Slashdot Mirror
HAMR Hard Disk Drives Postponed To 2018 (anandtech.com)
An anonymous reader writes: Unfortunately the hard disk drive industry is not ready to go live with Heat-assisted Magnetic Recording (HAMR). The technology is yet not reliable enough for mass production. Over the years, producers of hard drives, platters and recording heads have revealed various possible timeframes for commercial availability of drives with HAMR technology. Their predictions were not accurate. The current goalpost is set to year 2018. While solid state disks based on Flash memory keep seeing rapid improvements as well, HDDs still kick butt in scenarios where high areal density is more important than ripping transfer speeds. The areal density of HAMR products is predicted to exceed 1.5 Tb per square inch.
thats a lot porn per square inch :P
While solid state disks...
These are NOT disks.
Oh wait ... not yet
The summary lists HDDs as viable vs SSDs when "high areal density is more important than ripping transfer speeds" but in most applications it's the random access time that's more useful and SSDs are better than HDDs in this regard by several orders of magnitude.
Of course it's not reliable, who the hell thought that using a HAMR on hard drives was a good thing? That's what I use to destroy hard drives!
Get us bigger, faster, more reliable SSD storage. HDD platters were as good a solution in their day as CRT monitors were in theirs, but unless I miss my guess time's arrow has flown just a little bit since then, eh? Time for the storage equivalent of going from analog to digital (spinning platters. really. Carry my data on a gyroscope?)
As usual, we need to use STRK drives.
Were you trying to make a point? They very specifically say HDDs are better when storage density is more important than transfer speed. Thus, SSDs having faster access times means nothing.
So... you agree with the author of TFS? They never said anything that contradicts what you said, but there are definitely solutions where storage density is far more important than access speed. Remember, people still use linear tape drives because speed is the least important factor in backup and archival storage.
Slashdot still doesnâ(TM)t support Unicode after it was added to the HTML standard in 1997.
Dear Dice Editors:
If you are going to post a summary which says what *might* be possible in the future, it's helpful to know what the current state-of-the-art is. For example, if you are going to have a summary that says the areal density of HAMR products is predicted to exceed 1.5 Tb per square inch it would be nice to know that Seagate is already shipping a drive with 1.34Tb/in^2 according to Wikipedia.
As it turns out, context matters when giving statistics, or there is no reference to know if the statistic means anything. Given what I found in 30 seconds of using Google, that would mean that HAMR is expected to yield ~12% increase in density from the current state-of-the-art.
You're welcome.
Slashdot still doesnâ(TM)t support Unicode after it was added to the HTML standard in 1997.
Not in all cases, or even many. What do I care if the access time is 0.5 seconds longer for my 20+TB file of my research data? I would rather have it sit on one to two drives ( with backups OFC ) rather than spread across 20+ SSD drives, which, just by the number needed alone are more prone to failure.
In this case transfer speed isn't an issue either, as long as it isn't significantly slower than current HDD tech, since no matter what, data analysis is going to take quite a bit of time and it wouldn't really matter if it took 12 hours instead of 8 to transfer.
In other words, SSD VS. HAMR are made for entirely different use cases. You won't be needing the data density of HAMR for an OS drive ( well maybe you will with the bloat that everyone is putting into OS installs, I mean seriously 5-10GB just for the bare bones OS on Linux, Windows and Mac...) but the density would be useful in places that store and use extremely large data sets.
To err is human; effective mayhem requires the root password!
Actually, if you do more than gaming, there are MANY situations where capacity is far more important than blistering performance.
For example, on one of my home NAS boxes, as long as performance is adequate to stream 1080p video to my STBs, faster drives offer little to no additional value. Therefore the price/performance of spinning disk is FAR more attractive than SSD. I won't even go into all of the situations in my professional life where wasting money on extreme performance would be flat out irresponsible.
SSDs definitely have their place, but spinning disk is going to remain important for quite some time yet.
The drives are ready and work flawlessly, but they're being sued
by (non other than) MC Hammer over trademark infringement.
CATP === 'evermore'
I would love to speed up my Desktop Computer with a fast SSD but we live in an area where every year we have several power outages and from what I've read cutting the power to SSDs is far worse than with HDDs. I've never had a problem where I brick the HDD with a power outage.
