[HN Gopher] Is DWPD Still a Useful SSD Spec?
___________________________________________________________________
Is DWPD Still a Useful SSD Spec?
Author : zdw
Score : 44 points
Date : 2025-11-21 15:47 UTC (5 days ago)
(HTM) web link (klarasystems.com)
(TXT) w3m dump (klarasystems.com)
| igtztorrero wrote:
| The most common catastrophic failure you'll see in SSDs: the
| entire drive simply drops off the bus as though it were no longer
| there.
|
| Happened to me last week.
|
| I just put it in a plastic bag into the freezer during 15
| minutes, and works.
|
| I made a copy to my laptop and then install a new server.
|
| But not always works like charms.
|
| Please always have a backup for documents, and a recent snapshot
| for critical systems.
| lvl155 wrote:
| Always make backups to HDD and cloud (and possibly tape if you
| are a data nut).
| zamadatix wrote:
| I don't think one should worry as much about what medias they
| are backing up to as if they are answering the question "does
| my data resiliency match my retention needs".
|
| And regularly test restores actually work, nothing worse than
| thinking you had backups and then they don't restore right.
| serf wrote:
| to be perfectly fair though, this isn't a new failure mode when
| SSDs arrived on the scene.
|
| drive controllers on HDDs just suddenly go to shit and drop off
| buses, too.
|
| I guess the difference being that people expect the HDD to fail
| suddenly whereas with a solid state device most people seem to
| be convinced that the failure will be graceful.
| PunchyHamster wrote:
| We have a fleet of few hundred HDDs that is basically being
| replaced "on next failure" with SSD and that is BY FAR rarer
| on HDDs, maybe one out of 100 "just dies".
|
| Usually it either starts returning media errors, or slows
| down (and if it is not replaced in time, slowing down drive
| usually turns into media error one).
|
| SSDs (at least a big fleet of samsung ones we had) are much
| worse, just off, not even turning readonly. Of course we have
| redundancy so it's not really a problem, but if same happened
| on someone's desktop they'd be screwed if they don't have
| backups.
| toast0 wrote:
| > I guess the difference being that people expect the HDD to
| fail suddenly whereas with a solid state device most people
| seem to be convinced that the failure will be graceful.
|
| This is exactly the opposite of my lived experience. Spinners
| fail more often than SSDs, but I don't remember any sudden
| failures with spinners, as far as I can recall, they all have
| pre-failure indicators, like terrible noises (doesn't help
| for remote disks), SMART indicators, failed read/write on a
| couple sectors here and there, etc. If you don't have
| backups, but you notice in a reasonable amount of time, you
| can salvage most of your data. Certainly, sometimes the
| drives just won't spin up because of a bearing/motor issue;
| but sometimes you can rotate the drive manually to get it
| started and capture some data.
|
| The vast majority of my SSD failures have been disappear from
| the bus; lots of people say they should fail read only, but
| I've not seen it. If you don't have backups, your data is all
| gone.
|
| Perhaps I missed the pre-failure indicators from SMART, but
| it's easier when drives fail but remain available for
| inspection --- look at a healthy drive, look at a failed
| drive, see what's different, look at all your drives, predict
| which one fails next. For drives that disappear, you've got
| to read and collect the stats regularly and then go back and
| see if there was anything... I couldn't find anything
| particularly predictive. I feel disappear from the bus is
| more in the firmware error category vs physical storage
| problem, so there may not be real indications, unless it's a
| power on time based failure...
| jandrese wrote:
| For what it is worth the SMART diagnostics and health
| indicators have rarely been useful for me, either on SSDs
| or HDDs. I don't think I've ever had a SMART health warning
| before a drive dies. Although I did have one drive that
| gave a "This drive is on DEATH'S DOOR! Replace it
| IMMEDIATELY!" error for 3 years before I finally got around
| to replacing it, mostly to avoid having my OS freak out
| every time it booted up.
| toast0 wrote:
| Oh, the overall smart status is mostly useless. But some
| of the individual fields are helpful.
