Slashdot Mirror
Seagate Ships First 8 Terabyte Hard Drive
MojoKid (1002251) writes Seagate announced today that it has begun shipping the world's first 8 Terabyte hard drive. The 8TB hard drive comes only five months after Western Digital released the first ever 6TB HDD. Up until then, Seagate's high capacity HDDs had been shipping only to select enterprise clients. The 8TB HDD comes in the 3.5-inch form factor and, according to the manufacturer, features a SATA 6Gbps interface and multi-drive RV tolerance which makes it suitable for data centers. It's unclear what technology the drive is based on, or if PMR (Perpendicular Magnetic Recording) or low-resistance helium technology was employed.
That sure is a lot of porn...
WTF is that?
And I did RTFA, it doesn't mention it at all.
I'm surprised this wasn't posted by Timothy...
I remember when tape drives stored a few times more data than hard drives, and were priced about the same. I know I can back up to external USB drives (which I do using Snebu, but I which tape drives were more affordable.
Just like before I can lose entire tv series when the disk fails. But now it's the HD version of the series I will lose. That's called progress.
lucm, indeed.
Before SSD's were all the rage, a common thing to get a speed boost was to do 'short stroke' the drive. Essentially, all you do is only partition the first third of the drive and use that space.
The theory is that the head doesn't need to move around as much and speeds up the drive. I've never done it but modders used to swear by it.
Yeah, right. Media consumption is at an all time high. Between PVR, data mining markets, virtual machines, backup data, servers, and etc. At 12TB I still don't have enough space.
It's a bit weird to think about really. Back in the early 90's I worked for a company that had a "huge" 2TB array (which consisted of hundreds of drives). It was lol fun to keep the damn thing running because it had so many drives there were multiple failures almost every day.
Would it be trivial to design a drive that can be switched into a double-speed half-capacity mode?
High RPM drives tend to have smaller capacity if I remember correctly, and any drive can be short stroked to save on seek time.
Get Perpendicular!
Sdelat' Ameriku velikoy Snova!
But how would the host detect the redundancy if the guest operating systems use full disk encryption? The cipher modes they use are specifically designed to hide whether a sector is a duplicate of another sector.
I doubt it would be trivial: you can sacrifice capacity for some speed by reducing the amount of platter area you use(and thus how far back and forth the read/write head assembly needs to move); but RPM is still a serious constraint, and bumping that tends to get rather costly. 15k RPM has been the effective ceiling for years, and while increases in data density improve best-case read and write speeds they have no effect on how long you have to wait for a given chunk of disk to finish its rotation and come back under the read head.
It also doesn't help that SSDs are aggressively moving into the high speed area. If you applied the engineering tricks used in ultracentrifuges you could probably build a damn fast HDD; but doing so for less than the price of a really nice SSD would be a great deal more challenging.
I remember when Washington University in St. Louis installed a new RAID server back in the early 90's. It had a capacity of 1 Terabyte and only cost them $100,000. I remember thinking that was an amazing capacity. Now I've got 7 terabytes of external storage on my desk and it's almost completely full. If you build it, they will fill it.
I've heard that NSA wants to buy two or three.
Why hasn't the price of data centers come way down with new storage technology? For example, why not keep a few terabytes of offline storage in your desk drawer instead of paying $$$ for tapes? If tapes are more reliable then what level of duplication is needed for disks to be as reliable? This combined with the multiplier effects of no_AC_necessary solid-state ... why not big data center in small closet? If the data center is inefficient, why is it still around? Latin me that, my trinity scholard.
I just had my third 1tarabyte+ hard drive fail tonight. I remember when hard drives DIDNT fail. It wasn't even a think I thought could happen. It's nice they can get them so large now, but I don't want that much in one place. I'd rather have several smaller drivers raided waiting for the inevitable.
So I have multiple servers in different locations all using 3TB external USB3 Seagate drives (powered by AC adapter). At least 12 in total, one for each server used for BMR backups. In less than a year, ALL DRIVES FAILED!!! Either they started out with bad blocks and progressively got worse, or just died.
