tomman |
Posted on 20-03-28, 19:50 (revision 3)
|
Dinosaur
Post: #659 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
Remember SMR (shingled) HDDs? http://wp.xin.at/archives/2633 We thought those were a fad, and the industry would switch back to good ol' PMR while perfecting HAMR in no time (or maybe SSDs would finally beat spinning rust on $/GB, killing those for good), due to the following: - Abysmal write speeds once you've written several GBs to the drive, rendering those unsuitable for anything but archival purposes. - Drives would require special support from OS to deal with their particular layouts in order to minimize the write speeds impact. - More complex firmware architectures, introducing more points of failure (hi Suckgate!) Well, we're in 2020, and while spinning rust has been "going extinct" every single day since someone figured out how to sell mass storage devices built out of flash memory ICs, WD, Suckgate and Toshiba are still cranking them out because there is still money to be made. And most of that money is on -you guessed it!- SMR! Turns out that if you're willing to accept the write speed hits, you can get denser, cheaper HDDs. And that tech, formerly reserved for cold archival server-class drives, has permeated all the way down to Black Friday Amazon specials, aka consumer HDDs. But just like Advanced Format, drives now just sweep all the gory details under the rug, and pretend to be ordinary, no-frills devices - plug wires, write data, neglect backups. Well, almost ordinary: the only hint you get from those that they're not normal drives is a bit on the ATA IDENTIFY structures claiming that those drives support TRIM. Wait, what?! "Butbutbut wasn't TRIM only for solid state cards!?" Not exactly. Due to physics being a bitch, on SMR drives write heads are wider than read heads, this is why write speeds can tank when drive is being filled, as your lone "write a bunch of sectors" operation has become a VERY EXPENSIVE "read a bunch of sectors, including some that are completely unrelated to the ones you're going to write, modify them on memory, then rewrite several tracks at a time". In other words: you've got a really crappy SSD with moving parts, and therefore in order to preserve performance, you must treat those as such - if you're using Windows 7+ or most modern Linux distros, you're set: enable TRIM (periodic is highly recommended over continuous), set your filesystem mount options to "noatime", and try to not do any heavy random write workloads on those drives (I can guess swap space and mailservers/databases are right out of the question). Why am I speaking about SMR today? Because to my hands a brand spankin' new 1TB WD Blue laptop HDD has landed (long short story: $FAMILY_MEMBER went abroad a few months ago, and came back with this gift for me, but I was lazy and the drive remained sealed in its box all those months until this forced solitary confinement due to The Plague™). Cool, I'll finally be able to replace the failed Toshiba 1TB on my Steamlaptop™ (which only works mostly as a Plugging it to my IBM box for the initial torture test reveals another interesting fact: if you try to read drive as-is, it will report ludicrously high read speeds (verify rates up to 270MB/s!). But don't get fooled - that's just one of the SMR perks (more on that later, when we talk about data recovery on those drives), where an empty/security erased/TRIMmed drive actively LIES at you and doesn't actually hit the platters at all, instead it makes up zeroes out of thin air. Once you write actual data to the drive, read speeds fall down to more realistic ~100MB/s (not bad at all for a 5400RPM junk-class drive!), as the drive now knows that you've written stuff to them. Data recovery shops know those drives as "Palmer" (internal WD codename), and they hate them almost as much as modern Suckgates (locked firmware, requires very expensive "factory unlocked PCBs" just to be able to PARTIALLY work around said locks, self-encrypting BS, complex double-stage translator architecture means that a corrupted service area could lead to a drive reading only zeroes instead of actual data or giving read errors. In other words: "engineered for doom"). Soooo.... if you want to avoid SMR, you don't have many options nowadays: - Buy SSDs, or stick to low-capacity older-tech drives - WD Blacks should be PMR, but don't rely on the datasheets alone! As for Toshiba, they still make those ancient 2.5" MQ01s (not MQ04!!!) and rebadged 3.5" Hitachis, which are absolutely SMR-free. - Anything over 12TB (or in the case of 2.5", over 1TB/platter) is SMR for sure. Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
