So, does this mean that Seagate have finally figured out how to build drives properly? I've been avoiding them like the plague for years after bad experiences with earlier models.
Just 139 out of 10,000 12TB Seagate drives fail a year, and Western Digital's HGST brand has an even better rate of 51 in 10,000, according to cloud backup service provider Backblaze, which has 104,778 drives spinning in its data centre. It's not an exhaustive study; the firm listed just four brands in its estate, with models …
Not true, for a while there, almost all Seagate drives (even enterprise ones) were pretty shit.
I had many a NetApp/EMC, etc. etc. that came packed to the gills with Seagate drives, that started regularly failing on a very regular schedule starting just about half a year before the normal warantee period on those drives. I had a Sun Thor (48 SATA 1TB Seagate drives) that probably had lost 70% of its original drives.
The replacements started coming back with Seagate drives, that failed again after replacement.
Pretty soon the replacements starting came back with HGST or sometime Toshiba, and those disks never had to be replaced again.
The DL and DM series were utterly dire(*) but we've seen a noticably higher failure rate for all seagate units than for WD equivalents and the HGSTs just keep on trucking.
(*) Average life of a DM was about 10 months - the 3TB DMs were even worse than that. On top of that they seemed to be sluggish, which I suspect was a lot of time sorting head positioning.
They were rather solid 10-20y ago-- I still have the 320GB SATA disk I bought for the AthlonXP machine (iirc). It moved around, visited a gaming box which was turned into a HTPC then torn down... now it's mostly a cold store and I forget what's on it. But it works. The 500GB drive I got not long after, was sacrificed -- donated its PCB to a friend's near-identical drive which had a bad time. That one was a single volume with no MBR, all but filled with the one copy of lots of personal stuff after I setup LUKS, until I blew out the header with fdisk or something. Fun fun.
But then Seagate assimilated Maxtor and the word on the street was their drives got crappy all of a sudden. So I quit buying Seagate. Did that change in the meantime?
"But then Seagate assimilated Maxtor and the word on the street was their drives got crappy all of a sudden. So I quit buying Seagate. Did that change in the meantime?"
Seagate bought up connor, quantum and maxtor over time. ALL makers have good and bad periods (and particularly good or bad models), but Seagate have been reliably rotten for the last 7-8 years.
If this has changed it's a good thing, but the whole reaction to the Thai floods(*) leaves a bad taste.
(*) Both WD and Segate set about Increasing prices, slashing warranty periods whilst decreasing build quality. It took about 6 years for drive prices to recover to pre-flood levels but five-year consumer drive warranties are a distant memory.
That since they are no longer under such pressure to minimize cost for those larger models being shown, they don't have to cut corners they way they did when they needed to squeeze the last cent out of the cost of a 500 GB drive going into the $200 low end PC/laptop Black Friday special.
Hm. Indeed. It was introduced as it was a feature in the SAS specifications. Most people kind of know the difference between buying a SAS drive and buying a SATA drive. During this transition period, which will be decades long, they'll have to learn to spot the difference between a SAS drive and two subtle flavours of SATA. So from two semi-incompatible standards, they've created three, two of which are wholly incompatible but are given the same name. But (1) why, on physically identical connectors, did someone decide to make it active high? and (2) why didn't WD/HGST put a jumper in so one could easily disconnect the pin from the 3.3V supply that the specification previously assigned to it and which most leads and backplanes for the last 20 years have been built to support?
So they can resell you a drive when you need that feature, or resell you a drive if you got the wrong part number by mistake? Drives that can be used in either configuration would be damaging to their sales.
Either way, double the number of stock line items, double the number of sales. That's how it works doesn't it? Having massive inventories of very specific drive/controller combinations in the supply chain obviously has some hidden benefit to them.
I tend to agree back when Seagate announced that they were ceasing production of their 5400 RPM green drives and two reasons I remember they quoted at the time was that the faster drives they had available only used something like 0.5 watts more power which they considered insignificant and also there were too many SKU's, naturally after a month or two's absence perhaps then magically they suddenly had 5400 RPM "NAS" drives available for sale in quantity and of course at a much higher price.
"But (1) why, on physically identical connectors, did someone decide to make it active high?"
