hoakley December 13, 2022 Macs, Technology

Estimating the remaining life of a Mac’s internal SSD

Last week I looked at how you can estimate the working life of an SSD using a measure of the total amount of data written to it, together with its TBW, the manufacturer’s estimate of the total that can be written to the SSD before failure is expected. This article updates and extends that for Apple’s internal SSDs in particular.

In the distant past, Apple has offered some models with internal SATA SSDs, but all more recent SSDs fitted by Apple are NVMe, which support a different set of SMART indicators. For my previous article, I referred to Wikipedia’s account of those indicators, particularly indicator 07 Data Units Written, explained there as being the “number of 512-byte data units the host has written to the controller.”

That isn’t correct. According to the NVMe Base Specification version 2.0, Data Units Written “contains the number of 512 byte data units the host has written to the controller as part of processing a User Data Out Command; this value does not include metadata. This value is reported in thousands (i.e., a value of 1 corresponds to 1,000 units of 512 bytes written) and is rounded up (e.g., one indicates that the number of 512 byte data units written is from 1 to 1,000, three indicates that the number of 512 byte data units written is from 2,001 to 3,000).”

Thus to calculate the Data Units Written in TB, take the value given for NVMe indicator 07, multiply it by 512, then divide it by 1,000,000,000.

However, Data Units Written isn’t a complete account of all data written to the SSD. It excludes metadata and may exclude writes performed by the SSD itself for wear-levelling purposes. For NVMe SSDs that support it, SMART attribute 05 may be more useful, as it gives the manufacturer’s estimate of the percentage of life used. As the NVMe specification explains, this “contains a vendor specific estimate of the percentage of NVM subsystem life used based on the actual usage and the manufacturer’s prediction of NVM life.”

This is the figure given by DriveDx as the SSD Lifetime Left Indicator, which will differ from any estimated from Data Units Written even if a TBW is available for that SSD.

The raw value given for percentage of life used is a true percentage, but can exceed 100% in the event that the SSD continues to function normally beyond the manufacturer’s estimate. For example, if the manufacturer gives a TBW for that SSD of 600 TB, but it has written 900 TB so far without failure, the percentage given would be 150%. The specification even allows percentages greater than 254%, which are all given as the raw value 255 for attribute 05. Unfortunately, there’s no equivalent indicator for ATA SSDs.

Conclusions

When assessing the remaining life of a Mac’s internal SSD, if it’s NVMe, use SMART indicator 05 as the percentage of the life used.
If indicator 05 isn’t available or in doubt, calculate data written from SMART indicator 07, multiplying the raw value by 512 and dividing by 1,000,000,000 to get the result in TB.

37Comments

Add yours

1

Duncan on December 13, 2022 at 3:43 pm

“However, Data Units Written isn’t a complete account of all data written to the SSD. It excludes metadata and may exclude writes performed by the SSD itself for wear-levelling purposes.”

How can that not be included? Do they think those are somehow ‘magical bytes’ that don’t contribute to the wear on storage media? That’s like saying that all the time a vehicle isn’t actually moving but only idling in place doesn’t count towards the engine’s hour meter.

LikeLiked by 1 person
- 2
  
  hoakley on December 13, 2022 at 4:46 pm
  
  I’m sorry, that’s what the standard says.
  Howard.
  
  LikeLike
3

kapitainsky on December 13, 2022 at 9:19 pm

I would still love to see some real life stats how many SSD died because of wear off. I assume every tech dies – good backup is only way:)

LikeLiked by 1 person
- 4
  
  hoakley on December 13, 2022 at 9:21 pm
  
  I’ll leave you to ask Apple for those, then!
  Howard.
  
  LikeLiked by 1 person
  - 5
    
    kapitainsky on December 13, 2022 at 9:25 pm
    
    🙂 yes they have well guarded numbers (telemtrics) – and what we see is marketing fluff.
    
    LikeLiked by 1 person
  - 6
    
    Duncan on December 13, 2022 at 9:46 pm
    
    Conceivably someone could buy a cheap Mac with the smallest internal (soldered) SSD, buy 3-years of AppleCare, and then write to the drive until failure. If that happens during the warranty period then they’d get a new motherboard and a fresh drive out of it.
    
    Kind of a risky test to perform if one doesn’t have the extra money lying around, and it would only provide one data point.
    