|
| The ones for relocated sectors, pending sectors, etc.
| When those add up to N, it's time to replace and you can
| calibrate that based on your monitoring cycle and backup
| needs. For a look every once in a while, single copy use
| case, I'd replace around 10 sectors; for daily
| monitoring, multiple copies, I'd replace towards 100
| sectors. You probably won't get warranty coverage at
| those numbers though.
|
| Mostly I've only seen the smart status warning fire for
| too many power on hours, which isn't very useful. Power
| on hours isn't a good indicator of impending doom (unless
| there's a firmware error at specific values, which can
| happen for SSDs or spinners)
| seanw444 wrote:
| My experience has been the same. Hard drives fail more
| gracefully than SSDs.
|
| > The vast majority of my SSD failures have been disappear
| from the bus; lots of people say they should fail read
| only, but I've not seen it. If you don't have backups, your
| data is all gone.
|
| I just recovered data a couple weeks ago from my boss's
| SATA SSD that gave out and went read-only.
| jandrese wrote:
| I don't know how true this is, but it seems to me that SSD
| firmware has to be more complex than HDD firmware and I've
| seen far more SSDs die due to firmware failure than HDDs.
| I've seen HDDs with corrupt firmware (junk strings and
| nonsense values in the SMART data for example), but usually
| the drive still reads and writes data. In contrast I've had
| multiple SSDs, often with relatively low power-on hours, just
| suddenly die with no warning. Some of them even show up as a
| completely different (and totally useless) device on the bus.
| Drives with Sandforce controllers used to do this all of the
| time, which was a problem because Sandforce hardware was
| apparently quite affordable and many third party drives used
| their chips.
|
| I have had a few drives go completely read only on me, which
| is always a surprise to the underlying OS when it happens.
| What is interesting is you can't predict when a drive might
| go read-only on you. I've had a system drive that was only a
| couple of years old and running on a lightly loaded system
| claim to have exhausted the write endurance and go read only,
| although to be fair that drive was a throwaway Inland brand
| one I got almost for free at Microcenter.
|
| If you really want to see this happen try setting up a
| Raspberry Pi or similar SBC off of a micro-SD card and leave
| it running for a couple of years. There is a reason people
| who are actually serious about those kinds of setups go to
| great lengths to put the logging on a ramdisk and shut off as
| much stuff as possible that might touch the disk.
| fragmede wrote:
| > it seems to me that SSD firmware has to be more complex
| than HDD firmware
|
| I think they're complicated in different ways. A hard desk
| drive has to have an electromagnet powered up in a motor
| that arm that moves and reads the magnetic balance of the
| part of the drive under the read head and correlate that to
| something? Oh, and there are multiple read heads. Seems
| ridiculously complex!
| jandrese wrote:
| Yet somehow firmware bugs are endemic on SSDs far more
| than they were on HDDs.
| pkaye wrote:
| I worked on SSD firmware for more than a decade from the
| early days of SLC memory to TLC memory. SLC memory was so
| rock solid that you hardly needed any ECC protection. You
| could go months of use without any errors. And the most
| common error was erase error which just means to no longer
| use that back.
|
| But then as the years progressed, the transistors were made
| smaller and MLC and TLC were introduced all to increase
| capacity but it made the NAND worse in every other way like
| endurance, retention, write/erase performance, read
| disturb. It also makes the algorithms and error recovery
| process more complicated.
|
| Another difficult thing is recovering the FTL mapping
| tables from a sudden power loss. Having those power loss
| protection capacitors makes it so much more robust in every
| way. I wish more consumer drives included them. It probably
| just adds $2-3 to the product cost.
| namibj wrote:
| That's kind of that ZNS is for: make the SSD dumb but in
| exchange predictable; let the database on top that already
| uses some type of CoW structure handle the quantization of
| erasure blocks; expose all overprovisioning from the start
| and just give back less usable capacity after an erasure
| block for erased and skip over any read access sized blocks
| that got killed off there when mapping logical addresses to
| physical ones. That has to exist anyways because due yield
| reasons some percentage of blocks is expected dead from the
| factory.