tomman |
Posted on 20-03-31, 01:43 (revision 1)
|
Dinosaur
Post: #660 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
Yup, this drive does indeed support TRIM. blkdiscard is a SCARY command. If you wanted a one-second wipe to Also: for SMR HDDs, a zero-filled drive does read slower than a just-TRIMmed device, despite both read scenarios returning all zeroes. Nice way to fool naive consumers with fake benchmarks ("280MB/s from spinning rust?! MAGIC!!!") Will be deploying this thing to its intended laptop tomorrow, in the meanwhile I'll be performing more testing on this WD UPDATE: Drive now in active service, posting reply from it. I've yet to enlarge partitions (coming up from a 750GB Hitachi 5K750 which is now the backup) ...guys, don't let shingled drives fragment too much - just don't! The write performance hit when defragmenting is so severe that even for a very lightly fragmented drive, your defrag tool will easily stall for some time. For comparison, an ordinary non-SMR drive with 5% fragmentation will defrag in a couple minutes, at most. A SMR drive? Easily 5x that time, if not more. Remember: SMR drives are to be treated like very shitty fake Chinese refurb SSDs. I've enabled fstrim.timer on systemd to manually TRIM once a week - will need to check how to make it run during late nights, as once again the drive will stall beautifully (hope WD hasn't reinvented MCAS!), and I don't want that crap happening whenever I'm actually sit in front of my laptop! Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
Nicholas Steel |
Posted on 20-04-17, 11:45 (revision 3)
|
Post: #356 of 426
Since: 10-30-18 Last post: 499 days Last view: 14 days |
https://www.tomshardware.com/news/sneaky-marketing-toshiba-seagate-wd-smr-drives-without-disclosure Toshiba, Seagate and Western Digital sell SMR HDD's without any indication of the drives using SMR instead of CMR. AMD Ryzen 3700X | MSI Gamer Geforce 1070Ti 8GB | 16GB 3600MHz DDR4 RAM | ASUS Crosshair VIII Hero (WiFi) Motherboard | Windows 10 x64 |
tomman |
Posted on 20-04-18, 12:47
|
Dinosaur
Post: #668 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
My SMR WD is doing fine so far, but then I haven't bothered enlarging partitions or filling it up yet. But yeah, hardware OEMs are a bunch of sleazy scumbags - nothing new here. Buy SSDs, if you can! Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
tomman |
Posted on 20-04-25, 13:13
|
Dinosaur
Post: #674 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
And now Western Digital is doing damage control: https://blog.westerndigital.com/wd-red-nas-drives/ Well, fuck you too, WD. They're even pulling this SMR crap with Black drives, you know, the performance lines! Basically if your drive is not too large or not too small, you will get a SMR unit. If you're in the market for a new laptop HDD, this makes my suggestion easier: AVOID SPINNING RUST... or anything made by former spinning rust OEMs. Mind you, this is not the first time WD is caught pulling a Huawei: most of their recent external drives now hide most of their SMART attribute values (including Power On Hours!) via firmware trickery. Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
wertigon |
Posted on 20-04-28, 08:03
|
Post: #131 of 205
Since: 11-24-18 Last post: 156 days Last view: 27 days |
Speaking of hard drives; looks like soon you will need an NVMe drive to play latest AAA titles since SATA will be too slow in the newest consoles. On the plus side, the only things you will really have loading times for anymore is starting the game and non-graphics assets, since all graphics will be put on a built-in SSD in the GPU card. Imagine 256 GB GPU persistent memory with DDR2 equivalent IO speed... :) https://www.gamesradar.com/ssd-vs-hdd/ Only a decade left on the spinning rust and SATA ports, it seems! |
CaptainJistuce |
Posted on 20-04-28, 10:40
|
Custom title here
Post: #868 of 1164 Since: 10-30-18 Last post: 63 days Last view: 18 hours |
The disk will never die! --- In UTF-16, where available. --- |
tomman |
Posted on 20-04-28, 14:10
|
Dinosaur
Post: #675 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