    LikeLiked by 1 person
    - 7
      
      kapitainsky on December 13, 2022 at 9:52 pm
      
      100% correct. Apple does not limit warranty based on SSD writes – this is what e.g. any disk manufacturer does – e.g. 3 years/or 300 TB whatever comes first. I can do this with my computer – run some script to write SSD to death. I am sure Apple will replace it. What does it prove?
      
      LikeLiked by 1 person
    - 8
      
      hoakley on December 13, 2022 at 10:31 pm
      
      Interestingly, having just been testing a wide range of external SSDs from all the major manufacturers, none of them limited their warranty by writes. I don’t think that’s possible in EU consumer law, given that there’s no standard way the user can access those measurements. All external SSD warranties are based on years, ranging from 3-5. but no cap on the amount of data written.
      Howard.
      
      LikeLiked by 1 person
    - 9
      
      kapitainsky on December 13, 2022 at 10:04 pm
      
      This is quite old but overall message is still valid:
      
      https://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead/
      
      but somebody tried to wear off SSDs for science:) Result? most drives carry on for very long after manufacturer numbers are reached. It is game of chance obviously companies calculate very carefully to provide written guarantee.
      
      LikeLiked by 1 person
    - 10
      
      hoakley on December 13, 2022 at 10:33 pm
      
      Hmm, although interesting, that’s a very small number of SSDs, and the largest was 256 GB. It’s interesting reading statements though that few users write more than a few TB to an SSD each year – that would appear something of an underestimate for many Mac users now.
      Howard.
      
      LikeLiked by 1 person
    - 11
      
      hoakley on December 13, 2022 at 10:28 pm
      
      The snag is that your repaired Mac would only have the smallest SSD, so wouldn’t be good for much other than these tests.
      Howard.
      
      LikeLiked by 1 person
- 12
  
  name99 on December 14, 2022 at 12:31 am
  
  The honest truth is that we have extremely little USEFUL data about this.
  We (supposedly… I’ve never seen it published…) have a fair amount of data about the behavior of individual flash chips, but that has demonstrated little value when it comes to be behavior of the entire flash drive. Such studies as we do have tend to be irrelevant to the home use case because they concern the very different user environments of data centers.
  
  To get a feel for what’s published, one of the early studies was by Facebook and CMU, 2015, https://users.ece.cmu.edu/~omutlu/pub/flash-memory-failures-in-the-field-at-facebook_sigmetrics15.pdf
  Apart from the fact that this deals with much older flash technologies, the two most notable points are
  – they say NOTHING about actual flash lifetime/”wearout” statistics but various ways they say things suggest they think this factor is essentially meaningless, or at least the way it’s currently recorded and presented is useless
  – they seem to think that of things you can control and monitor, temperature is by far the most important. Of course for the home user this is much less of a problem in that your SSD is probably not writing 24/7 the way it may be at a data center.
  
  On the other side of recency we have something like
  
  Click to access atc22-lu.pdf
  
  which has as its interesting points the claims that
  – for current drives, unlike the past, Write Amplification Factor is NEGATIVELY correlated with failure.
  – failure seems to be Poisson more than anything else (ie more or less constant probability of failure from purchase to the end of time, eg every year of use there’s maybe a 5% probability of failure and that doesn’t change)
  – realistic failures appear to be driven far more by environment (excess temperature or noisy power) than by the data patterns or volume
  – the REAL way that modern SSDs “fail” is “grey failure” which is that they become slower and slower. In a way this is especially good news for the home user (for the data center it’s more tricky) because it means that if it’s the flash itself that is going bad, it will manifest as the drive feeling slower and slower, but with so single point of failure. At some point you’ll just get sick of this and upgrade.
  – “grey failure” has no correlation with anything SMART is measuring, and very rarely ends in outright death.
  
  I’d say this all matches my experiences. I’m now at the point where I’ve probably used maybe 30 SSDs in various ways (inside Macs or as external drives) over 15 or so years. Of that pool
  – two failed outright. The first after perhaps 4 years of aggressive use, the second after perhaps 12 years of less-aggressive use. In both cases these were third party SSDs, and the failure was unrelated to anything SMART talked about, they were just out-of-the-blue total failures (probably controller or DRAM died?)
  – one failure which was “fraud”. The drive was purchased as 4TB drive; it died at the point that more than 2TB was written to it. Probably one of the (all too common) scams whereby a small drive has some attribute faked to appear to be a large drive… Always buy flash drives from a company with an easy return policy, and always, as soon as you get it, FILL the entire drive, make sure what you wrote at the end is readable, wipe it, and do that again.
  