| dale_glass wrote:
| > I just put it in a plastic bag into the freezer during 15
| minutes, and works.
|
| What's that supposed to do for a SSD?
|
| It was a trick for hard disks because on ancient drives the
| heads could get stuck to the platter, and that might help
| sometimes. But even for HDDs that's dubiously useful these
| days.
| ahartmetz wrote:
| Semiconductors generally work better the colder they are.
| Extreme overclockers don't use liquid nitrogen primarily to
| keep chips at room temperature at extreme power consumption,
| but to actually run them at temperatures far below freezing.
| butvacuum wrote:
| Complex issue- analog NAND doesn't work anything like the
| Logic in CPUs.
|
| Far more often it's the act of simply letting a device sit
| unpowered itself that 'fixes' the issue. Speculation on
| what changed invariably goes on indefinitely
| rcxdude wrote:
| It could be due to a dodgy connection - changing temperature
| might make the two halves of a broken conductor touch again.
| ssl-3 wrote:
| > It was a trick for hard disks because on ancient drives the
| heads could get stuck to the platter, and that might help
| sometimes.
|
| Stuck heads were/are part of the freezing trick.
|
| Another other part of that trick has to do with printed
| circuit boards and their myriad of connections -- you know,
| the stuff that both HDDs and SSDs have in common.
|
| Freezing them makes things on the PCB contract, sometimes at
| different rates, and sometimes that change makes things
| better-enough, long-enough to retrieve the data.
|
| I've recovered data from a few (non-ancient) hard drives that
| weren't stuck at all by freezing them. Previous to being
| frozen, they'd spin up fine at room temperature and sometimes
| would even work well-enough to get some data off of them
| (while logging a ton of errors). After being frozen, they
| became much more complacent.
|
| A couple of them would die again after warming back up, and
| only really behaved while they were continuously frozen. But
| that was easy enough, too: Just run the USB cable from the
| adapter through the door seal on the freezer and plug it into
| a laptop.
|
| This would work about the same for an SSD, in that: If it
| helps, then it is helpful.
| Havoc wrote:
| After getting burned by consumer drives I decided it's time for a
| zfs array from used enterprise ssds. Tons of writes on them but
| full mirrored config and zfs is easier to backup so should be ok.
| And the really noisy stuff like logging im just sticking into
| optanes - those are 6+ dwpd depending on model which may as well
| be unlimited for personal use scenarios
| hangonhn wrote:
| Do you just source these from eBay? Any guidelines for what's a
| good used enterprise SSD? I had considered this route after I
| built my ZFS array based on consumer SSDs. The endurance
| numbers on the enterprise drives are just so much higher.
| Havoc wrote:
| Yeah ebay. In general I've found buying enterprise stuff off
| ebay to be quite safe...all the shysters are in consumer
| space
|
| >Any guidelines for what's a good used enterprise SSD?
|
| Look at the sellers other items. You want them to be data-
| center stuff.
|
| Look at how many they have to sell - someone clearing out a
| server won't have 1 drive, they'll have half a dozen plus.
|
| Look for smart data in the post / guaranteed minimum health.
|
| I mostly bought S3500/3600/3700 series intel SSDs. The
| endurance numbers vary so you'll need to look up what you
| find
|
| >The endurance numbers on the enterprise drives are just so
| much higher.
|
| That plus I'm more confident they'll actually hit them
| 0manrho wrote:
| > Any guidelines for what's a good used enterprise SSD
|
| Micron, Samsung and Intel (enterprise, branded DCxxxx) / SK
| Hynix / Solidigm (Intel sold it's SSD business to SK which
| they merged into Solidigm) are the go to's for brands. HGST
| can also be good.
|
| The best guideline is buying from reputable sellers with a
| decent volume of business (eg, >1000 sales with high ratings
| on ebay) that focus on enterprise hardware and have a decent
| return/DOA policy.