Seagate, never again! The article below doesn't show just how bad Seagate drives are when used every day.
http://www.pcworld.com/article...
Life is not for the lazy.
It would take me 20 months to fill that up.
Visit the Arcade Restoration Workshop @ http://www.arcaderestoration.com
If it's the outer most tracks, sure. More surface area = more bits. As such, that 1/3rd would be physically narrower making the actuator arm not having to swing as far back and forth when reading/writing to that partition. As a bonus, the outer most track also has faster throughput as more bits fly under the head vs the inner track regardless of RPM.
Life is not for the lazy.
Considering how awful QC and MTBF have been with Seagate in the past 10 years or so, I really can't think of a good reason to buy this drive.
Damn_registrars has no butt-hole. Damn_registrars has no use for a butt-hole.
I would MUCH, MUCH rather have half speed double capacity. Just about all my storage comes much closer to write once, read mostly.
I suppose, but if your data is only small, a good OS will probably put it all together at the beginning of the drive anyway.
Plus, OSes perform better when they have got a lot of space to work with. So not all usage scenarios would improve.
If you have 8 TB of capacity in the form of two 4TB drives, you can trade speed for capacity via RAID. With RAID 0, each druve gets half of each KB, doubling throughput.
I've often wondered about building a drive with TWO sets of read/write heads. All drives going back decades read one cylinder at a time. Why not add another set of heads intge other side of the platters and read two cylinders at a time. Rotational latency eould be cut in half. One set could be used for the inner tracks and one for the outer tracks to reduce seek time.
Expanding on that, why does the head read from ONE point. The arm could be lined with a row of ten read heads. We can put millions of pixels on a four inch screen, why can't we put ten reading sensors on a two-inch arm?
love to see the rebuild time on my RAID stripes.......................zzzzzzzz
Slashdot ran a story that 60TB Disk Drives Could Be a Reality In 2016 based on promises made by Seagate.
Rather than a doubling, it seems like hard drives are continuing to go up by 2TB in the first half of the year and an additional 2TB (for a total of 4TB increase) by the second half of the year. So, it seems more likely that in 2015 we will see 10TB and 12TB drives and then in 2016 there will be 14TB and 16TB drives. While a 16TB drive is impressive, it is still only a quarter of the size promised. Also, at a rate increase of 4TB per year, we will be at 30TB by the end of the decade which is still half the 60TB prediction.
While there are some applications which could take advantage of the additional storage space, there are more applications that could take advantage of the improved performance provided by SSD. So the million dollar question seems to be at what price point and density does SSDs have to reach before the industry phases out hard drives? I don't think hard drives can mature fast enough any longer to survive to the end of the decade.
12 TB, snort. I've got over 100 TB worth of 2 and 3 TB drives on-line or on-call a boot away. The most critical part is mirrored RAID-Z2 (4 drives' worth of redundancy per data item), and most of the rest is ad hoc replicated via rsync, some of it several times, so there is nowhere near 100 TB of data stored, but there is a lot.
I would definitely be happy with 64 of these 8 TB. At least for a while.
The Pillar Data Axiom SANs did this. DEC filed a patent for it back in '92. http://www.google.com/patents/...
I'm sure they already have 'two or three' on order!
Say you have an 8TB drive with 6 platters - the option could be to pair up the platters and write alternate bytes to each, doubling sustained read and write
That would require the head to be right over both tracks at the right moment. I'm not sure the heads are physically aligned that precisely. Or are you suggesting to separate the head assemblies for the top 3 and bottom 3 platters and do RAID 0 in a box?
Would it be trivial to design a drive that can be switched into a double-speed half-capacity mode?
There's a word for it... "Velociraptor".
There's even a word for a drive that's "triple" speed... "Cheetah".
In any case, you wouldn't decrease the capacity on account of the faster rotational speed... you'd just use a faster DSP capable of doing its thing in less than half the time as a slower drive. From what I recall, the Cheetah's storage density per platter was basically the same as any other 2.5" drive.