Because Google, Amazon, ect had been cheaping out with their custom back planes for years, because the 3.3v pin wasn't used by HDDs they decided to not bother including it on their backplanes. So when they decided to use that pin for remote shutdown they were faced with the option of either making the drivers not work in their old servers or in consumers old computers (which did supply the 3.3v), so they chose to make it the consumers issue.
Active low logic would mean that the unconnected pin floats and the drive remains powered. That doesn't explain it, unless their back planes tied pin 3 to ground which wouldn't make sense as many, many vendors actually physically tied pin 3 to its neighbours (1 & 2) that were ALSO 3.3V in the spec, but were supposed to mate after pin 3 did.
3.3V in the SATA spec, I should make clear. The specification for the SAS connection had them as N/C, so they would float anyway. The whole thing just doesn't make any sense to me. And it is claimed that the SCSI Trade Association determined that writing the spec the way it did wouldn't be a problem.
"With the introduction of 12G SAS, a new SAS standard, SAS-3, redefined P3 (Pin 3) from “3.3V Power” to “POWER DISABLE”, i.e. “Reset”. At that time, the STA (SCSI Trade Association) researched the marketplace and determined that there were no conflicting legacy concerns."
WD were key contributors to that association and the standards forum. Their suggested solution for those that were silly enough not to spot that HUH721010ALE604-0F27606 was so radically different to HUH721010ALN604-0F27609 when the textual description of the two variants in the reseller channel is actually identical, is that you use a MOLEX adaptor. Which isn't really an option if you have a backplane connector. Worse, some hot-swap backplanes actually work fine with these drives until you power cycle the device and it puts the expected 3.3V onto that line during the POST and enters an error state as the expected drives don't appear or it bums out as there's a shadow BIOS (or equivalent) on the drive itself which is supposed to load and fix that particular quirk.
One really wonders about the thinking behind this... I mean, just why? What was the reason? If there is a good reason, it'd be more acceptable, or at least understandable. Why isn't there a jumper for PWDIS or (overline) PWDIS? Or a DIP switch? Or even a register option somehow?
Interesting - the results pretty well mirror my own experience, even though I'm dealing with far fewer drives (still hundreds of them.) Overall, I won't buy a Seagate drive if I can possibly avoid it; for well over a decade they have consistently been the worst in terms of reliability for me.
Most annoying is the way they seem to avoid logging real SMART failures - even obviously knackered Seagate drives will often pass an extended SMART test.
Then again, I won't buy a spinning drive at all if I can possibly avoid it - SSDs may fail in a more brutal manner but they have been orders of magnitude more reliable in my experience, even fairly ancient ones.
Two of my three WD Red 3TB discs I bought in Dec 2013 have died, thankfully not at the same time as they're part of a RAID5. One of those replacements died in the warranty period, which was replaced by WD. Thanks to Linux SW RAID allowing me to put the old disc on USB and still assemble the RAID, I was able to migrate to data to a new disc without degrading my array, only pulling data from the other discs when the copy hit bad sectors.
I only buy Enterprise rated drives for my home computer.
If you don't, you're just asking for trouble.
"Enterprise" drives are designed for different usage patterns, and are likely not to perform so reliably in a "consumer" environment ... given their significantly higher price methinks you are making a false economy.
You'd do better to buy two consumer drives and run them in a RAID1 array.
Can you quantify that "extra strain" as I can only see two potential problems and that is the head loading/unloading cycles and thermal shock from going from cold to hot and back again. The first is not a problem as even consumer PC's have at least 400K to 600K head unload cycles so daily shutdown will only account for 365 of these a year so no problem there until you get aggressive APM spinning down idle hard drives in some cases in just 8 seconds so some people have racked up hundreds of these cycles in a matter of hours especially in external hard drives, enterprise hard drives usually have more unload cycles available than consumer drives even though they don't need them. Secondly I haven't seen any statistics where tests are done on hard drives which are say fired up and run for maybe half an hour to warm up to their working temperature and then turned off for maybe half an hour to cool off and this cycle repeated over and over until failure occurs so I presume this is a non-issue either unless you can point me to some research somewhere.
I use whatever consumer drive I can buy cheaply, so that includes a couple of older Seagate drives, one of which has been making some nasty clicking noises for a few years now.
However, I'm not worried about data loss because I'm using RAID, and don't forget that the 'I' stands for 'inexpensive'.
If you want uptime, use some form of RAID, if you want your data, have backups, but you only need enterprise drives in a server that's run hard 24/7.
