? Except for some esoteric conlangs(?), no languages have checks for interpretation integrity. Just re-reading it, you're not necessarily going to detect the misinterpretation taking place, and you'll happily send that rotten interpretation to your brain until the misunderstanding is pointed out in some other way.

Re:Backup won't help you

[bws111]

I read it multiple times. Still can't see the point.

While the system is running you are making backups and no data is being lost. If your SSD is powered off for an extended period it starts to lose data. If you have any sort of reasonable data management you would now assume the data on the SSD to be unreliable and restore the backups before it is used. What, exactly, is the problem? Or do you think 'bring it online and wait til someone complains before restoring backups' is a reasonable data management technique?

Re:Backup won't help you

[netlag1]

The SSD will detect bit rot the same way a HDD does, they use ECC

Re:Backup won't help you

[Anonymous Coward]

Point: Backups should be taken *after* power is removed, not before?

Re:Backup won't help you

[jarfil]

Sometimes a HDD/SSD will detect bit rot... and fix it using its ECC, only the result will still be wrong. That's when btrfs checksums kick in to pull a copy of the original block from a RAID drive.

Re:Backup won't help you

[Rockoon]

He thinks that adding a few more bits to a byte is magical.

Re:Backup won't help you

[netlag1]

From the Intel data sheet on their series 530 SSD: "Uncorrectable Bit Error Rate (UBER): 1 sector per 10^16 bits read". So it's extremely unlikely a rotted bit will go unnoticed. This is even better than most HDDs, which are 1 per 10^15 on enterprise drives and 10^14 on consumer drives

Real enterprise has not gone to SSD

[fluffernutter]

It's too expensive to buy at that volume and not yet proven.

Re:Real enterprise has not gone to SSD

[fluffernutter]

To the people saying 'no datacenter goes without power for a week'. I know one that did due to a natural disaster. THey spent 1000's of hours bringing that datacenter up. Do you think they would have wanted to have been messing around reconfiguring raid arrays and rebuilding on top of that? It is not just the data you lose, it is the configuration.

Re:Real enterprise has not gone to SSD

[Anonymous Coward]

If you own a datacenter that loses power for a week, you don't own a datacenter. If you are an engineer that is waiting for 5 year old technology to be proven, you don't work in the enterprise.
If you are somone that once aspired to principle and logical thought but have somehow succumbed to fraudulent misrepresentation, penny pinching and bit rot of morality, then you are one of us. Welcome to the "enterprise"

Re:Real enterprise has not gone to SSD

[Anonymous Coward]

It's too expensive to buy at that volume and not yet proven.

2012 called. http://www.wired.com/2012/06/f...

Re:Real enterprise has not gone to SSD

[jazzis]

Like this? http://en.wikipedia.org/wiki/E... Major dumb AC!~

Re:Real enterprise has not gone to SSD

[Anonymous Coward]

Like this?
http://en.wikipedia.org/wiki/E...

Major dumb AC!~

Sandy has nothing to do with a data center losing power for a week. Data centers have generators and fuel dumps to keep running, and the financial resources to get fuel shipped in.

9/11 might be a valid link, WTC collapsing is tough to mitigate, but that was a decade ago. Class 1 data centers have geographically diverse "hot sites" in case the primary site is lost.

Re:Real enterprise has not gone to SSD

[jazzis]

http://www.infoworld.com/artic... Get a Clue AC. Everything is not as clear cut as you believe. Think Wall Street. Data centers were flooded dumb ass. Onsite backup is no good if under water.

Re:Real enterprise has not gone to SSD

[oobayly]

Which is why you don't keep your data centre in the basement, in fact nothing of importance should be kept in the basement. Sandy and Fukushima are prime examples of that.

On this site there was a report after Sandy from a guy who literally had his team carry fuel up flights of stairs to the generators - the data centre and the generators were out of harms way, but the fuel was in the basement, along with the pumps. I'm guessing that storing fuel half way up a building is frowned upon. I don't know the feasibility of having the pumps beside the generators - it'd be a bitch to prime them, but in theory once the generator is running it doesn't matter if your fuel bunker is 6ft under water.