  Of the rest, the worst-case SSD I have is probably my 2012 Fusion Drive iMac, where the internal HD died while the SSD carried on (and still carries on) just fine. I cracked the Fusion drive, put the OS on the 128G SSD and my user environment on an external SSD, and life goes on. That 128G SSD is supposedly at 62% of its lifespan after 10 years of reasonably hard working, swapping and storing /tmp and whatever, and shows no signs of any issues, including grey failure slowing down.
  
  Honestly I think in any normal circumstance, worrying about “SSD wear” is a waste of time. My experience and every paper I have read agree that
  – SSDs fail rarely (ie they are much more reliable than hard drives) AND
  – when they fail, like hard drive failure, it’s generally something you can’t predict and didn’t expect, leading to the drive going from alive to totally dead with no blinken-lights even working (not alive but bad data or alive but can’t interact with one block of flash, or whatever, as the lifetime wear model would predict) AND
  – the majority of realworld SSD “failure” like some (but not the majority) of HD failure, is performance degradation, and you can live with it as long as you want to, until the slowness makes you want to pay for the speed you used to have BUT
  – SSDs are, realistically, so fast that, while this slowdown may matter for data warehouses, it is unlikely to be anything you will encounter in real life. By the time it kicks in (if ever) your Mac will be 10 years old and may have been given away or it’s doing something undemanding rather than acting as your primary machine.
  
  LikeLiked by 2 people
  - 13
    
    hoakley on December 14, 2022 at 9:23 am
    
    So on the basis of that recent paper given at Usenix, and anecdotal experience, you think that the information provided by Samsung and other major manufacturers is incorrect?
    I think I’d expect a little more substantial evidence before rejecting such as generally accepted account of the way that SSDs age and eventually fail.
    The emperor might have no clothes, but the merest glimpse of his flesh isn’t compelling.
    Howard.
    
    LikeLike
    - 14
      
      kapitainsky on December 14, 2022 at 10:06 am
      
      I have checked on Samsung UK and DE websites – write endurance is provided for warranty purpose e.g. “5-Year Limited Warranty or 300 TBW Limited Warranty”. So it has little to do with end of SSD life.
      
      LikeLiked by 1 person
    - 15
      
      hoakley on December 14, 2022 at 9:13 pm
      
      As I pointed out, that isn’t the case for external SSDs. Look at the warranty for Samsung’s T7 Shield, for instance, and there’s no mention of any TBW limit. I think the difference here is that internal SSDs are treated as components for assembly, while finished external storage is a consumer product.
      Howard.
      
      LikeLiked by 1 person
    - 16
      
      name99 on December 14, 2022 at 6:05 pm
      
      I merely gave two papers from a tradition that, if not exactly extensive, has many more examples. You can for example look up the CMU+Facebook paper in Google Scholar then follow the many references to it.
      
      Accelerated aging and what we learn from it have their value, but that value is not total. Specifically, the main thing we seem to have learned from studies of how individual flash chips fail is how to construct very good codes to deal with these failure modes. Which is great! And which is why we now see so much grey failure as the codes have to work harder over time to recover the data (but they do not fail!)
      
      However all the codes cover is the failure modes of individual bits (or strings, or pages) that can be formalized; codes cannot cover failure that results from bad capacitors or solder joints that work loose or broken wire bonds, all the sorts of engineering cost tradeoffs which seem to now be the dominant cause for SSD failures.
      
      This is very similar to the evolution of HDs, the main difference being that the public is less aware of the HD side. Pretty much since the 90s, HD evolution has also been about using stronger and stronger codes so as to allow for smaller and smaller bits. (The basic issue is that
      – the magnetic grains on current media are of random size, orientation and placement SO
      – once the average area of a bit becomes of order the average area of a grain, there’s a good chance that any particular grain may be “misplaced” relative to where it “needs to be” and we need coding to deal with these inevitable non-readable bits.
      This becomes even more of an issue when you start to use overlapping tracks as in shingled media… There are two classic papers by Roger Wood, ex-IBM, that discuss these issues, first at 1 terabit per square inch [randomness of media is the problem] then at 10 terabits per square inch [overlapping tracks and so 2D coding becomes the issue].)
      