SSDs obviously made the highest-performance spinning disks almost irrelevant, but personally, I used to think it would have been awesome if Seagate had taken the Cheetah platform, added two more independent sets of actuators and read/write heads, and wired it all up to look like 3 SCSI drives with sequential SCSI IDs so you could have single-drive RAID-5 performance in a luggable laptop (think: inch-thick Alienware/Sager/Clevo) or SFF desktop. Heat would be an issue... but really, a Cheetah didn't throw off any more heat than the mini-PCIe discrete video cards found in some gamer/mobile-workstation laptops now. In MY laptop, at least, the GPU's cooling system is bigger than the CPU's.
One thing I'd LOVE to see, and even think there's a market for, would be a single-platter drive suitable for mounting in the optical bay of mobile workstation laptops (say, 120mm diameter, 7mm or thinner). I rarely use optical discs, but having another 4tb or so that's always with me would be nice to have. Basically, it would be 7mm thick Quantum Bigfoot from the late 90s... and Jesus, with that much diameter per platter, just imagine how many terabytes you could pack into a multi-platter drive that fully-consumed a 5.25" quarter-height drive bay. It's almost scary to think about something like a 256-tb 5.25" single-bay hard drive.
I'm also kind of surprised that nobody ever made a thin-but-3.5" drive for laptops (which would obviously need a larger drive bay... but modern laptops, even thin ones, have SHOCKING amounts of horizontal acreage under the keyboard that could easily be put to good use for bigger cheap drives).
In the early 1990s, AIX allowed you to partition drives (physical volumes) where a logical volume could be residing on the inner or outer part of a drive. That way, DB indexes and critical tables could be placed where access was relatively fast, while the stash for archive logs, program files, and stuff not really accessed could be placed on the outer part. Not SSD speed, but it was a way to help with database performance, especially if one had a lot of spindles.
As you mention, 15k SAS drives are going to be rapidly undercut by SSDs. The price difference is no longer 10x or 20x when looking at cost/gigabyte, the price difference is now only 2-3x.
Pay 2x-3x the amount for a SSD of the same size as the 15k SAS, and you gain 50x improvement in your IOPS. For workloads where that matters, it's an easy choice to make now. As soon as you say something like "we'll short-stroke some 15k RPM SAS drives" - you should be considering enterprise level SSD instead. Less spindles needed, less power needed, and huge performance gains.
The only downside of SSDs is that write-endurance. A 600GB SSD can only handle about 120TB of writes over its lifespan (give or take 20-50% depending on the controller, technology, etc). The question is - are you really writing more then 60GB/day to the drive (in which case it will wear out in 5 years).
And more importantly... will you care if it wears out in 4-5 years? That you could handle the same workload using fewer spindles and less power likely pays for itself, including replacing the drives every 4-5 years.
Wolde you bothe eate your cake, and have your cake?
Anyone else remember when 10MB was a decent size disk and 30MB was huge? Man I'm getting old...
Isn't it the outer portion, rather than the inner portion, given that you can reach more per revolution if it's written to the outer edge, on account of the greater circumference? And if so, then yup, this is a viable technique for speeding up read times. OS X actually implemented something similar as far back as 10.4, where it'd move the OS and other frequently-used files to the outermost portions of the platter in order to improve read performance. I never really noticed a difference, personally, but Apple clearly thought there was one, given that they implemented it into their OS as a standard feature.
If they could strip the capacity across the platters, don't you think they'd do that already? It would mean a 12x increase in sequential read/write with a 6 platter drive.
I see in this drives future, let me see my crystal ball.....2 years from this day. Yes....
The drive shall fail.
Your mystical fortune says...let me see...
Use backups.
That'll be $75.
No, you can't see my third nipple.
It is pitch black. You are likely to be eaten by a grue.
A bit off topic, but what would be the recommended file system to use on a drive like this when you're using it for backups? Something with built-in file checksums or is using ext2/3/4 and writing a script to generate and validate CRC files better?
I bought a 4TB WD My Book yesterday and am slightly concerned about the high failure rates for the 3TB version of the drive. Something about bad controllers...