Yes I've read that some SSDs come with Power Loss Protection and also a power backup is a possible solution, and I realize that Mobile devices (laptops etc with there own built-in battery backup) are the main uses of SSDs, but I am just puzzled as to the popularity of a component (SSDs) that is so fragile compared to established tech (HDDs).
Yeah. 2TB in a 3.5" drive HDD, and only 500GB in a 2.5" SSD drive. Errr... wait a second, the HDD has the worst storage density! So why would I care about the storage density on the disk or the silicon?
SSDs are better than HDDs for GB/cm3 already, but consumers for the most part don't want to buy $8000 SSDs. (Large companies/data centers are already buying them as fast as they can be built).
Laptop (2.5") drives are limited to about 2TB, and they're 15mm thick. SSDs are hitting 2TB using a 7mm Z-height..
Really, the only play for HDDs is $/GB, which already has nearly evaporated at the low end when you consider all the other contributors to cost in a system.
More data, damnit!
HDDs still kick butt in scenarios where high areal density is more important than ripping transfer speeds.
There is more to SSD than ripping speed, though that is a huge consideration. Denser, slower drives are still useful, but only to a point where the amount of data needing transference, then you start to run into other bottlenecks. Spinning drives are dying, because there hasn't been much improvement over the last 10-15 years (still SATA?) on the one thing that will change in the next couple years, namely bus speeds and getting data from point A to point B, and as dense as HDD are promising to be, it may be better to simply speed up the bus to take advantage of the IOPS available.
The way I have been looking at it for the last several years is fairly unique (IMHO). I see data in terms of distance / speed from core processor, not in density or anything else. Critical data needs to be closer (speed/distance) to the CPU and that is where SSD kick spindle HDDs in the ass. If the CPU is constantly waiting for data, you're wasting cycles.
CPU > L1 Cache > L2 Cache > L3 Cache > RAM > RAM Cached Storage > Local Storage (SSD/HDD) > Network Storage (SSD/HDD) > Cloud Storage > Archival Storage
Keeping the pipelines full of relevant data is key. The faster you can move data to the key point in the pipelines, and that is something HDDs just cannot keep up with any longer, and why people are moving to SSD even though they are less dense. There is a payoff in efficiencies at each level.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
My point was that the performance tradeoff between SDDs and HDDs is more about random access performance (which connotes random I/O) than it is about transfer speed (which connotes sequential I/O).
Remember, people still use linear tape drives because speed is the least important factor in backup and archival storage.
Some people still use pen/pencil and paper and maybe stone tablets ... Sometimes durability and longevity are the most important factors. Just sayin'. Although, to be fair, carving out my Twitter feed in marble is a huge PITA.
It must have been something you assimilated. . . .
You'll care because unless you have 20+TB of memory in your system that allows your 20+TB file to be read in its entirety without any rotational latencies then the per-I/O access time differential between an SSD and HDD will be multiplied by several million I/Os.
Not in all cases, or even many. What do I care if the access time is 0.5 seconds longer for my 20+TB file of my research data?
You may not need instant access to that much data. But then again, if your 1/2/sec longer is multiplied thousands of times a day, five days a week, 4.25 weeks a month ... That is almost 3 hours each month of wasted time, and almost a full week a year of wasted time. My guess, is that you're actually not waiting 1/2 second per, times 1000 a day, it is more likely that you're wasting 3-5 seconds per, several thousand times a day, but since you aren't measuring, you will never know.
However, based on my Anecdotal evidence (personal experience), there is even a greater cost to your short sightedness that you may be ignoring (or ignorant of). Speed is more than just time, sometimes the difference between whether or not you actually do something. And not doing something, because it takes a long long time, is lost opportunity, and opportunities lost can never be recaptured.
And until you actually experience the speed difference, you're not really able to appreciate what it really means. Booting in 10 seconds vs 2 minutes might not seem impressive, until you actually see it.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
I wasn't arguing that capacity vs performance is not a valid trade-off for applications. I was arguing about what the performance trade-off actually is (access time vs throughput for most applications).