|
| You should expect these drives to be partially worn
| (regardless of the SMART data, that often gets wiped) if for
| no other reason than the secure erasure process mandated by a
| lot of org's data security policies resulting in multiple
| intensive disk writes, but also due to actually having been
| used. Drives that have recently been released (within 12
| months, eg Micron 7600) are suspect as that implies there was
| a bad batch or that they were misused - especially if they
| aren't write focused drives. Not uncommon for a medium to
| smaller-end large business pinching pennies and buying the
| wrong drives and then wrecking them and their vendor/VAR
| rejecting warranty claims. That said, that's not always the
| case, it's entirely possible to get perfectly good recently
| made drives from reputable 2nd hand market sellers, just
| don't expect a massive discount in that case.
|
| Otherwise best advice I can give you, is redundancy is your
| friend. If you can't afford buy at least 2 drives for an
| array, you should probably stick to buying new. I've had a
| few lemons over the years, but since availability on the
| second hand market for any given model can be variable and
| you tend to want to build arrays from like-devices, you
| should purchase them with the expectation that at least 1 per
| batch will be bad just to be safe. Worst case scenario you
| end up with an extra drive/hotspare.
| butvacuum wrote:
| And exclude China from the eBay regions. All the drives
| I've had with reset smart data came from China.
|
| I'd rather see 3Pb of 5Pb writes used than an obviously
| pretend 2Gib written.
| 0manrho wrote:
| Disagree. You should never trust SMART data from second
| hand drives, full stop. No matter if it's wiped or not.
|
| If you're US domestic market, then yeah, you can usually
| avoid Chinese vendors. If you're EU or elsewhere, China
| can often be the main/only source of affordable drives vs
| domestic market. Really depends (I don't shop for
| international buds/clients, but I constantly hear about
| how the homelabbers across the pond have significantly
| higher prices/lower availability for surplus enterprise
| gear in general)
|
| Stick to the rules on reliable vendors with a return
| policy, buy in bulk with the expectation that some will
| be bad (not a given, but good to be prepared), and the
| only issue from buying from china is delayed shipping
| times.
| butvacuum wrote:
| It's not about the numbers as much as it is the LIE.
| There's no legitimate reason to take the extra steps to
| wipe a drives internal state clean. Like rolling
| odometers back, it has one purpose: fraud.
| mdtancsa wrote:
| dropping off the bus is the best case fail really. Its more
| annoying when writes become slower than the other disks often
| causing confusing performance profiles of the overall array.
| Having good metrics for each disk (we use telegraf) will help
| flag it early. On my zfs pools, monitoring disk io for each disk,
| smartmon metrics help tease that out. For SSDs probably the worst
| is when there is some firmware bug that triggers on all disks at
| the same time. e.g. the infamous HP SSD Failure at 32,768 Hours
| of Use. Yikes!
| PunchyHamster wrote:
| we had ones that turned into that failure mode at like 80% life
| left. Zero negative SMART metrics, just slowed down.
|
| My hunch is that they don't expose anything because that makes
| it harder to refund on warranty
| mgerdts wrote:
| This article misses several important points.
|
| - Consumer drives like Samsung 980 Pro and WD SN 850 Black use
| TLC as SLC when about 30+% of the drive is erased. At this time
| you a burst write a bit less than 10% of the drive capacity at 5
| GB/s. After that, it slows remarkably. If the filesystem doesn't
| automatically trim free space, the drive will eventually be stuck
| in slow mode all the time.
|
| - Write amplification factor (WAF) is not discussed. Random small
| writes and partial block deletions will trigger garbage
| collection, which ends up rewriting data to reclaim freed space
| in a NAND block.
|
| - A drive with a lot of erased blocks can endure more TBW than
| one that has all user blocks with data. This is because garbage
| collection can be more efficient. Again, enable TRIM on your fs.
|
| - Overprovisioning can be used to increase a drive's TBW. If
| before you write to your 0.3 DWPD 1024 GB drive, you partition it
| so you use only 960 GB, you now have a 1 DWPD drive.
|
| - per the NVMe spec there are indicators of drive health in the
| SMART log page.