Welp, seems my post was a bit misunderstood. I was actually thinking transfer rates. Say you have an 8TB drive with 6 platters - the option could be to pair up the platters and write alternate bytes to each, doubling sustained read and write.
It could also be an option to turn on when you start using the drive, and if it gets half-filled up, it should be possible to decouple them and get the full size.
The tendency for many consumers is to have an SSD boot drive and a platter storage drive - but that platter drive takes some time to fill up, why not double speed it until it's half full?
I'm not 100% certain, but I believe the problem is that the hard drive head assembly moves as a single unit, which means all of the heads for all the platters must move in unison. But the precision required to move the heads to the precise spot on the tracks where the data is recorded is such that it would be too difficult to design the heads in such a way that when one was over its track, all of the others would be *guaranteed* to be over their tracks on their respective platters. To do this you'd need to have the heads each on their own arms with their own voice coils to keep them all on track simultaneously. But that would add enough cost to the drive, it would be cheaper to just buy two half-capacity drives and stripe them yourself.
Basically, I think its possible, but not economically logical to make hard drives in a way that would allow for this kind of in-box striping. That's what RAID is for.
But like you said that is totally dwarfed by SSD.
I doubt this would be cheaper than a fast 15Krpm 4TB 2.5" drive to manufacture and the 4TB drive would probably be faster overall. Sure it'd work on a 3TB consumer drive and probably be a good tradeoff, but on "the largest capacity drive in the world" I have my doubts it'd be economical and most certainly not double the speed.
I was promised a flying car. Where is my flying car?
My Altair 8800 running CP/M had a washing machine size hard driver. Air compressor,
big power draw, etc. Had a 5 MB fixed platter, and a 5MB removable platter. I think
was made by Shugart. Interface was parallel port (not printer port, but similar).
In its day, it was the cat's meow.
I still have that Altair, but not the drive. I replaced with a 5MB drive to a parallel port,
8" then 5". I also experimented with IDE interface and a 3.5" drive, but I do not
remember the capacity. Then the Altair got stored away. I have a bunch of them,
and some IMSAI 8080 (a better computer).
Yes, Jeannie ... I *am* an old geezer.
Actually I think the goal is to use the innermost tracks on the disk. The linear read speed is slower, you're correct that the outer tracks are faster, but the inner tracks have lower seek times.
You're dreaming if you think HDDs don't fail without warning.
Back in the day, my college campus mainframe, a Burroughs B6700, had (in addition to its more conventional "disk pack" drives) a head-per-track (HPT) drive. The disk was several feet in diameter and the whole surface was covered with read/write heads (they didn't need to move).
Can't find specs on the B6700 version, but here's a blurb about the older B5500 version (from http://www.retrocomputingtasma...)
-- Alastair
Interesting point.
I suppose, but if your data is only small, a good OS will probably put it all together at the beginning of the drive anyway.
It depends on the file system. It's hard to say what strategy makes a "good OS". NTFS puts files sequentially, which gives the benefit that you will have lower seek times if you do not have that much data on your hard disk. The downside is fragmentation. Now, ext4 spreads the files over the volume, which avoids fragmentation efficiently. The downside is constantly high seek times across files.
We'll just put your data onto your new Seagate drive... aaaand it's gone!
Worst. Signature. Ever.
Seriously, burn your nerd card. Tar and feather this bastard!
I'm guessing the reason drives don't have multiple read-write heads is because that would significantly increase the cost of the drive and potential for mechanical failure, to the point where you're probably better off simply getting a second drive and using RAID as you indicated.
The row of read heads probably wouldn't work - at least for simultaneous reads - because the distance between them wouldn't properly align at different angles of the arm as it swung across the platter. You'd need to design an arm that tracked linearly across the drive, adding weight and complexity. Even if you could do that, the real benefit would be if you could get them to read parallel tracks. I'd imagine that the multiple magnetic heads would tend to interfere with each other if placed that close (you couldn't stagger them or you'd have alignment problems again). If they're placed far apart to not interfere with each other magnetically, then they have little benefit, because the odds of them aligning on multiple chunks of data across the drive that a client wishes to access would be much lower then, largely negating the advantage.