Re:Real enterprise has not gone to SSD

[Anonymous Coward]

9/11 might be a valid link, WTC collapsing is tough to mitigate, but that was a decade ago. Class 1 data centers have geographically diverse "hot sites" in case the primary site is lost.

Get a Clue AC.
Data centers were flooded dumb ass. Onsite backup is no good if under water.

Get a Clue, Jazzis. Do you even know what the phrase "geographically diverse" means? Hint: It doesn't mean putting the backup tape on top of the server you just backed up. For an extra hint, read the rest of the original post's sentence.

My strategy

[vikingpower]

I develop on a Lenovo laptop, which ( only ) has an SSD. Anytime I get something done I consider of "historic importance" ( which may simply be simply be a file added to a project, a new method added to a class, an enum having gotten a new value, an interface's signature changed ) I do a commit to version control, with a relevant comment. Version control lives on a server with spinning disks ( RAID-5 ). That server "backs up" to another server - with spinning disks - by rsync, once per hour. THAT server gets backed up once a week.

The build server uses two SSDs in RAID-0, and gets its input from the version control server.

So my single point of failure is the VC server's RAID controller.... Yes, it would be better to have that RAID-5 array mirrored, by RAID-1. to another RAID-5 array. It would take an entirely new server to realize that, sadly enough. I still have to come up with a better idea, though. Not being dependent, in any way, on SSDs for business continuity is already an achievement in and by itself. Or so I believe.

It boils down to

1) not being dependent upon SSDs

2)Moving dependency upon SSDs away to dependency upon other hardware, namely RAID controllers....

Re:My strategy

[Anonymous Coward]

While you are thinking about failure modes, also consider software malfunctions and malicious attack... In this chain of client-server-server-backup, how much privilege does an upstream element have to initiate IO on the next tier? If your development machine becomes compromised, can an attacker leap-frog the same path using your (stolen) credentials and get to the point of overwriting your version control repository or second server tier of backups? Can a bit of social engineering bypass all of this and damage your only real backups, i.e. those supporting data recovery to a point before the attacker took over your systems?

I'm not saying your backup plan is worthless, but once you have something as complex as you describe, it is worth thinking about these other risks when considering data loss scenarios. For anything network-connected, a virus/script-kiddie/ex-employee attack may be more likely than many of the other data-loss causes like correlated hardware failure, theft, fire, or natural disaster. For example, I've seen a number of people who deploy a big NAS box and call it a near-line backup archive, but they do it by allowing dozens or hundreds of client machines to write their backups directly, with no protection in place to prevent a rogue client from deleting or modifying all of those supposed backup files for the whole organization.

This is the kind of risk analysis that leads people to take hard line stances about off-line and off-site tape archives being the only "real" backups. It takes much more concerted effort to destroy all point-in-time recovery options if they are stored in media that is physically removed from any read/write mechanism and replicated to different cold storage locations, so a single social engineering/burglary/arson action cannot sweep up all the copies either.

This will solve one problem of security...

[Anonymous Coward]

If the drive goes bad, remove it, put it in the oven and cook @ 55C for a couple of weeks and all the data will be effectively erased. Cool.

Re:This will solve one problem of security...

[redback]

or drill a hole in it

Re:This will solve one problem of security...

[Zero__Kelvin]

You are thinking of traditional Disks. A hole in an SSD is unlikely to do much to keep a determined and competent attacker from accessing most or all of the data on the drive.

Re:This will solve one problem of security...

[netlag1]

A microwave oven for just a few seconds will do it

Re:This will solve one problem of security...

[edmudama]

one hole per package of NAND will be sufficient

Scenario

[Kjella]

Bring laptop with SSD to Death Valley, leave it in the car stuck in the sun and go hiking. How long until your data is in trouble? However, I just looked at the specs for the Samsung 840 EVO, since it was the first to pop up:

Temperature
Operating: 0C to 70C
Non-Operating: -55C to 95C

I would assume the 95C is with data? It would be a rather small caveat if the drive survived but your data was fried.

Re:Scenario

[Anonymous Coward]

You know what the say about when you assume..

That 95 C figure looks suspiciously like the standard specification of non-operating temperature for electronics in general - That is, the temperature at which capacitors start going bad and solder joints start going soft.

I very much doubt it takes the longevity of the data stored on the drive into account.