Therefore the price/performance of spinning disk is FAR more attractive than SSD.
HDD rarely outperform SSD. I think you're talking about price/capacity of HDD vs SSD. And if my suspicions are correct, this will really start to change in the next couple years (2-3). If my suspicions are correct, you'll see > 16 TB SSD drives that are cost / TB comparable to slower spindle drives within that time frame.
As the SSD densities increase, and the price continues to drop, as the technology continues to improve, the signs are all there, HDD are at the end of their line. My guess, in 5 years, HDD sales will be tanked.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
But but but we have 3D printers and computers got better and like someone was wrong about airplanes once, so therefore anything is possible!?
That's what I've been told right here!
I think you're talking about price/capacity of HDD vs SSD.
Yes, that is indeed what I meant to say. Caught the error too late, and there's no longer an edit button here....
Although, to be fair, carving out my Twitter feed in marble is a huge PITA.
Real men use basalt. Marble is for wimps who only want their carvings to survive a millenium or so.
Areal density doesn't mean shit. Volumetric density is what counts.
*sigh* Yet another person who thinks they know what "most applications" means. The problem isn't that you choose wrong; it's that you actually think there exists a correct answer.
Everyone, stop saying bullshit like "most applications" and start saying "my applications" (and then for bonus points, mention what those applications are).
My dominant application for large storage is video playback. In this application, both the random access time and total throughput requirements are already vastly outclassed by even the lowest-performing drives. Thus: the only thing I really care about are terabytes per dollar, with a secondary consideration for warranty duration. Your blah-blahs-per-second numbers are totally irrelevant to me (for most of the drives that I buy).
I get your point but 2TB 9.5mm 2.5" drives have been around for a while
It appears that 15mm gets you 4TB these days, assuming that's not just two drives crammed in the case
Looks like you can get a 4TB SSD in that same form factor (it's also 15mm if the spec sheet is to be believed)
.
SSD fanbois have been saying this for 10 years, and there is still not the slightest sign of it happening.
Consumer:
HD $33.33/TB
(you couldn't get a 6TB for twice that much a year or two ago)
SSD $317.80/TB
About 10x. OK, so it was 20X or more 10 years ago. Wake me up when it gets below 2X, and I will REALLY pay attention when it gets below 1X.
Enterprise is similarly dismal. Both HD and SSD cost is through the roof for enterprise.
There has never been an edit button on Slashdot. There is a continue editing button before you click submit.
APK likes to ask for responses to the same things over and over. Maybe he just likes the responses?
I think it's helpful to envision hard drives as serving two roles: one for fast, persistent data storage and access (OS, program files, documents) in which you want to prioritize access speed and throughput, and another for mass storage, in which you want as much storage capacity for as little cost as possible. SSDs excel in the first role, while spinning HDDs excel in the latter.
So, I wouldn't characterized spinning HDD drives as "dying"... they're just becoming more specialized, like PCs. The significant improvements have been made to capacity (or cost per GB), which is what's important in spinning drives now - much more than transfer speed. There's only so fast you really need to pull data off a media server, for instance. And capacity, as well as cost per GB, has *dramatically* improved over the past 15 years.
A computer is really just a big set of cascading slower and larger data caches with the ultimate destination as the registers on the CPU. I've also long viewed the computer architecture like that, although I admit I never viewed networks and the internet as part of that system (that definitely makes sense). Here's how I'd revise your graph:
CPU > L1 Cache > L2 Cache > L3 Cache > RAM > Fast Local Storage (SSD) > Mass/Local Network Storage (HDD) > Cloud Storage > Archival Storage
I also wouldn't differentiate between RAM and RAM Cached Storage, because you're making a distinction based on usage type, where everything else is about the type of hardware you're working with.
Irony: Agile development has too much intertia to be abandoned now.
The whole thing hard drives are counting on right now is cramming more data into a device, and at a lower cost, than SSDs. SSDs have yet to stop their progress up the Moore's Law ladder, and hard drives have never been on it. At some point in the not too distant future, cost might be the hard drive's only advantage. Not long after that, all they will have to count on is "SSDs fade if you put them on the shelf too long". The market for archival hard drives is fairly limited. HAMR was supposed to postpone the inevitable just a bit longer, but if they can't get it out the door on time, it may not be worth bothering at all. Those who need massive archives aren't all that concerned about them taking up space, by and large. The current technology is good enough for that.