In theory, I guess that creating an array of read heads spread out across the arm but activating only one at a time (whichever is closest to the cylinder being read next) might be a way to improve latency. I can't think of a reason offhand why this wouldn't have been tried except that the added complexity might have a negative impact on reliability, or perhaps the increased weight of the arm would tend to negate this advantage. Or, it might be difficult to re-align the arm to arbitrary positions as opposed to the predictability of a single read head.
Generally speaking, in such a mature industry that's obsessed with squeezing more performance out of their hardware, it's probably safe to assume that engineers have already thought of and rejected such ideas as being impractical for some reason or another.
Irony: Agile development has too much intertia to be abandoned now.
first third wouldn't really do it on multiplatter drives though?
you'd have to do 100gb there and skip and 100gb and skip and 100gb... just the right way.
wouldn't surprise me if modders used to swear by bullshit though.
what I had do do once was to skip 600mbyte in the middle of a 3.2gb drive because that area was a broken platter or head and would crash the drive if tried to access - it worked just fine when I formatted around that area though...
world was created 5 seconds before this post as it is.
From TFA
PMR (Perpendicular Magnetic Recording) or low-resistance helium technology was employed.
They actually use a variant of PMR that's based on magnetic monopoles. The reason why they're "shipping only to select enterprise clients" is because there's a limited supply of those, looted by the Red Army from a secret Nazi lab in 1945 and only recently rediscovered in former NKVD archives in a bunker outside Moscow.
Not a lot of people know that...
Honestly if this were a WD article someone would come up with the same anecdote and a different brand. Every manufacturer has had bad batches. I too have had a Seagate fail. I also had a WD fail. Like 4 IBM drives fail, a Quantum drive fail.
I wouldn't trust a Seagate drive with 100 MB of data, let alone 8 TB.
That's not necessarily true. You can get the same amount of space in a smaller number of tracks around the edge of the disk, so the horizontal movement for the largest seek is going to be smaller. Seek times on mechanical disks are based on three factors. The first two are related: the time it takes to move the head between tracks (proportional to its distance) and the time it takes for the head to settle and be able to be lowered again (dependent on its speed). The third is the time it takes for the correct sector on the track to spin under the head. In the middle, you have fewer sectors per track, so you need to move the head more often (this is where the upper bound on seek times comes from).
I am TheRaven on Soylent News
As you mention, 15k SAS drives are going to be rapidly undercut by SSDs. The price difference is no longer 10x or 20x when looking at cost/gigabyte, the price difference is now only 2-3x.
Pay 2x-3x the amount for a SSD of the same size as the 15k SAS, and you gain 50x improvement in your IOPS. For workloads where that matters, it's an easy choice to make now. As soon as you say something like "we'll short-stroke some 15k RPM SAS drives" - you should be considering enterprise level SSD instead. Less spindles needed, less power needed, and huge performance gains.
The only downside of SSDs is that write-endurance. A 600GB SSD can only handle about 120TB of writes over its lifespan (give or take 20-50% depending on the controller, technology, etc). The question is - are you really writing more then 60GB/day to the drive (in which case it will wear out in 5 years).
And more importantly... will you care if it wears out in 4-5 years? That you could handle the same workload using fewer spindles and less power likely pays for itself, including replacing the drives every 4-5 years.
I don't know what you're talking about. You can definitely write more than 120TB/600GB=240 times to the same bits.
I'd guess 2TB, before it fails.
it was possible to do this with PqMagic few years ago. It's a shame that it was discontinued, it was simply the best disk partitioner I have ever used.
Religion: The greatest weapon of mass destruction of all time
Inner tracks have higher seek times. The angular velocity across the whole drive is identical so it takes just as long for the platter to spin half way on both the inner and outer tracks. So the seek delay due to waiting for the platter is identical. However, the head also needs to move and the outer tracks require less head movement per amount of data (since each track stores more data). As a result the seek delay due to head movement is lower.