Re:Scenario

[Anonymous Coward]

The data may very well be gone if you leave it for an hour or so at 95C (not completely gone, but corrupted).

I had this by accident. A USB stick was left in my pocket during a hot wash, the washed stick still worked fine, but most files could not be opened anymore.

After format it seemed to be okay again.

Re:Scenario

[Anonymous Coward]

I did exactly this (death valley, laptop in car). Laptop and SSD (with no problems at all) lives just fine---it was hot to the touch after a long day hike, but worked just fine.

mine seems OK

[Anonymous Coward]

I have one that I use in the trunk of my car. I had a problem with another part of it and it didn't get powered up for a long time and still works fine as my carputer's boot drive.

Real world use cases needed

[ITRambo]

I skimmed though Seagate's paper. At first glance, it appears to have a hidden agenda. That is to make hard drives preferable to solid state drives for everyday use, as Seagate primarily sells hard drives. For long term storage that is what is generally done anyway due to cost, along with tape. For performance, 120+ GB SSD's cannot be beaten, one on one, by any hard drive. Does anyone know of data that indicates that (non-OCZ) drives lose data over time during use, not when powered off for many months after use? Seagate's paper may not be useful if it doesn't correspond to real world use.

M-Discs?

[Anonymous Coward]

Is anyone using M-Discs for backup? They got a lot of press when first released, and seem to still be an option, but I see very little discussion of them.

Re:M-Discs?

[damnbunni]

They're too expensive and inconvenient to back up any serious amount of data to.

If you have some personal data, photos, whatever you want to save, they're fine for that, but it just takes too many discs to back up a goodly chunk of things.

(I use plain BD-R, not M-disc, but when I wanted to back up some things just-in-case before working on my backup drive, it took me all day to write about 30 discs. If I wasn't doing something else at the time and just swapping as needed, it would have been horribly frustrating.)

Backups

[Rich0]

If you have switched to SSD for either personal or business use, do you follow the recommendation here that spinning-disk media be used as backup as well?

First, anything stored on any kind of drive should be backed up if you care about it.

Second, if you do backup, who backs up to SSD? You don't need frequent fast random-access on backups, and SSD is about the most expensive storage technology around per-GB. Anybody doing bulk storage is going to be doing it on either hard disks, tape, or something optical.

So, if you're backing your data up, you'll be backing it up to something safe most likely.

Of course, this does bring up the need for the ability to verify the integrity of your data at-rest, and right now I'd say ZFS/btrfs are the best way of accomplishing this. You could also do hashing above the filesystem layer, but that requires a lot of overhead if your files change frequently. If your files don't change much than something like tripwire would be fine. You'd want to run that more often than you rotate your backup media so that you don't discard the last-known-good version of anything.

Re:Backups

[Rockoon]

The typical home user that does backups doesnt need much.

I suggest that the vast majority of backups by home users today are on thumb drives and/or flash cards. Not the greatest method by a long shot, but you can't expect someone with only a couple gigs of photos and music to invest in a real backup solution.

Re:Backups

[Anonymous Coward]

The typical home user that does backups doesnt need much. I suggest that the vast majority of backups by home users today are on thumb drives and/or flash cards. Not the greatest method by a long shot, but you can't expect someone with only a couple gigs of photos and music to invest in a real backup solution.

External USB HDs are cheap these days. I just saw a 1TB USB drive in expensive Australia for $89 AUD, you could probably halve that price in the US.

Re:Backups

[m.dillon]

No, its a stupid recommendation. Spinning rust doesn't last very long on a shelf. It will rapidly go bad mechanically if you keep switching between shelf and active. SSDs are far superior and data retention is going to remain very high until they really dig into their durability. If you still care, there's no reason why you can't just leave them disconnected from a computer but still powered... they eat no real current compared to a hard drive. SSD-based data retention should be 30+ years if left powered... impossible to test as yet :-)... but no reason why not.

However, for backup purposes there is still an issue of cost. Using SSDs for bulk backup storage can be expensive... it wouldn't matter for a big business so much but cost can be a big issue for individual users.