      I think it’s mainly history that has not led to any sort of panics on the HD side about “fragile data” in the way that we occasionally see around SSDs. And on current HDs, like SSDs, the primary failure modes are the engineering stuff that can’t easily be modeled and worked around with good codes, not the failure of individual sectors or tracks, which are dealt with, and hidden, by coding and firmware.
      
      An alternative *possible* failure mode for SSDs that’s less likely for HDs is the SSD in a cupboard issue, the SSD that was written then placed in some non-powered-up state for many years before being read. Some have stated concerns about this. I’m unaware the extent to which those concerns are justified; or of any work on this (for obvious reasons, it’s not the sort of thing that Google or Facebook or Ali Baba have much professional interest in tracking).
      
      LikeLiked by 2 people
    - 17
      
      hoakley on December 14, 2022 at 9:36 pm
      
      Thank you.
      I’m amazed then that we still refer to the completely mythical concept of wear-levelling. And that the only well-documented account of testing SSDs to the point of failure (admittedly on smaller SSDs, and seven years ago now) found that they failed when they had reached high TBWs, rather than slowing to a grinding halt.
      Maybe when everyone else has caught up with this revolutionary change in understanding, we’ll be able to assure users that their SSD could be the longest-lasting component in their Mac. I just hope the evidence is more scientific than the papers I’ve been pointed at. That Usenix paper, for example, was excellent for discussion at a conference, where I’m sure it kept attendees talking well into the night. But it has holes so large you can drive a truckful of SSDs through them. A great start, now let’s see the finished study, perhaps. How seldom do we ever see those completed.
      Your final point isn’t, I think, a failure mode at all, but loss of data. And it is generally known, and I have pointed it out in several articles. SSDs are not archival storage, and do need to be powered up periodically or they may lose data. I don’t know of any evidence that it leads to failure of the SSD, though. Do you?
      Howard.
      
      LikeLiked by 1 person
    - 18
      
      name99 on December 14, 2022 at 9:51 pm
      
      “And that the only well-documented account of testing SSDs to the point of failure (admittedly on smaller SSDs, and seven years ago now) found that they failed when they had reached high TBWs, rather than slowing to a grinding halt.”
      
      What was that paper?
      I do suspect (but cannot be sure without having read it) that the points I made are in fact relevant – that error coding back in those days was a lot less strong than it is today, meaning that error code “failure” was a lot more prevalent. There was a time when arguments over how much coding and what sort of codes to use were very much live subjects, discussed at places like the Flash Memory Summit, and this activity was doubtless informed by both lack of historical experience and the new demands of MLC then TLC. But we seem to be past those days, with the type and amount of coding required now pretty well understood.
      
      A secondary issue, particularly back then (and again I can only speculate without having read the paper) is the issue of temperature. There is the general point that temperature hurts electronics of all sorts, including both flash chips and flash drive assemblies; and back then manufacturers were more cavalier than they are today (or to put it differently, better processes have allowed for lower power chips) so drives ran hotter.
      Perhaps even more important, test to failure is likely maintaining the drives at problematically high temperatures 24/7 in a way that doesn’t reflect most use cases.
      This becomes even more detached from reality if the temperatures are deliberately kept high for accelerated failure. As I keep trying to say, accelerated failure and other techniques that track the failure of individual cells are important and have their place (eg for designing error coding); but this type of technique does not provide a good model for the *many other ways* that drives can fail; and these many other ways now seem more important (ie more prevalent) than flash cell failures.
      
      LikeLiked by 1 person
    - 19
      
      hoakley on December 14, 2022 at 10:04 pm
      
      It was the Techreport SATA SSD ‘test to the death’, the final summary being here.
      If you’re trying to investigate failure linked to the number of P/E cycles, using high working temperatures to try to ‘accelerate’ failure is plain dumb. But I’m sure the engineers working for SSD manufacturers are very well aware of that.
      Howard.
      
      LikeLiked by 1 person
    - 20
      
      name99 on December 15, 2022 at 11:09 pm
      
      Thanks for the reference, Howard. That’s my afternoon reading!
      