How is the Riemann zeta function like Trump rallies? Both have an endless number of trivial zeros.
It will come with hurd pre-installed.
Moderating "-1, Disagree" is simple censorship. Have the guts to post your opinion. -- Spazmania (174582)
CPU > L1 Cache > L2 Cache > L3 Cache > RAM > RAM Cached Storage > Local Storage (SSD/HDD) > Network Storage (SSD/HDD) > Cloud Storage > Archival Storage
And what IS that "cloud storage" you ask? Spinning magnetic hard drives. What is "Archival Storage?" Since the Archival Disc has failed to appear, the answer is tape or, for many, magnetic hard drives.
In my personal experience, high latency for user interactive IO, like opening a file, can be jarring and dramatically reduce my work throughput. I can lose my train of thought if the interactivity is not fluid. It really depends on the context.
Ok, how about this. The most common activity on computing devices these days is web browsing. The average, content-rich page has 100+ separate URIs. Browsers cache those resources in discrete files to make browsing faster. I'll give HDDs the benefit of the doubt and say all those files will be close enough to not require anything more than a track/head switch, thus no seek latency. The average rotational latency of a 7200RPM HDD is 4.17ms. So when a user revisits their favorite web site this means 417ms of rotational latencies to load the page of 100 cached URIs, almost a half second. Most current SSDs can do 30K random IOPs a second - many can do much higher but for the benefit of this comparison I'll say 30K, which works out to 33us per I/O. So for the web page with 100 cached URIs the total resource load time will be 3.3ms, as compared to 417ms from an HDD. That's 126x faster.
Interesting that you chose the cheapest HDD you could, but a pretty expensive SSD.
SSDs are at about $240/TB. That's still pretty expensive, but it's still only 7.3x more than HDDs atm.
More so, your "20x or more 10 years ago" is technically correct, thanks to the "or more". In reality though, a decade ago 512*MB* of flash storage (not even a true SSD) cost $40, aka $80,000 per TB. So the gap has come down from 2400 times more expensive per GB to 7 times more expensive per GB in a decade.
In fact, just one year ago, a 128GB SSD cost you $140 - aka $1,120 per TB, or 33 times more expensive.
Based on the rate of decline in the gap, it's not hard to see why people are saying SSDs are going to get into the price range of HDDs pretty soon. Based on the current progression in the cost of SSDs it'll be about 1.5 to 2 years.
Isn't 10 years a small exaggeration ? I think the first real mainstream SSD was in 2009 (intel X25 80GB).
Yes, there used to be. That's been a long time ago, though. Quite possibly long enough that most of the current population never saw it.
That is still a different use case. Large data set analysis is always going to be CPU limited, to the speed you can cram the data through the CPU power you have available and analyse what you need.
These drives are being designed to store very large amounts of data, and on release day should have a smaller failure rate than spanning the whole data set across multiple standard HDDs or SSDs. Every drive you add to an array means another percentage of failure multiplied against all of the other failure chances. This means the more SSD drives you have, the higher the chance of at least one failing, hence what I said about the data being on one or even two drives instead of twenty plus drives.
Not only the original drives are prone to failure, but any backups you make with smaller capacity drives are also prone to failure, meaning more drives for backup redundancy will be needed as well. And remember what happens when you add more drives to the chance of failure?
This is completely ignoring the complexity needed for a large array of drives, FS spanning of the pool, interlink latency between pool nodes, power and cooling usage of large drive pools... ETC.
To err is human; effective mayhem requires the root password!
RAM Cached storage is not the same as RAM, though they use the same medium. RAM Cache storage would be great for holding things like NTFS/FAT table information, so that it didn't need to be read every time a hard drive was accessed. It is a cached (non-working) copy of a Hard Drive.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
Yes, and I'm sure that someone needing massive drive storage space is only going to be doing some light web browsing versus storing a huge amount of data that they will then feed through a ( usually ) slower CPU set. When you need an ungodly huge storage pool you are not going to be _as_ concerned about latency as you will be about storage room and reliability. Here is a hint, these are not going to be markets to casual home users anytime soon. MAYBE they will be marketed to enthusiasts, but it's definitely geared significantly more for the enterprise / professional markets.