SSDs don't go bad the way HDDs do. With a HDD maximum reasonably-safe life is 3 years whether powered or not (and swapping between powered and shelf will radically reduce its durability). With a SSD only durability really matters. A business can easily justify buying the required SSD storage in bulk with a marginal cost calculation, but it might be too big a hunk of change for an individual.

Personally speaking I still use HDDs for my backups, for reasons of cost, but I expect in the next few years that will change as SSD prices continue to drop. I just bumped up from 2TB x 3 (active, on-site backup, off-site backup) to 4TB x 3. My storage needs are going up more slowly than the technology is dropping in price. The two will meet in a few years and I'll be 100% SSDs. I'm already 100% SSDs for everything else. No point even contemplating a HDD any more except for bulk backup storage or software test rigs.

-Matt

Re:Backups

[Rich0]

Using SSDs for bulk backup storage can be expensive... it wouldn't matter for a big business so much but cost can be a big issue for individual users.

Cost matters for ANY user.

If it supports a revenue-generating activity a business is more likely to be willing to invest in something. That doesn't mean they're going to buy something other than the most cost-effective solution.

If you only have a little data to store for a long time, then archival-quality DVDs are going to be your cheapest option I would think. If you're doing short-term disaster-recovery backups (think rotating over the span of a month or something like that), then online backups to spinning hard drives at a remote datacenter are probably your best option - and for that you're going to have raid/etc and the drives won't be powered off. For long-term backups of large amounts of data (think rotation after many months) tape is probably your best option.

Why would anybody do long-term backups to SSDs? Backups typically involve sequential read/write, and SSDs are no faster than that than any of the competing options. They're FAR more expensive per GB stored.

Re:Backups

[Rich0]

The typical home user that does backups doesnt need much.

I suggest that the vast majority of backups by home users today are on thumb drives and/or flash cards. Not the greatest method by a long shot, but you can't expect someone with only a couple gigs of photos and music to invest in a real backup solution.

Anybody with those kinds of needs should just be using a service like backblaze, carbonite, dropbox, google drive, and so on.

Some of those are free, some cost money, and any of them are going to be more effective than your thumb drive that burns up in the same fire as your PC.

What I want to know is

[Anonymous Coward]

as a regular computer user buying consumer-grade SSDs, what are the chances of my SSD losing data if I power everything off and go on a ten day trip?

Re:What I want to know is

[Anonymous Coward]

Seriously. :D Relax a bit, guys. That reminds me of the guys who always recommended to avoid "wearing" the SSD by putting the browser cache in ramdisk and not swapping on SSD. And then we would see the Tech Report test in which they grind in petabytes of data.

Well, to answer your question. The chance is 0. People would already be talking about it in hordes if it happened like that.

Re:What I want to know is

[edmudama]

The chance is the same AFR of the rest of the product, but yes, it's very small.

Your worst case is that you cycle your SSD to 100% of its capability (which basically no user does anyway) inside a freezer, then put it on your dashboard as you park your black-on-black sports car in death valley for a 6 month hiking trip.

If you're not doing all 3 of those things simultaneously I wouldn't worry.

Now I'm not as excited about Uber as I used to be.

[__aaltlg1547]

Slide 10:

Unrecoverable Bit Error Ratio = ( number of data errors ) / ( number of bits read )

Does anyone still return faulty drives?

[Anonymous Brave Guy]

You should be leasing it via a support contract: the premium is with a tech on the other end of the phone who's out in a couple hours to replace a dud drive and have your RAID rebuilt before the day ends

I always think the idea of giving back a storage device that has had real data on it under some long-term warranty or rapid on-site service agreement is mostly marketing spin.

Every company I work with has a simple policy on this, for basic security/privacy reasons: a drive that is DOA can go back, but anything that has ever been touched by real data is written off and securely destroyed. Any warranty longer than a few days is therefore worthless to us, as is any rapid service support contract if all it's going to do is swap the failed drive out for a new one.

Obviously this is only applicable to drives where recovering data is potentially possible. If everything (including things like swap space) was encrypted before it hit the drive's controller then returning the unit is in theory an option. But usually it's not worth the trouble to verify that this was the case throughout whatever working life any drive had, and permanent encryption introduces a significant performance hit.

I've seen enterprise support contracts from major brands that specifically say a failed drive won't have to be returned for exactly these reasons.

Re:Does anyone still return faulty drives?