Echoing others on this:
Most Enterprise drives are intended for RAID operation and towards that end they have the sector recovery timeout set to 7 seconds before marking it bad and moving on (consumer drives can spend up to 3 minutes trying to get data out of a flaky sector)
Using an Enterprise drive unRAIDed in a consumer environment without changing that timer risks losing data if a sector goes bad.
The other differences between enterprise and consumer drives are usually:
- a screw in the lid fixed to the end of the splatter hub (supports both ends of the platter assembly)
- more powerful error correction (10^10E15 instead of 10^10E14 - which practically translates to a silent ECC failure every 40TB instead of 4TB). This may or may not be a firmware thing as my understanding is that enterprise and consumer drives in any given family share the same spec CPU
- stronger magnets/coils (to give better seek stability and more powerful centering - usually achieved by simply turning up the power to the existing assembly)
- more sensors for vibration (important for use in racks and chassis with lots of drives)
- enhanced operational temperature ranges (why, I'm not sure. Most enterprise environments are closely controlled)
The flipside of that is Enterprise drives reliably pull 12-14W when idling instead of 5-7W, so things DO get hotter and putting Enterprise drives in the wrong case (poor airflow) can cook things. They're usually significantly noisier too thanks to higher seek acceleration and seeking power consumption can peak over 20W so you need to be careful with power supplies.
Their service lives aren't usually any longer than consumer ones (it used to be common for our enterprise drives to die _sooner_ than their consumer siblings) and our experience is that they're not as tolerant of a case being kicked/knocked/etc as consumer drives.
All in all: Unless you have an actual need for an enterprise drive, I wouldn't bother. You're paying anything up to 300% premium for virtually no discernable improvement in performance or reliability in a home environment under most circumstances and much higher running costs.
You can more easily take care of the ECC issue by using ZFS, which will give higher effective recovery rates anyway.
Those were the only drives since the advent of winchester hard drives that ever had so many fail so quickly that I actually lost data!
I must be a lucky so and so. I've owned no lemons from GM, or from Seagate. Timing must be everything. I've had one seagate 2.5" drive start to vibrate very slightly after 3 years 24/7, and I pulled it offline to use as a cold backup. It's still working, as all the 20 or so IDE drives I recently pulled from machines I should have decommissioned quite a long while ago (pentium 2 class) - they were just sitting in storage till a friend wanted them to recover any gold he could. Checking the drives, they worked, the systems still booted for crying out loud. I pulled some interesting nostalgia off them, and now they're gone, never having failed. All these ran for a couple years in a software development business...
Deathstars didn't turn up their toes for mechanical reasons.
They had the same 49.7day uptime bug as Windows95 had and for the same reason. Counter rollover.
32 bits unsigned max value is 4,294,967,295; that many milliseconds comes out to about 49.7 days
Believe it or not I still see this one showing up on various pieces of kit. Most recently: Enterprise Wireless Access Points, but it's shown up in tape robots, Hardware RAID arrays (as in the self-contained drawers), NASes and other systems over the years.
I used to see a VERY high failure rate of WD drives for my customers, many already failing before they got them out of the box, literally. One lady had one fail in her laptop so for me to do the data recovery she got one of the middle-[priced ones and brought it straight to me, roughly a 5 minute drive. Opened the shrink-wrapped box, plugged the drive in, no go no spin no nothing. Check the wall-wart, that was fine, tested the USB lead on another device, also fine. Sent her back for another one. Spun up, started transferring data for an overnight transfer, came back to a few megs transferred and an IO error. Two more drives failed before we gave up and went with Seagate - the last WD one lasted a week, but had the 5-6 hour transfer of her data from our recovery disk then was sat on a shelf till she came to collect. When I went to show her the data was there, safe and sound, tic tic tic.. (not tick of death but clicks of badly de-virginised surface).
From what I know her Seagate drive is still fine some 5-6 years later.
I agree, I haven't see a drive fail in several years now and yet they use to die all the time - I still have a pile of NAS drives that I bought several years ago to keep the company servers running - I haven't even opened the boxes.