For what these drives are designed for, as long as latency and throughput is within an except-able range no one gives a crap how much is there. It all boils down to storage space per unit of area and power consumption. Having ten 5 watt draw drives is still going to cost more that having a single 40 watt draw drive for the same amount of storage space.... just as an example.
To err is human; effective mayhem requires the root password!
That presumes that the only people who buy HDDs vs SSDs are those needing huge storage space. The price disparity between the two extends all the way down to the lower capacities, which is why most systems still ship with HDDs instead of SSDs.
As another commenter pointed out, the 1.5Tbit/in^2 number in the posting (which is taken from the original article) is pretty bogus. Seagate's 2TB 7mm 2.5" drive has an areal density of 1.32Tbit/in^2, and it's probably a safe bet that they (and WD) can wring another 15% density improvement out of SMR technology in the next year or two.
For those commenters bemoaning the fact that the highest density drives today are SMR rather than "regular" drives, get over it - the odds of conventional non-HAMR, non-shingled drives getting much denser than the roughly 1TByte per 3.5" platter we see today are slim to none:
To get smaller bits, you need a smaller write head. That smaller write head has a weaker magnetic field. The weaker field means the media has to be more easily magnetizable (i.e. has lower coercivity). The lower coercivity media needs to have a bigger grain size (size of the individual magnetic domains), so that grains don't flip polarity too often due to thermal noise.
Since a bit can't be smaller than a grain, that means that smaller your write head is, the larger your minimum bit size is. Eventually those two lines cross on the graph, and it's game over.
Two ways of getting out of this are SMR (shingled magnetic recording) and HAMR (heat-assisted magnetic recording):
SMR - stop making the write head smaller, but keep making the bits smaller. Overlap tracks like clapboards on the side of the house (where'd this "shingle" nonsense come from?), allowing small bits with large write heads. Of course this means that you can't re-write anything without wiping out adjacent tracks, which means you need something like a flash translation layer inside the drive, and because of that, random writes might be *really* slow sometimes. (I've seem peak delays of 4 seconds when we're really trying to make them behave badly)
HAMR - Write your bits on low-coercivity media with a tiny, wimpy head, and store them on high-coercivity media with tiny magnetic grains. How do you do this? By heating heating a high-coercivity media with a laser (say to 450C or so) to reduce its coercivity to reasonable levels, then letting it cool down afterwards. But you need a big laser (20mw?) on each head, which causes a whole bunch of problems. Which is probably why they're delaying them.
Oh, and you can overlap tracks on HAMR drives, creating an SMR HAMR drive, with even higher density but the performance problems of both technologies. Which they'll probably do as soon as HAMR hits the market, because with today's SSDs the market for fast HDDs is dying a very quick death.
HDD and SSD drives have different performance characteristics, however the differences in failure rates isn't really all that well known. Yes, if you add drives, you're adding to the number of possible failures, but you're also mitigating the chances of a single point of failure ruining an already bad day. That is why you have Backups and Redundant copies and disaster recovery plans. The more valuable the data, the more money you'll spend securing that data.
You can RAID for protection, you can RAID for increased capacity, and you can RAID for both. The type of drive and failure rates mean very little when you're at that level of Enterprise, except for replacement planning. Personally, I hold onto a 2 - 3 year replacement cycle for all "critical" data drives. 4-6 years on non-critical data drives. No drives older than six years, simply because of increasing failure rates.
And since we're talking failure rates, I suppose you have seen BackBlaze reports on drive failure rates which pretty much mirror my expectations before seeing them. 3 - 4 year spindle drive lifetime before failures start impacting long term data protection. But lets just take your use case (data protection, then speed), carrying 20TB of data will likely use a nice RAID (Enterprise like SAN/NAS) using 4 TB drives (common), you'd need at least 48 drives in RAID hold one 20TB file of research data. Minimum. And if you're doing it right, you'd have two NAS of equal, spanning two different locations in mirror (Disaster Recovery) (Mind you, I'm really ballparking it here). So now you have 96 drives with an average failure rate of 3%, means you'll have three drives fail, this year. That is three times your data is in danger until the drive(s) is/are replaced and it syncs back up (no small feat). At 42 months, the failure rates increase substantially.