[lgftsa]

All my HP FC SAN arrays have DMR support for the warranty period and again when under maintenance. On a 25 drive chassis, with 900Gb 2.5" SAS drives, it's about 5% of the total cost. All have dual controllers and power supplies, of course.

The disks are arranged as 4x 6-disk RAID6 sets, then presented to VMWare as 4x 3.6Tb VMFS disks.

The 25th disk? It goes on the shelf and is the cold spare for that array. When a disk fails, it goes in straight away and a phone call to HP sees a new disk is sent out overnight.

DMR: Damaged Media Retention. We send a load of old server and dektop media, plus USBs, etc, to a rated destruction facility once or twice a year. Yes, it will shred.

Re:Does anyone still return faulty drives?

[Dogers]

You DO know that the big vendors have options that let you keep the drive instead of requiring you to return it?

Re:Does anyone still return faulty drives?

[Anonymous Brave Guy]

Yes, I do. Did you actually read my post to the end before replying?

Re:Does anyone still return faulty drives?

[Dogers]

Okay, mayhaps I tldr, soz :)

Why not use those contracts, since I've still got your attention?

Re:Does anyone still return faulty drives?

[Anonymous Brave Guy]

Just to be clear, I've nothing against those contracts at all. On the contrary, I think they are the only sensible policy for most drives these days.

I realised that one of my earlier sentences was missing an important comma, but the intended meaning was that any warranty or support contract that only swaps out a failed drive (as distinct from providing a new one but still allowing retention of the failed one) has very limited value to anyone I work with these days. There is just too much risk of data leakage, from commercially sensitive plans (direct economic damage) through personal data (damaging to staff/customers, and regulatory/legal issues for the company) to financial data (potentially very serious regulatory issues).

Personally, I also securely destroy any out-of-use drives I've had in devices at home as well, for similar reasons. The idea that I'd voluntarily send something like a laptop that's had financial records or personal e-mails on it off to some kid at a PC store to swap out a failed drive is just crazy to me.

what about new ssd's that you buy?

[Anonymous Coward]

I just bought a new SSD from newegg.com. I don't know how long it's been sitting in their inventory but I could easily see it having sat around for months on end. Does this mean that I should assume I'm going to have lots of bad "sectors" on it immediately upon receipt?

Re:what about new ssd's that you buy?

[jones_supa]

Generally storage devices are filled with zeroes. But if it also uses inverse logic (charge means 0, no charge means 1) then some bit could indeed be flipped. This is just speculation though. In either case, when you format the disk, those rubbish bits are ignored anyway. They will eventually be refreshed as well.

A Data Point

[PRMan]

As a data point, at work I took an 80GB Intel SSD from the IT guys' desk drawer (I needed it for a project and that's all he had available as he had a bulk order out). He told me it would probably be blank, since he hadn't used it in over a year. I found a whole PC worth of data on there which I had to format. He was surprised and until reading this article, I really didn't understand why. At least in my case, everything was there and could be read just fine, over a year after it had been written. It was in a 70F office both while in use and in storage.

Charge trap vs floating gate NAND...

[AcquaCow]

Newer 3D NAND is using a charge trap design which basically solves the electron leakage issue found with the older floating gate NAND...

Also, the move to the newer 3D NAND brings us back up to 40nm processes vs the 10nm gates we are currently working with, allowing for much better reliability.

Disclaimer: I've been selling enterprise flash storage for the last 6 years.

Re:Charge trap vs floating gate NAND...

[TheDarkMaster]

Interesting. Do you know which SSDs are coming with 3D NAND?

Re:Charge trap vs floating gate NAND...

[AcquaCow]

Pretty much all of them as we move into 3D NAND... I know samsung has one out already as the 850 EVO... everyone should be right behind them with similar products.

Phew

[Limitless_Potential]

I knew there was a reason I turned my computer on every day for next to no reason. my data is always gonna be safe!

Re:Phew

[jones_supa]

Remember that merely turning it on is not enough. You have to keep it on long enough to give time for the SSD's internal refresh cycle to walk through all of the cells.

hum

[Anonymous Coward]

No longer use WD and Seagate which have a short life span. I switched 5 years ago to Hitachi brand drives and it has been running without issues.