After having 5 seagate drives fail all within one year, including a momentous xt that died 2 weeks after I got it, it's replacement died a week after that, and THAT replacement died last week (less than 6 months across all three), I will never buy seagate again.
They are peddling crap, and I'm surprised they haven't been hit with a class action lawsuit yet.
Medium business with two locations. Each locations houses 3-4 servers, running about 15-20 Virtual Machines on each host. Every essential system is virtualized. Another server, lower specs, but loaded with plain 7200 rpm enterprise class drives (Not 10k RPM drives like the VMHosts) run Microsoft DPM 2012 R2. We have it constantly backing up. Our email and file servers are backed up on the hour or every other hour. All others that are more "set it and forget it" systems that dont change or store changing data are backed up once a day or so. The entire VM. Should a VMHost fail all child VM's can be restored immediately. Likewise I have recovery points going back 2-3 months for our main data drive and email using DPM with regular drives. I can get anything near instantly rather than having to search a tape. I can see tape would be useful if something was deleted years ago and needed to be recovered. However until then I'm drive only. Likewise all our VMHost servers are RAID5 or RAID6, and even our DPM server is RAID5 so if a drive fails we're okay. If two fail at once.. it's a backup. We also try to mix batches of drives in it as well or add them spaced apart so they have different operating hours and time to replicate if it has to rebuild from a single drive loss. (Why ive been switching our main servers to RAID6, as any weaklings would die sometimes during rebuild of a raid5 array or even raid1 array which leaves out SOL).
As you mention, 15k SAS drives are going to be rapidly undercut by SSDs. The price difference is no longer 10x or 20x when looking at cost/gigabyte, the price difference is now only 2-3x.
Pay 2x-3x the amount for a SSD of the same size as the 15k SAS, and you gain 50x improvement in your IOPS. For workloads where that matters, it's an easy choice to make now. As soon as you say something like "we'll short-stroke some 15k RPM SAS drives" - you should be considering enterprise level SSD instead. Less spindles needed, less power needed, and huge performance gains.
The only downside of SSDs is that write-endurance. A 600GB SSD can only handle about 120TB of writes over its lifespan (give or take 20-50% depending on the controller, technology, etc). The question is - are you really writing more then 60GB/day to the drive (in which case it will wear out in 5 years).
And more importantly... will you care if it wears out in 4-5 years? That you could handle the same workload using fewer spindles and less power likely pays for itself, including replacing the drives every 4-5 years.
I don't know what you're talking about. You can definitely write more than 120TB/600GB=240 times to the same bits.
Yes, but to all bits? Remember the drive will move around physical where a logical data cell is stored. Each time you write you are almost guaranteed it will be written to a new place and the old just marked free until all cells have been as used as that one.
Curious. Back in the stone ages (12 years ago) we had a 53 GB 12-platter drive (The box said "Solve your disk space storage problems forever!") that had a head fail. I was able to recover 22/23rds of the data, but it was clear that the data was recorded from one platter to the next all the way through the stack, and then the heads moved. Back in that day (I don't know if it's still true) one side of one of the platters just contained alignment information.
I love Mondays. On a Monday, anything is possible.
Your post contains:
2 paragraphs.
6 sentences.
375 words.
On average, your post contains:
3.00 sentences per paragraph.
62.50 words per sentence.
For comparison, typical English text contains:
4.49 sentences per paragraph.
38.58 words per sentence.
Chuuch. Preach. Tabernacle.
> You're dreaming if you think HDDs don't fail without warning.
More often than not, they fail WITH warning. Or rather, they give you some indication that it's time to replace a drive and you aren't stuck scrambling at the last minute because it was a surprise.
If this stuff is sneaking up on you, you are probably not paying attention.
A Pirate and a Puritan look the same on a balance sheet.
GParted would probably be a very familiar tool for you...
bork bork bork!