|
| - Almost all current datacenter or enterprise drives support an
| OCP SMART log page. This allows you to observe things like the
| write amplification factor (WAF), rereads due to ECC errors, etc.
| Aurornis wrote:
| You're also missing an important factor: Many drives now
| reserve some space that cannot be used by the consumer so they
| have extra space to work with. This is called factory
| overprovisioning.
|
| > - Consumer drives like Samsung 980 Pro and WD SN 850 Black
| use TLC as SLC when about 30+% of the drive is erased. At this
| time you a burst write a bit less than 10% of the drive
| capacity at 5 GB/s. After that, it slows remarkably. If the
| filesystem doesn't automatically trim free space, the drive
| will eventually be stuck in slow mode all the time.
|
| This is true, but despite all of the controversy about this
| feature it's hard to encounter this in practical consumer use
| patterns.
|
| With the 980 Pro 1TB you can write 113GB before it slows down.
| (Source
| https://www.techpowerup.com/review/samsung-980-pro-1-tb-ssd/...
| ) So you need to be able to source that much data from another
| high speed SSD and then fill nearly 1/8th of the drive to
| encounter the slowdown. Even when it slows down you're still
| writing at 1.5GB/sec. Also remember that the drive is factory
| overprovisioned so there is always some amount of space left to
| handle some of this burst writing.
|
| For as much as this fact gets brought up, I doubt most
| consumers ever encounter this condition. Someone who is copying
| very large video files from one drive to another might
| encounter it on certain operations, but even in slow mode
| you're filling the entire drive capacity in under 10 minutes.
| nyrikki wrote:
| > You're also missing an important factor: Many drives now
| reserve some space that cannot be used by the consumer so
| they have extra space to work with. This is called factory
| overprovisioning.
|
| This has always been the case, thus why even a decade ago the
| "pro" drives were odd sizes like 120g vs 128g.
|
| Products like that still exist today and the problem tends to
| show up as drives age and that pool shrinks.
|
| DWPD and TB written like modern consumer drives use are just
| different ways of communicating that contract.
|
| FWIW I'd you do a drive wide discard and then only partition
| 90% of the drive you can dramatically improve the garbage
| collection slowdown on consumer drives.
|
| In the world of ML and containers you can hit that if you say
| have fstrim scheduled once a week to avoid the cost of online
| discards.
|
| I would rather have visibility into the size of the reserve
| space through smart, but I doubt that will happen.
| mgerdts wrote:
| > You're also missing an important factor: Many drives now
| reserve some space that cannot be used by the consumer so
| they have extra space to work with. This is called factory
| overprovisioning.
|
| I think it is safe to say that all drives have this. Refer to
| the available spare field in the SMART log page (likely via
| smartctl -a) to see the percentage of factory overprovisioned
| blocks that are still available.
|
| I hypothesize that as this OP space dwindles writes get
| slower because they are more likely to get bogged down behind
| garbage collection.
|
| > I doubt most consumers ever encounter this condition.
| Someone who is copying very large video files from one drive
| to another might encounter it on certain operations
|
| I agree. I agree so much that I question the assertion that
| drive slowness is a major factor in machines feeling slow. My
| slow laptop is about 5 years old. Firefox spikes to 100+% CPU
| for several seconds on most page loads. The drive is idle
| during that time. I place the vast majority of the blame on
| software bloat.
|
| That said, I am aware of credible assertions that drive wear
| has contributed to measurable regression in VM boot time for
| a certain class of servers I've worked on.
| justinclift wrote:
| > With the 980 Pro 1TB you can write 113GB before it slows
| down.
|
| 113GB is pretty easily reached with video files.
| kvemkon wrote:
| Now that PCIe 5.0 SSDs are available since 6+ months and you
| could backup your SSD with 15 GB/s but:
|
| > you're still writing at 1.5GB/sec.
|
| Except of few seconds at the start, the whole process lasts
| as if you had PCIe 2.0 (15+ years ago). Having so fast SSDs
| there is no chance to make a quick backup/restore. And during
| restore you're second time in a row too slow.