All of that is to say, that failure is preventable. I don't treat a single drive failure as a threat per se, unless I am running very thin (no Hot Spare, no Cold Spare, simple RAID). So when you're talking about failures of either SSD or HDD, I don't see the difference, other than as a math problem to solve. People using 4 x 8TB drives in RAID 5 to hold 20TB of data are asking for trouble. First off, they probably don't value the data in terms of Real Dollars (1 Million, 5 million, 10 Million), and are being stingy to save a few pennies, risking millions in effort. It works, until it doesn't, then it wasn't worth anything.
So, when I talk about the value of the data, the time accessing it, the protections to prevent disaster and having proper backup/redundancy/recovery it isn't a simple "4 x 8 TB drives should be enough" calculation. Of course, if you don't have the budget for SSDs or proper NAS/SAN systems to protect your data, then when it fails don't blame THIS IT GUY, go chase the bean counters who don't give a shit.
The TL;DR version, "Good IT is expensive, bad IT is costly"
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
This year is going to be interesting. When you finally have a 16 TB SSD drive released, which is higher capacity and faster than any HDD. Once those are released, the end is just a matter of production ramping up. The other thing most people don't realize is that and SSD has 100,000 IOPS, where a Spindle drive is somehwere under 1000 IOPS, even for the best/fastest drive. Those IOPS count. 100 times faster is a big deal. One second becomes 100 seconds (not exactly, but illustrative). Then you have to get the data off the drive, and SATA is just not going to cut it with SSDs, which means that there is going to be a huge push for increasing Bus speeds to move the data to the CPU. The problem with Standard HDDs is that they are really 20 year old designs that have been milked along long time. SSDs are game changer, yet most people are still just looking at capacity/cost, and not any of the other benefits that SSDs offer.
My point is, SSDs are worth it in just about every case scenario that requires access to data and getting that data to the CPU. The bottleneck is the HDD, not capacity.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
Yes! And Seagate was the st506 company ... And I got a quantum 700mb scsi2 drive
I'm kind of surprised that the hard drive industry has not created bigger (i.e. size, not just capacity) drives. It seems that a large portion of hard drives these days are going into huge arrays in data centers. All the data that needs super-quick access times is moving to SSD. The multi-TB near line data is staying with HDD storage. It seems to me that the industry could put out a drive with something like 5 inch platters; 20 platters per drive; a really good motor; redundant heads per platter; and an extra, backup circuit board. They could fill it with Helium and get the cost down to below 1 cent/GB. No one would put one in their desktop, but data centers might really like it.
If HDD capacities could go up, but HDD makers pivot from just shoveling more bits into less space to redundancy and reliability, HDDs will take the niche that tapes have now. Especially if a HDD maker could guarantee an archive life of a tape, perhaps having a dual-head mechanism (I remember seeing some older drives which actually had two sets of heads, each independent of the other and in an active/active configuration.) Doing this would keep HDDs around as backup media or media to stick in the NAS (for a low-end NAS for stashing archived files or backups, a 5200 RPM WD Red or two are just fine.)
I can see HDD makers also "smartening" up their controllers, by going with more SSHD style drives... mainly by adding SSD as a cache or write landing zone... where the I/O first lands on the SSD, then gets shuffled off to the spinny disks. Of course, there is the slowdown once the SSD winds up full... but done right, it can be a way to help with disk I/O for all but the worst sustained random writes. While doing this, HDD makers might be able to add more features, such as autotiering (think the Apple Fusion drives, except done on the controller) and block level deduplication (although this changes a drive from just a being a raw device into presenting the disk as a LUN, as opposed to just a raw chunk of storage.)
I do agree that SSDs, even now, are pretty much a "why not?" choice for primary storage. Even a USB 3 attached SSD with its limited throughput has an advantage over a HDD, just because multiple tasks don't have to fight for control of the drive heads.