Why would they have lower seek times? It seems like lateral, track-to-track movement would be at the same speed regardless of position. And since rotational velocity is constant, the average time for a sector in the current track to come around should be identical. What's missing from that line of thinking?
Dewey, what part of this looks like authorities should be involved?
And then LBA came along in 1996 and completely mooted the strategy.
Dewey, what part of this looks like authorities should be involved?
One thing I'd LOVE to see, and even think there's a market for, would be a single-platter drive suitable for mounting in the optical bay of mobile workstation laptops
' Thinkpads T-series laptops have had that capability since the early 2000s. I'm pretty sure that current models still let you swap out the DVD drive for a 2nd SATA drive slot.
The problem with any solution that attempts to be multi-vendor is that every laptop has a slightly different form factor for their optical bay tray - there is no standard.
Wolde you bothe eate your cake, and have your cake?
Like this one?
In a nutshell, the smaller production runs make this type of drive more than twice as expensive as two regular drives, with similar performance for independent workloads. (But if you have a particular workload, you may benefit.)
Bill Gates once thought 640K memory would be more than anyone would need.
Sorry but I've dealt with more failed drives at the shop than you've had hot meals and if they fail "without" warning?
Then YOU sir are not paying attention! Before a HDD fails you will see several rather blatant warning signs, warning about delayed write fails being the most obvious but there is also temp spikes on the drive (as the motor heats up trying and failing seeks) and SMART changes (not talking SMART fail, which is usually at the end, we are talking large changes in the SMART values which can be read by one of several free programs such as HWMon or HDTune) not to mention most modern drives get REALLY noisy when they are getting ready to croak.
Compare this to the "dirty little secret" of the SSD world which is the majority of SSD fails are NOT the flash chips themselves but the SSD controller chip. When that fails? NO warning, NO chance to back up your data, just flip the switch and...poof. this is why I tell my customers they should use a religiously adhered to backup system along with cloud computing to insure no data loss.
ACs don't waste your time replying, your posts are never seen by me.
Sorry but I've dealt with more failed drives at the shop than you've had hot meals and if they fail "without" warning?
Unnecessary hyperbole.
Then YOU sir are not paying attention! Before a HDD fails you will see several rather blatant warning signs, warning about delayed write fails being the most obvious but there is also temp spikes on the drive (as the motor heats up trying and failing seeks) and SMART changes (not talking SMART fail, which is usually at the end, we are talking large changes in the SMART values which can be read by one of several free programs such as HWMon or HDTune) not to mention most modern drives get REALLY noisy when they are getting ready to croak.
I never said that HDDs never give warnings. I claimed that HDDs can fail without warning. I've had a few die with controller failures. It's not always a mechanical failure. I've also seen mechanical failures where the SMART information didn't contain any errors. For example, sometimes a head can just crash (rare, but can still happen even on stationary drives). You're making some dangerous assumptions on the types of ways that HDDs can fail, which if you really had dealt with more failed drives than I've had hot meals, you'd know that they aren't always predictable.
Compare this to the "dirty little secret" of the SSD world which is the majority of SSD fails are NOT the flash chips themselves but the SSD controller chip. When that fails? NO warning, NO chance to back up your data, just flip the switch and...poof. this is why I tell my customers they should use a religiously adhered to backup system along with cloud computing to insure no data loss.
I hope you give that same advice to HDD customers. And why are you suggesting that backing up from a drive showing signs of failure is desirable? If I see signs of failure, I don't trust the data coming off it. I junk it, either rebuilding the array or recovering from backup.
Yes. I'm really handwaving something I vaguely remember, but the worst MLC drives fail at 1000 writes to a byte. It gets worse on lower nodes (20nm, quicker failure), but the nice thing is there is a little extra space added to compensate for that. Once you hit 3% failures, it's time for a new drive-- you'll begin losing the rest quickly.
There is wear leveling used but that's a touch problem to solve perfectly. I think 2000-3000 writes is about normal.
SLC drives much higher. I don't recall. You'd have to google.
Intel has the best flash controller by far however-- it's the only one that reliably survive random power failures without data corruption or loss.