|
| It's crazy that instead of using slow PLC at the time of slow
| PCIe 1.0, back then fast SLC was in use. Now with PCIe 5.0
| when you really need fast SLC, you get slow TLC or very slow
| QLC or even worse PLC coming.
| markhahn wrote:
| Text is wrong about CRCs: everyone uses pretty heavy ECC, so it's
| not just a re-read. This also provides a somewhat graduated
| measure of the block's actual health, so the housekeeping
| firmware can decide whether to stop using the block (ie, move the
| content elsewhere).
|
| I'm also not a fan of buy bigger storage concept, or the
| conspiracy-theory on 480 v 512.
|
| It sure would be nice if when considering a product, you could
| just look at some claimed stats from the vendor about time-
| related degradation, firmware sparing policy, etc. we shouldn't
| have to guess!
| saurik wrote:
| > I'm also not a fan of buy bigger storage concept, or the
| conspiracy-theory on 480 v 512.
|
| I don't understand why this is being called a "conspiracy
| theory"; but, if you want some very concrete evidence that this
| is how they work, a paper was recently published that analyzed
| the behavior and endurance of various SSDs, and the data would
| be very difficult to describe using any other theory than that,
| comparing apples-to-apples on drives that have better write
| endurance, they are merely overprovisioned to allow the wear-
| level algorithm to not cause as much write amplification while
| reorganizing.
|
| https://news.ycombinator.com/item?id=44985619
|
| > _OP on write-intensive SSD._ SSD vendors often offer two
| versions of SSDs with similar hardware specifications, where
| the lower-capacity model is typically marketed as "write-
| optimized" or "mixed-use". One might expect that such write-
| optimized SSDs would demonstrate improved WAF characteristics
| due to specialized internal designs. To investigate this, we
| compared two Micron SSD models: the Micron 7450 PRO, designed
| for "read-intensive" workloads with a capacity of 960 GB, and
| the Micron 7450 MAX, intended for "mixed-use" workloads with a
| capacity of 800 GB. Both SSDs were tested under identical
| workloads and dataset sizes, as shown in Figure 7b. The WAF
| results for both models were identical and closely matched the
| results from the simulator. This suggests that these Micron
| SSDs, despite being marketed for different workloads, are
| essentially identical in performance, with the only difference
| being a larger OP on the "mixed-use" model. For these SSD
| models, there appear to be no other hardware or algorithmic
| improvements. As a result, users can achieve similar
| performance by manually reserving free space on the "read-
| intensive" SSD, offering a practical alternative to purchasing
| the "mixed-use" model.
| justsomehnguy wrote:
| > Is DWPD Still a Useful SSD Spec?
|
| Yes.
|
| You need years from that SSD? Buy a drive with DWPD > 3.
|
| You are a cheap ass and have the money only for a DWPD 0.3 drive?
| Replace it every year.
|
| You are not sure what your usage would be? Over-provision by
| buying a bigger drive than you need.
|
| And while we are at it: no, leaving >= 25% of the drive empty
| _for the drives > 480GB_ is just idiotic. Either buy a bigger
| drive or use a common sense - even 10% of a 480GB drive is 48Gb
| already, for a 2048GB drive it's _204GB_.
| 0manrho wrote:
| > Controller failure: by far the most common catastrophic failure
| you'll see in SSDs
|
| In _consumer_ drives. Often not even a hardware failure, but a
| firmware one, but to most consumers, this is splitting hairs as
| the drive is still "Dead" as the common ingress points to fix
| this are not present/disabled on consumer class drives (thus the
| blurb at the end of that section about physically swapping
| controllers). Also, cell failure is far more prevalent than
| controller failure in instances where the drives lack a DRAM/SLC
| cache (aka transition flash) layer. Controllers still fail, even
| at the hardware level, for enterprise and consumers alike though,
| it's a prevalant issue (pro tip, monitor and rectify the thermals
| and the prevalence of this problem drops significantly)
|
| > Failure to retain charge: typically, only seen in SSDs, thumb
| drives, and similar devices left unpowered for long periods of
| time.