Weren't Stadia and friends supposed to kill gaming PCs and consoles as we know them? Nothing to install = no fancy SSDs to sell. Ah yeah, the China Plague™ reminded everybody of the fact that streaming is, indeed, a terrible way to enjoy interactive applications when 75% of the planet is locked down inside their homes and therefore every server in the world is forced to ration bandwidth so they can cope with the increased load of people downloading THE WORLD at once. As for SMR: it does have its uses (anything that is "write once, read many" fits the bill perfectly), it's not a universal replacement for all other magnetic recording technologies in the market, no matter how many pennies on the dollar it saves for the beancounters at Western Deceiver, Suckgate and Toshitba. So if you're right now on the market for that NAS expansion, carefully review both the market offers AND your particular use cases. If you're just a data hoarder, SMR is fine. But for high-performance RAID, constant VM thrashing, or simply changing the failed boot drive on $GRANNY's laptop, avoid SMR and go for CMR (if looking for massive storage), or ditch the disk and make the jump to an all-solid lifestyle (if you don't need that much storage) Thankfully I'm not one of those data hoarders anymore (it complicates your backup strategies when you have limited access to storage media, for starters). As for my WD SMR laptop drive: so far I haven't hit the SMR woes during daily usage, but then I don't heavy VM thrashing, or recompile the Linux kernel several times a day. Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
CaptainJistuce |
Posted on 20-04-29, 05:55
|
Custom title here
Post: #869 of 1164 Since: 10-30-18 Last post: 63 days Last view: 18 hours |
It actually occurs to me that shingled recording drives should benefit a lot from developments intended to help flash drives, as they have a simidar weakness(read small chunks, write big chunks). ... Do shingled drives support trim? --- In UTF-16, where available. --- |
Nicholas Steel |
Posted on 20-04-29, 07:16
|
Post: #358 of 426
Since: 10-30-18 Last post: 499 days Last view: 14 days |
Posted by CaptainJistuceYes. AMD Ryzen 3700X | MSI Gamer Geforce 1070Ti 8GB | 16GB 3600MHz DDR4 RAM | ASUS Crosshair VIII Hero (WiFi) Motherboard | Windows 10 x64 |
tomman |
Posted on 20-04-29, 17:12 (revision 1)
|
Dinosaur
Post: #676 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
Posted by CaptainJistuce Quoting myself: Posted by tomman And from the actual drive itself:
Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
CaptainJistuce |
Posted on 20-04-29, 22:49
|
Custom title here
Post: #870 of 1164 Since: 10-30-18 Last post: 63 days Last view: 18 hours |
I feel dumb now. --- In UTF-16, where available. --- |
Nicholas Steel |
Posted on 20-04-30, 03:40
|
Post: #359 of 426
Since: 10-30-18 Last post: 499 days Last view: 14 days |
Western Digital has just now released this new blog post on how SMR works: https://blog.westerndigital.com/dmsmr-device-managed-shingled-magnetic-recording/ AMD Ryzen 3700X | MSI Gamer Geforce 1070Ti 8GB | 16GB 3600MHz DDR4 RAM | ASUS Crosshair VIII Hero (WiFi) Motherboard | Windows 10 x64 |
tomman |
Posted on 20-04-30, 12:37 (revision 7)
|
Dinosaur
Post: #677 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
More damage control. But hey, they didn't forgot to bring the buzzword bingo! Using machine learning and pattern recognition in our system, the drive detects types of data so that we can coalesce similar data types together to maximize performance. They also said how good SMR is for data integrity at the user-area level... Data integrity — Because DMSMR drives put data down sequentially, track ECC (accumulation of parity information on a per-track basis) can be placed at the end of each data track. It enables the drive to correct up to 8K bytes of data per track during read, with near on-the-fly performance. This exceptions-handling capability is especially beneficial in event of shock, fan vibration, grown defects or other less-than-ideal operating environments. This development of data protection power during read comes naturally in a SMR architecture. ...while failing to address the more vulnerable, highly complex firmware architecture that will nuke your data as soon as the LBA translation tables get corrupted, which seems to be a popular issue with some of those WD drives. (For reference: this could happen too to older WD drives, but most of the times it was an easy fix, doable at home even, with free or cheap tools! While