"Cap based" storage

[bagofbeans]

You can make reliable fast access NVM using DRAM plus battery or cap based backup to run the refresh engine during power-off. So not complete nonsense.

Re:"Cap based" storage

[gweihir]

Complete nonsense for almost all scenarios then. You can buy these, but they are extremely expensive and usually come as a 19" unit. And complete nonsense because these are not what is called "SSD" today.

Seriously, you can nit-pick all day, the fact of the matter is that the "contribution" by luther349 was of negative value and I pointed that out.

author's interests

[etinin]

They must be talking about their own low quality SSDs. For online storage, the crappiest Samsung SSD will probably outlast your data. The best ones will outlast any old fashioned hd out there. As for offline storage, if you won't touch the data for months, using HDs is just stupid. You're better off with tapes.

archival optical

[drinkypoo]

There are some pretty fancy optical discs which are supposed to last for decades if kept in a reasonable environment, and which are written in ordinary drives. They're a cheap intermediate step before tape.

Re:archival optical

[Cramer]

Depends on the quality of the recordable optical media. In fact, the same is true of pressed CD/DVD/Bluray.

There's no news here.

[m.dillon]

These tests explicitly state that the SSD is rewritten until it reaches its endurance rating before the retention test is done. At that point the flash in a consumer would not be expected to retain data unpowered for more than 1 year.

If you write your data to a fresh SSD once, multiply the number by at least 10.

-Matt

NOT consumer drives

[Sarusa]

From TFA - consumer SSDs can expect 2 years, which is better than lots of HDs which probably won't spin up. Enterprise SSDs are faster but more ephemeral.

On the other hand I wouldn't count on this - cell drift is what causes the Samsung 840 slowdown after just a month.

And yes, I back all my stuff up constantly since I don't want to lose it. To platter drives just because it's much cheaper and speed doesn't matter.

Temperatures and data retention

[Anonymous Coward]

Those tables say that for any give 'Storage Temperature' the data retention time increase for higher 'Active Temperature'.

For example, if the storage temp is 25, and active temp is 55, then data retention is 404 weeks. But if the active temp is 40, then data retention is 105 weeks.

Operating your SSD in hotter environments increases data retention time?

How does this work?

Re:Temperatures and data retention

[viperidaenz]

More charge gets transferred to each transistor? The insulating layer doesn't break down as much? Some weird physics effect?

Re:Temperatures and data retention

[Anonymous Coward]

The insulating layers work less well as temperature increases, so yeah, maybe some kind of temperature compensation going on in the write cycle.

That was my first guess but it's very much a guess.

But if they're pumping in more charge at higher temps (because the charge is leaking out faster) it seems like there are going to be issues when you're reading the cell. IOW, if you pump in different levels of charge at different operating temps, then a bit written at one temperature might not read as the same value when read at a different temperature.

Backup

[Jatheon Technologies]

I do backup of course on my 1 TB Hitachi. But, this thing about SSD and data loss sounds really unbelievable. That means, i have Samsung 840 EVO, and when i go on vacation for 10 or 15 days, i won't be able to boot my Win7 when i get back to my home because it was temperature difference meanwhile while i was away. No logic at all but we will see...

I write my data on basalt

[vandamme]

It's good for thousands of years, even at temperatures in the thousands of degrees. Although the density is a problem.

ZOMG, don't store SSDs at 131F, OR ELSE!!!

[Anonymous Coward]

If you store your SSDs at 131F for a week and lose data, you only have yourself to blame. Consumer SSDs, per the article, will retain data for up to 1 year when stored without power under NORMAL circumstances. Enterprise SSDs only retain for 3 months which was a little surprising to read, I would have expected them to store data longer not shorter.

Lesson to be learned: do not use SSDs for archiving data, use them only for active boot/application drive, keep backups in different formats.

Nightly backups

[MoarSauce123]

I pull backups from my SSD every night to a network share. Using the free version of Macrium Reflect....excellent tool by the way. The file server runs a script that drops any backup older than a week. I have a 3TB drive for quick access archiving, some of the more important files get backed up to an enterprise grade SCSI drive and then I use 'abandoned' hard drives (anything that is too small to occupy a slot in a live system) for additional backup/archiving offsite.