|
| Also happens to flash that see lots of writes, power cycles, or
| frequent significant temperature fluctuations. This is more
| common on portable media (thumb drives) or mobile devices
| (phones, laptops, especially thin ones)
|
| > Now, let's take a look at the DC600M Series 2.5" SATA
| Enterprise SSD datasheet for one of my favorite enterprise-grade
| drives: Kingston's DC600M.
|
| Strange choice of drive but okay, especially considering they
| don't talk about any of it's features that actually make it an
| enterprise version as opposed to their consumer alternatives:
| Power loss protection, Transition flash/DRAM cache, controller
| and diagnostics options, etc etc.
|
| > Although Kingston's DC600M is 3D TLC like Samsung's EVO (and
| newer "Pro") models, it offers nearly double the endurance of
| Samsung's older MLC drives, let alone the cheaper TLC! What
| gives?
|
| For starters the power regulation and delivery circuitry on
| entrprise grade drives tends to be more robust (usually, even on
| a low-end drive like the DC600M), so that those writes that
| _wear_ the cells are much less likely to actually cause wear due
| to out-of-spec voltage /amps. Their flash topology, channels,
| bitwidths, redundancy (for wear levelling/error correction) etc
| etc are also typically significantly improved. all of these
| things are FAR more important than the TLC/SLC/MLC discussion
| they dive into. None of these things are a given just because
| someone brands it an "Enterprise drive" but these are things that
| enterprises are concerned with where consumers typically don't
| often have workloads where such considerations really make a
| meaningful difference and they can just use either DWPD or brute
| force by vastly overbuying capacity to evaluate what works for
| them.
|
| > One might, for example, very confidently expect 20GB per day to
| be written to a LOG vdev in a pool with synchronous NFS exports,
| and therefore spec a tiny 128GB consumer SSD rated for 0.3
| DWPD... On the surface, this seems more than fine:
|
| Perhaps, but let me stop you right there as the math that follows
| is irrelevant for the context presented. You should be asking
| what kind of DRAM/Transition flash (typically SLC if not DRAM) is
| present in the drive and how the controller handles it (also if
| it has PLP) before you ever consider DWPD. If your (S)LOG's
| payloads fit within the controllers cache size, and that's it's
| only meaningful workload then 0.3DWPD is totally fine as the
| actual NAND cells that comprise the _available capacity_ will
| experience _much_ less wear than if there were no cache present
| on the drive.
|
| Furthermore, regardless of specific application, if your
| burstable payloads exceed whatever cache layer your drive can
| handle, you're going to see much more immediate performance
| degradation entirely independent of wear on any of your
| components. This is one area that significantly separates
| consumer flash with enterprise flash, not QLC/TLC/MLC or how many
| 3d stacks of it there are. That stuff IS relevant, but it's
| equally relevant in enterprise and consumer, and is first and
| foremost a function of cost and capacity than endurance,
| performance, or anything else.
|
| This is an example of how DWPD is a generic that can be broadly
| used, but when you get into the specifics of use, can kinda fall
| on it's face.
|
| Thermals are also _very_ important to endurance /wear and
| performance both, and often goes overlooked/misunderstood.
|
| DWPD is not as important as it once was when flash was expensive,
| drive capacity limited, and their was significantly more overhead
| in scaling them up (to vastly oversimplify, a lot less PCIe lanes
| available), but it's still a valuable metric. And like any
| individual metric, in isolation it can only tell you so much, and
| different folks/context will have different constraints and
| needs.
|
| Note, kudos for them bringing it up that not all DWPD is equal.
| Some report DWPD endurance over 3 years instead of 5 to
| artificially inflate their DWPD metric, something to be aware of.
|
| TL;DR: DWPD, IOPs, Capacity and price are all perfectly valid
| ways to evaluate flash drives, especially in the consumer space.
| As your concerns get more specific/demanding/"enterprise", they
| come with more and more caveats/nuance, but that's true of any
| metric for any device tbh.
___________________________________________________________________
(page generated 2025-11-26 23:02 UTC)