the only equivalent for those SMR drives involves the purchase of thousands in American money of Soviet-made software with heavy DRM and -possibly- more $$$ in a new PCB for your drive since HDD OEMs doesn't want people tampering inside their drives since the NSA had their fun with firmware-level spyware, and Suckgate got into the data recovery biz). Oh, that article does mention some WD models that are NOT listed on WD's blog (mainly external drives) At least they're being honest (for once!) with the main disadvantage of SMR: Data movement requires idle time — One of the most discussed about topics in DMSMR drives, is data movement. This is, by LBA Indirection nature, the other side of the same coin. To be able to move the data freely requires the drive to have idle time to perform these tasks in the background. Without the ability to preemptively maintain disk space, the drive, under specific conditions, may take longer to complete a command as its resources dwindle. ...hence reminding you that knowing your workloads is of vital importance when picking your next HDD. I'm not against SMR at all (once again, they have their place in our data hoarder lives). I'm against scumbag companies pretending that everybody of us are dumb (just like dictators and presidents!), while they keep lowering the quality of their products without us realizing all and any possible savings. UPDATE: And now the Toshiba beancounters (which are used to deceiving customers, just ask their nuclear guys) are feeling the pressure, as the company is now disclosing which of their HDD product lines are using SMR: https://toshiba.semicon-storage.com/ap-en/company/news/news-topics/2020/04/storage-20200428-1.html So far Suckgate has yet to make their move (aside of teling Ars that "our NAS drives aren't shingled, no sir!"), and considering how terrible are their product lines (particularly how nasty their mobile drives have become) this isn't surprising at all. UPDATE 2: The Toshiba P300/DT02 series is actually a rebadged Hitachi... kinda, sorta? Well... I thought the WD/Toshiba HDD divestment agreement ended with the DT01 series (which IS still in production - damn, those are probably the LAST GOOD HDD drives still made!), but if you look at the pics, you can see that Toshiba is still evolving that former Hitachi design (which also implies that the WD/HGST merger stuff let them with full IP rights over those designs, instead of just allowing Toshiba to build rebadged drives under license, a-la Excelstor). However, those drives don't seem to be fully Hitachi under the hood: - For starters, HGST didn't used that chassis for their native air-based 6TB drives. - Look closer at the DT02 PCB: that thing looks more like a native Toshiba PCB (barcode label designs are clearly unique to OEMs), and it's less shiny (again, a trait of native Toshiba PCBs) - Firmware revision KQ000A, which is clearly a Toshiba version string (Hitachi ones look like "MXO1A1BB", that is, 8 characters... a trait inherited from the IBM generations) - These drives are SMR, of course. To date, no Hitachi/HGST drive is known to be using SMR (and that includes anything released under the WD brands) ...so, what the hell, Toshiba?! Time to look up some reviews! Of course, being SMR it means you won't be slapping those Half-tachis inside your NAS or DVR anytime soon! Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
tomman |
Posted on 20-05-30, 16:37
|
Dinosaur
Post: #717 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
Some nerds figured out that SMR drives are indeed "unfit for NAS usage" and you should avoid any SMR drive if you plan to deploy them on your NAS - particularly if said specific SMR specimens says "Western Digital Red" and "NASware" on the label. They're just too slow for extended writes of big chunks of data, and most NASes out there just can't wait so they will assume the drive died and mark your RAID as failed. Oh, and your RAID rebuild times go straight to the moon and back. ...then they got the lawyers involved, 'murica style: https://hardware.slashdot.org/story/20/05/29/2340240/western-digital-gets-sued-for-sneaking-smr-disks-into-its-nas-channel Too bad the most likely outcome of this lawsuit -should WD lose- is a $5 coupon good for a discount in your next HDD purchase from WD. Oh, and a bunch of slimy lawyers getting stinkin' rich :/ For desktop usage, SMR is fine - my WD is doing fine so far and I haven't changed my usual desktop usage patterns so I haven't experienced long stalls, and I let my machine TRIM the drive weekly around midnight. Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