Large SSDs are still 10x the cost per GB of old magnetic drives. SSDs (MLC/TLC) are also starting to run into feature size issues where making the features smaller creates cells that can only survive a few hundred rewrites or less.
SSDs will, however rapidly kill the 15k RPM and 10k RPM niches where performance matters more then bulk cost. SSDs are only 2x-3x more expensive per gigabyte then 15k RPM SAS drives, but offer 25x-100x the IOPS, which means for a given IOPS, you need a lot less spindles (less power, less heat, less cabinets needed). I don't think you'll see 15k SAS drives offered much longer, especially after enterprise class SSDs get below $0.20/GB.
It's the time to get the entire file that matters, so the sum of access time and read time, thus with files beyond a tiny size not "several orders of magnitude" or even a single order of magnitude compared with an array of spinning disks. SSD's are fast but the comment doesn't make any sense unless referring to RAM instead of storage.
For now spinning platters of rust are cheaper for large volumes and not a lot slower in use - but the SSDs are catching up and having a lot of memory is a lot better than both.
It depends - sorts can be I/O limited while filtering, transforming etc is very much CPU bound. Comparisons of data can go either way depending on how much can be kept in memory.
That's often what happens and a large array at the other end of a network link can be much quicker at feeding in the data than a single local SSD (eg. 10Gb/s and 40Gb/s stuff is a lot cheaper than it used to be).
They'll still be a place for spinning rust with that lower IOPS in places like storage of large files with few users (eg. 10 instead of hundreds) until the price of SSDs at large volumes go down. So I don't dispute that they are nice, just not always worth it for now.
It's similar to how you can solve just about any corrosion problem by coating things in gold. Gold costs, but if the price is dropping you can use it more. SSDs used to be the gold plated solution and now they are only that at the large volume size end of town.
With a well designed filesystem that's what memory is for (with an optional extra of using an SSD as cache too if you want but that is a lot slower than memory).
Yes.
Look at the price lists at say the 80GB to 120GB scale which is plenty for an office desktop computer and you'll see it has already happened. A spinning disk is now for if you want to put a lot of stuff on a disk.
Let's take a 64KB file, not small by any means. The fast HDDs presently do just over 200 MB/s sustained for reads. On a 7200RPM HDD the average rotational latency will be 4.17ms. Let's be conservative and say the average seek time only adds 3ms to that, although it's often higher...let's round the sum down to an even 7ms. That's 7ms before the drive starts transferring any data. At 200 MB/s and assuming the entire file is in a contiguous block on the transfer time for the 64KB file will be 312.5us. So the transfer time represents just 4.46% of the total I/O time for an HDD. Now for that same 64KB file on an SSD. Taking an average read IOPs rate of 30K (although most SSDs are much faster), the access time (which is mostly interface overhead) comes to 33.33us. Most SSDs do well above 200 MB/s but let's keep it at 200 MB/s to help the HDD out a bit. So 33.33us access time + 312.5us transfer time = 345.83us. HDD = 7312.5us, SSD = 345.83us. The SSD is 21.14x times faster for a 64KB file. An order of magnitude (1st) is defined as 10x.
That's where we very strongly disagree so you probably don't really get what I wrote about. Even "blank" MS Office documents end up larger than that due to the size of the template file.
So even your tiny file is not "several orders of magnitude" faster to access, since that implies 1000x or maybe 100x at a stretch if used incorrectly instead of "a couple of orders of magnitude".
"Several orders of magnitude" faster - that's what a ramdisk or keeping the stuff in memory some other way is for.
So I'm able to do complete lifecycle calculations of I/O execution times yet I'm not able to understand the difference between what you subjectively call small vs larger I/Os?
Double the I/O size and the SSD is still an order of magnitude faster. Reduce the I/O size to 4KB or 8KB for more random DB workloads or pagefile operations or filesystem metadata fetches and we're at 2 orders of magnitude faster.
Double a tiny file is still a tiny file. We are talking past each other and you don't seem to understand what I wrote.