But on the other hand, I powered up an ST-506 this weekend and I think my hearing is still affected - it was absolutely deafening ... OK, so it was in a Xerox 820, about 40 years old and at least it still rotated ... until I slung it in the recycle bin.
since 1996 or so, the only drive I've had going wrong (apart from a seperate IDE controller which went in 96 ish) was a maxtor 40GB, and that was pretty much my own fault because I ran it in a 'quiet' enclosure without the metal heat sink brackets, it overheated and started developing errors. mostly running 500GB sata western digital blues now (apart from my hitachi 2TB external). the annoying thing with two IDE western dig's I have is they can't be used on docking stations because of the jumper configuration.
Used to be if you had a WD drive it would take two to six months for it to fail. Bought one a couple months ago, it was already dead fresh out of the box.
I got a drive for a friend in a boxing day sale. They didn't have the Seagate he was after so after MUCH discussion I got him a WD for the same price. He was desperate, and though I had misgivings I got the drive (it was a good special).
This thread reminded me and I rang him to check. Such language!
Oh, the drive has been returned, he got a replacement but also tresspassed from the store. Apparently he made some offers about where to place the old drive that the manager didn't quite fancy partaking in. Thankfully I suggested we don't trust it and give it movie machine duty instead of main file store duty. Thankfully more so that he listened.
Not dead out of the box, but it didn't last a month. At least WD are consistent!
Icon - SOP for WD!
I keep thinking I need to build/buy an external storage thingy for home (RAID 10). I was thinking of buying drives from different manufacturers so they don't all fail due to the same cause at the same time. Mirroring across identical drives (from the same batch) seems like a recipe for trouble.
> Mirroring across identical drives (from the same batch) seems like a recipe for trouble.
Yup. I've seen a couple of RAID5 setups collapse for precisely this reason. "Go to backup."
Suggest you look at ZFS rather than RAID. Step-change in what you can do, and serious step-change in reliability, at the bit-level.
Check out the 1-bit error examples in the photos there. (I lost 99% of 20yrs' photos to these errors, coming back from UK to Aus).
But note the ability to, eg: decide to upgrade OS, do Snapshot, Upgrade, discover nightmare, do RevertToSnapshot, and continue as-was In Under A Second without faff. (vastly better lowlevel approach than LVM, btw)
I'd be more interested in more metrics: more platters or denser platters would have any bearing - no pun intended - on failure rates. Or any particular construction detail, for that matter. Things like heat dissipation, the sort.
The data seems to be too random, with one model or another being lucky.
I thought I wouldn't comment on this post, as the topic is really enterprise oriented and I'm just a home user with variously 12 or 15 drives installed in or connected to variously 6 or 7 PC's and notebooks in a home network. But I see other small-time users commenting below, so I felt I had to jump in.
Everything I have is always running 24/7.
Everything I have that came with a hard drive, I immediately replaced the drive with a higher capacity Seagate drive. Always. 1 or 2 TB 2 1/2 inchers in the notebooks, and 2 or 4 TB 3 1/2 inchers in the desktops. I guess I standardized on Seagate years ago because of the free DiskWizard cloning thing they provide. DiskWizard has been totally bulletproof for me for cloning. Never a hiccup.
I have never had an ounce or a blip or a jot or tittle of trouble with a Seagate drive--ever. My AFR has been zero. Even if I had one go tits-up and screech itself to death tomorrow, I would still be high on Seagate. (Maybe the next one, not so much...) It does, in fact, surprise me that these unbelievably complicated $100 products work so well.
The reason I felt compelled to make this hearty endorsement here (I don't work for Seagate, etc. etc.) is that it seems to me that bellyachers and bad-mouthers are often the main people that ever post anything. Some people love to boast that, in their huge profound experience, a wide-spread, huge volume, successful product like a Seagate drive is crap. An AFR in the 1% range, to me, for something like an HDD, either absolutely or relative to competitors, is totally fine with me. If the price is right, I'll keep buying, even as an enterprise user. IT'S THE ENTIRE REASON FOR A SOLID BACKUP REGIME. To call Seagate drives "crap" or "shit" (same thing) is fairly unintelligent. Huh ?
Largely this. My HDD failure rates have always been pretty low. I had a Deathstar fail (but then, I had 3 of them, so I was pushing my luck). I've had a single Samsung drive fail. My home server currently runs WD Caviar Green 3TB, and I've lost 2 of them, I think. But that lives in the attic, so it gets cold-cold in winter and hot in summer. Not the best environment.
I think all the manufacturers are pretty interchangeable for the most part.
Biting the hand that feeds IT © 1998–2019