      BTW have you seen
      https://varnelis.net/on-the-destruction-of-the-kingdom-of-doggerland/
      
      This is an elaborate joke, illustrated with plenty of AI art (as always the fingers are the giveaway in a few pictures) but I think will appeal many of your interests!
      
      LikeLiked by 1 person
    - 21
      
      hoakley on December 16, 2022 at 7:52 am
      
      Thank you – that’s a wonderful article.
      Howard.
      
      LikeLike
22

name99 on December 14, 2022 at 6:06 pm

Yikes, no editing!
one sentence above should be:
And which is why we now see so much grey failure as the codes have to work harder over time to recover the data (but they DO NOT fail!)

LikeLiked by 1 person
- 23
  
  hoakley on December 14, 2022 at 9:39 pm
  
  I apologise, but WP doesn’t trust those who kindly write comments to be able to edit them. I have made that correction for you.
  Howard.
  
  LikeLiked by 1 person
24

baron on December 15, 2022 at 12:45 am

I was disappointed to see that I was unable to install “smartmontools” from the smartmontools-7.3-1.dmg on my Monterey MBP 14″ M1Pro, because the developer was not identified said the dialog.
Have I missed something?

LikeLiked by 1 person
- 25
  
  hoakley on December 15, 2022 at 12:53 pm
  
  This is because neither the installer package nor the disk image has been signed, but the latter has been quarantined.
  Copy the disk image to a different volume and strip all its extended attributes, particularly the quarantine flag and the macl xattr. Then double-click it to mount it, and the package should run fine.
  However, this also works fine if the installer package turns out to be malware – which is why macOS won’t let you use it.
  Howard.
  
  LikeLike
  - 26
    
    baron on December 15, 2022 at 3:17 pm
    
    Thank you for your kind explanation.
    I will give it a try, being careful about where I download from and being reassured by my backups.
    
    LikeLiked by 1 person
    - 27
      
      hoakley on December 15, 2022 at 6:40 pm
      
      I have an article coming in the morning with fuller details.
      Howard.
      
      LikeLike
28

baron on December 15, 2022 at 7:01 pm

Thank you.

LikeLiked by 1 person
29

name99 on December 16, 2022 at 9:27 pm

So after reading the TechReport series, we seem to see very different things in the same words!
Even apart from the fact that the series dates from 8 years and so multiple tech generations ago, the primary message that comes through the series, to my eyes, is just how long-lived flash is! To quote from one article
“You won’t believe how much data can be written to modern SSDs. No, seriously. Our ongoing SSD Endurance Experiment has demonstrated that some consumer-grade drives can withstand over a petabyte of writes before burning out. That’s a hyperbole-worthy total for a class of products typically rated to survive only a few hundred terabytes at most.”

And that’s essentially my point. Whatever legal justification organizations have for stating flash life in various terms (either as warranty or via particular SMART indicators) real-SSDs usually last far far longer than that supposed life (and, as far as I can tell from te trend in papers) that “excess-over-warranty” life is increasing rather than decreasing.

A secondary point is that if you look at the details of each failure (insofar as they are available), again the death pattern is not “SMART indicates that every sector available and has been remapped then you die”. There were some devices that, kinda sorta, followed that pattern (just with the bad smart indicators kicking in much later than expected), but there were also multiple devices that just failed, with no real SMART warning, and no obvious reason.

You can interpret this depending on your predilections! One interpretation is that SMART did in fact warn you (in almost every case) that SOMETHING was going wrong; and if it’s important to you that a drive not fail at a (possibly maximally inconvenient time) you could heed those warnings (and also, probably, give up half or two thirds of the actual working life of the drive…)
A second viewpoint is that, in fact, most of the drives appear to have failed in a “random” way that’s not obviously at all correlated with flash failure. (I’d expect a genuine “flash failure” to give a drive that could still be read from, just not written to. The closest we seem to get to that is the Intel drive, which may, possibly, have failed like that, but the situation is clouded by some strange choices apparently made by Intel firmware, together with what appears to have been a bug in that firmware…)
Certainly I regard it as unproven by this article that SMART statistics of how many writes to a drive are predictive of anything, and I think the author of the article would agree with me.