Nicholas Steel |
Posted on 20-05-31, 12:35 (revision 1)
|
Post: #362 of 426
Since: 10-30-18 Last post: 499 days Last view: 14 days |
And apparently Western Digital is the only one to put SMR drives in a NAS oriented product line. AMD Ryzen 3700X | MSI Gamer Geforce 1070Ti 8GB | 16GB 3600MHz DDR4 RAM | ASUS Crosshair VIII Hero (WiFi) Motherboard | Windows 10 x64 |
tomman |
Posted on 20-06-23, 12:06 (revision 1)
|
Dinosaur
Post: #732 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
It's official now: the Mac is https://www.apple.com/newsroom/2020/06/apple-announces-mac-transition-to-apple-silicon/ https://www.phoronix.com/scan.php?page=news_item&px=Apple-Going-In-House-Silicon This now puts those longstanding rumors to rest: Intel is dead to Apple, and so are Real Computers for them too. A matching macOS release has now been announced too ("Big Sur", WTF is with those silly codenames!?) with Rosetta 2 for limited x86 compatibility. Oh, and the convergence of macOS and iAppliancesOS is happening too, as those new Macs will natively run iDevices apps (so I guess those will be getting touchscreens out of the box?) Kiss goodbye your very expensive creative tools and Steam games, Mac guys. Enjoy the ride while it lasts. No word on the future of the Mac Pro. Apple could have revived PowerPC (remember that they bought a PPC arch licensee years ago, PA Semi), but nope, they had to switch to cellphone CPUs :/ Today is a sad day for the future of personal computers, no matter what lies those ARM pundits want us to believe. Oh, this also kills Boot Camp and seriously threaten the future of bare metal Linux on Macs (for those that actually see Macs as "overpriced half-decent PCs" instead of "Think Different™") - Apple's "solution" to the latter is "you can always have VMs" instead of "wipe macOS and install Debian". Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |
BearOso |
Posted on 20-06-23, 15:45
|
Post: #162 of 175 Since: 10-30-18 Last post: 1451 days Last view: 1451 days |
Posted by tomman Don't worry about that. Just sit back with your Netflix and Youtube and Facepage and let the Geniuses at Apple handle the programming. |
Nicholas Steel |
Posted on 20-06-23, 17:20
|
Post: #367 of 426
Since: 10-30-18 Last post: 499 days Last view: 14 days |
If they wanted to distance themselves from Intel, wouldn't it be far simpler to just go with AMD? AMD Ryzen 3700X | MSI Gamer Geforce 1070Ti 8GB | 16GB 3600MHz DDR4 RAM | ASUS Crosshair VIII Hero (WiFi) Motherboard | Windows 10 x64 |
tomman |
Posted on 20-06-23, 20:48
|
Dinosaur
Post: #733 of 1317 Since: 10-30-18 Last post: 1 hour Last view: 29 min. |
Posted by Nicholas Steel Apple wants to control the entire supply chain. Since Intel won't license x86 to them and they are still butthurt with PPC, their next best choice is ARM, where they already hold a design license allowing them to own and mess up to the last transistor on their custom A-series silicon. For those still waiting those Threadripper cheesegraters: wake up, it won't happen, the dream is over. Your options are sucktastic Wintendo Diez which breaks all your shit twice a year, and sucktastic Loonix with no proprietary applications you want to run. But then, it's not only Apple ditching Intel at the "very expensive designer computers" end: at the very niche but highly lucrative supercomputer market, Fujitsu is moping the floor with them with their newest baby, Fugaku. Yes, you've read it right: it's a half-EXAFLOP (that is, 500PFLOPS) beast... and that's because it's not finished yet - the final design calls for a whooping ONE EXAFLOP, and that will be for early 2021. When it's done, it will be more powerful than the next 10 machines on the TOP500, with plenty of muscle to spare. And it's being done with ARM64 cores - 150K of 'em. Because Fujitsu (which was the last SPARC holdover besides Orrible) has seen the writing on the wall. But then, supercomputers are not exactly the first things you would think when Joe Gamer, Fred Videncoder or Tim Starbucks from Javascripts'R'Us thinks about "high performance computing". Also, IPC is a thing, and supercomputers cover very different use cases we're not likely to see at home or work. Licensed Pirate® since 2006, 100% Buttcoin™-free, enemy of All Things JavaScript™ |