I am not claiming that flash does NOT wear out!
What I am claiming is that,

(a) in the real world (by multiple studies, and I’d include this as one of them) it does not wear out in a way that can usefully be predicted. You CAN, as a precautionary measure, replace drives after a certain degree of wear; but that’s no different from doing the equivalent thing with an HD (replace after any one sector goes bad, or whatever). Mainly what you are doing is choosing to replace the drive after a certain amount of time, which may be valuable insofar as items within the drive (capacitors apparently being especially problematic) age and eventually die.

(b) it’s unclear that most flash drive deaths (whether or not accurately predicted) are the result of *flash wearing out from too many writes*. Certainly most of the deaths do not match what I would expect such a death to look like.
Of course who knows what the firmware is designed to do in the event of “write to one particular flash chip ALWAYS returns an error”? But I would not expect the designed response to be “drive goes dead, drops off SATA, and becomes invisible to the motherboard and the OS”; which is what most of the failures looked like. To me that looks like failure in some other part of the drive.

Perhaps this is an issue of semantics? If the drive firmware is not robustly written, so that the result after (I don’t know, let’s say 256 failed pages) is that rather than switching to some alternative, slower but more robust, remapping system, the firmware simply calls some sort of abort() routine, after which point it looks like it’s dead, is that “flash was driven to failure” or is that “bad firmware — but there’s no economic incentive to fix these sorts of bugs”? I’d call it the second, perhaps you’d call it the first.

It is interesting that grey failure, while noted by this set of tests, is not considered to be a big deal. That may be a consequence of the testing environment (testing overall bandwidth rather than p99 tail latencies, which is what data centers care about); or it may be that this has become more of an issue is newer faster (and especially enterprise) SSDs that push every aspect of the design a lot more aggressively?

LikeLiked by 1 person
- 30
  
  hoakley on December 16, 2022 at 9:45 pm
  
  So what is your response to someone whose expensive Apple internal SSD reports that 10% of its expected life has been used in the first 6 months? Knowing that the only way that SSD can be replaced is with an exchange logic board, which will cost at least half the price of a replacement Mac?
  I’d be interested to know how you factor in SLC cache use, which many now consider to be a crucial factor in determining the write speeds of high-performance SSDs. There’s an expectation, for instance, that most of the superb performance we see with Apple’s internal SSDs is due to their use of SLC cache. How does that affect ‘wear’ or ‘life’?
  We know that SSDs do ‘wear out’ and fail eventually. Of those fitted inside Macs this year, I expect a minority will still be running well in ten years time. I’m no great fan of SMART indicators, but we have nothing else that we can use to assess the state of our storage. Just telling a user that we don’t know enough about the ageing of their SSD, but it should be fine for a good few years yet, really isn’t going to be useful, is it?
  A lot of this comes down to an actuarial approach, but without a lot of data, you can’t use that. People know that their death can’t be predicted accurately, but the concept of life expectancy remains valuable, and in some instances essential.
  Howard.
  
  LikeLike
  - 31
    
    name99 on December 16, 2022 at 10:14 pm
    
    Once again that’s where we differ. What I’d tell them is that
    (a) SMART attributes, for the most part, have basically zero predictive ability. (Rapid changes in some SMART attributes may have some predictive ability, but that’s a different issue, and SMART “lifetime” or “wear” is not one of those attributes.
    
    (b) some things (like when your computers flash drive will die) simply cannot be predicted by science. You are welcome to visit an astrologer or psychic, but an honest scientist/engineer cannot answer this question. The cost of the computer does not change that. Buildings can cost billions of dollars; that doesn’t make it somehow easier to predict when an earthquake will hit them, no matter how much we’d like to know that.
    
    (c) unlike you, I expect the VAST majority of Apple’s internal SSDs in the current crop of Macs to be running just fine in ten years, and this is something just not worth worrying about. You might as well worry about, say, the fan in that Mac Studio dying, which I suspect statistics will bear me out as being more likely to fail.
    
    We HAVE a good pool of statistics already! Apple has been selling SSD-based macs for over ten years now, and those have all (even the fusion based ones) been supporting VM and /tmp. It has presumably happened, but I am unaware of a single case of an Apple-Internal SSD failure, either personally or reported as an internet rant (as opposed to a variety of other failures I have seen on my personal machines – including multiple internal hard drives, one laptop fan, one iMac GPU card going bad [ultimately bad capacitor…], and one laptop WiFi card breaking its solder connections).
    
    People cannot predict their deaths, yes. But if an average 30yr old is living in terror that they will probably be dead within three years, that person has a skewed view of reality which is making their life worse than it needs to be.
    
    LikeLiked by 1 person
    - 32
      
      hoakley on December 16, 2022 at 10:48 pm
      
      You seem concerned with prediction. I’m not, neither is this article. It refers to estimating the remaining life, which is completely different from predicting failure.
      I’m now 68. No one can predict my death. However, I can look up, given various aspects of my health, lifestyle and location, my life expectancy. That’s the concern. And that’s something we can normally do quite well in the face of great uncertainty. Prediction is completely different.
      I haven’t gone near the use/misuse of SMART indicators as predictors, because it’s so complex, and as you write, prediction isn’t possible with any reasonable reliability.
      We know that SSDs will all eventually fail. One important distinction is whether they simply fail by time, i.e. irrespective of the number of P/E cycles, but just on the basis of chronological age. That TechReport study and many others argue strongly against that. Alternatively, do they fail in relation to the number of P/E cycles or any other measure of ‘wear’ from use? I think there’s pretty strong evidence that’s the case, but maybe you disagree, in which case I’d be interested to know what you think is the wear correlate.
      I’d be very interested to see your statistics. I have absolutely no idea what the AppleCare and other warranty replacement figures are, and I’d love to discover them. After all, if you don’t know how many logic boards have been replaced in a large cohort of Macs, you can’t even start being objective about failure rates. From what you write, you’re referring not to statistics, but to anecdote. And telling people to ignore all they’ve been told in favour of something different, on the strength of anecdote is hardly well-evidenced or credible.
      You still haven’t explained how wear-levelling fits into all this. Is it a big lie, marketing bullshit, or what?
      Howard.
      
      LikeLike
    - 33
      
      name99 on December 16, 2022 at 11:08 pm
      
      I fail to see the difference between “predicting end of life” and “estimating end of life”.
      
      When the tools at hand predict (or estimate) a lifespan that is of order ten x to a hundred x smaller than a realistic lifespan, I fail to see what use that is. If your actuarial tables basically say “most people live longer than a year; often to many many years” what’s the value in talking about human lifespan as “around a year”?
      
      The issue is not that mortality tables are not helpful in suggesting how long a 68yr old man might live; the issue is that you look at the mortality tables and say “I have 20 months left to live” and I say “no, you are misinterpreting the meaning of that abbreviation, in fact the tables say you have 20 years left to live”.
      
      I suspect we have reached an impasse here. For reasons I do not understand, you think that flash wear statistics have some sort of useful value (beyond basic facts we all agree on, like “flash writes eventually kill flash chips”) and I don’t. I’ve tried to give you multiple lines of evidence for my belief, from various different academic papers, to personal experience, to suggesting your personal experience of SSD’s in Macs over at least the past ten years. I’ve even tried to show that the set of techreport articles you consider to be dispositive say something very different from what you seem to believe.
      At this point there’s nothing more to be said. Others can read the discussion and come to their own conclusions.
      
      LikeLiked by 1 person
    - 34
      
      hoakley on December 16, 2022 at 11:26 pm
      
      So the statistics you referred to about Mac SSD failures are?
      Howard
      
      LikeLike
    - 35
      
      name99 on December 16, 2022 at 11:55 pm
      
      Multiple reference, including the very reference that you gave me, the techreport reference, all say the same thing about the lifetime of flash drives.
      
      Why do you imagine that flash inside a *Mac* is somehow a different sort of flash drive from any other flash drive (apart from probably having, on average, slightly higher quality components)?
      
      LikeLiked by 1 person
    - 36
      
      hoakley on December 17, 2022 at 10:36 am
      
      “Why do you imagine that flash inside a *Mac* is somehow a different sort of flash drive from any other flash drive”
      Where do I say that I imagine that?
      Howard.
      
      LikeLike
    - 37
      
      hoakley on December 17, 2022 at 10:42 am
      
      ‘I fail to see the difference between “predicting end of life” and “estimating end of life”.’
      I didn’t use either of those phrases. Where are you quoting them from? The words that I used are: “estimating the remaining life” (which is easy to read from the title of this article) and “predicting failure”, which is derived from your previous arguments, and what you were writing about.
      So why change the words, as they don’t mean the same, do they?
      Howard.
      
      LikeLike

·Comments are closed.

Conclusions

Share this:

Related