Consider doing stuff by the GB and you'll get what I meant. Spinning stuff is used for large files now, SSDs won on the small desktop with a web browser.
There are plenty of cases where random access isn't as big an issue, and density is.
First off, yes, random access is good - for random access patterns. Like an OS drive. And games that load lots of little files randomly.
But there are cases where users want more storage, and sheer size of files means they won't be seeing much random access to be a problem. Media storage, for example - if you're storing your 50GB Blu-Ray rips, or just holding movies up that are 1GB or larger, then they would most likely not be very fragmented to begin with, so seek time is less of an issue.And in which case, well, being able to store 6TB or more on a disk is much more useful.
Heck, a lot of people are into movie editing, and movie files are big, and usually only linearly accessed, which means you want to store a massive amount more than random access.
It's really all about the usage now. Bulk storage, hard drives are still the cheapest, while fast storage SSDs have it.
The only problem is, SSDs can be pricey if you want more storage - 2TB hard drives can be had for $50, but are nearly $1K in SSD form, and sometimes you need bulk storage.
Heck, even building something like Android requires a 200+GB source code tree. Holding a few build trees is possible on a cheap 2TB hard drive. Practically impossible on an SSD, even 512GB cheap ones.
Because you are exceedingly bad at mathematics.
Heh... I'm pretty sure there used to be an edit button back when you opened the reply in a new window. I'm not sure how long it lasted but I sort of recall the option to edit - I think it was time limited but I do not recall. I even remember some debate as to why it was taken away but I could be conflating that with another site. I'm thinking early 2000s?
I had an older account back then and I no longer have access to the email nor do I recollect the username but I used to use /. and then took a few years off in the mid-2000s and only posted as AC - if at all. (I was simply too busy. I was also dealing with a bunch of proprietary stuff and didn't want to give indications about the direction the work was taking that could be attached to me.) So, if I posted, it was as an AC and the UID was forgotten, the email lost, and the account is now gone. I think it was a 4 or 5 digit account? I didn't spend much time commenting back then, I was pretty busy.
At any rate, I *think* there was an edit button. I seem to recall that it was in the lower-right and you could click to edit your post. There's some chance that I'm misremembering but I think it was here? Perhaps someone else recalls it as well or can confirm that it was not a feature. I want to say that people were trolling with it and that might have been why they got rid of it? They'd type one thing outlandish, wait for replies, and then type something innocent so that the replies were "wrong" and would be considered stupid.
Meh, I may well be mistaken but that's my recollection. Err... I've put a whole variety of illicit substances into my system between then and now. There's a good chance that my memory is faulty.
"So long and thanks for all the fish."
until the price of SSDs at large volumes go down.
That is what I am suggesting. However, I also believe that we are 2-3 years (short term) away from that. Once you see multiple venders each making 16 TB SSD (first this year), and knowing that HDD aren't likely to reach that size anytime soon, then you'll realize that HDDs are on the cusp of disappearing altogether. This will be especially true if the MTBF of SSDs increase well beyond spinning drives, simply because they do not "wear out" and start failing after 42 months of non-stop use.
It isn't just about initial cost. It is the full Total Cost of Ownership that matters. Spinning drives take more power to spin, cool and otherwise keep running, SSD use much less. I can hardly find a use case scenario except as Tape Drive replacement for HDD (archival) (cold on the shelf backup).
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
We haven't seen a standard faster than SATA 3 emerge for hard drives because there is no need for one. It is already faster than the drives are. SSDs are starting to become available with faster interfaces because they can actually use them; current high end drives can exceed the maximum data rate of SATA.
"HDDs still kick butt in scenarios where high areal density is more important than ripping transfer speeds"
I recently installed a bunch of 4TB SSDs in a server - yes, for ripping transfer speeds but more importantly because you couldn't get 4TB rotating media in enterprise 2.5" format until very recently (and those drives are 12mm thick, vs SSDs being 9mm, which is important for airflow when the 12mm drives draw 3 times as much power)
By the time you can get 4TB 9mm spinning drives, those 4TB SSDs will be 8 or 16TB and based on the current cost curves, the 4TB SSD will be 1/4 the price it currently is (I paid about $1500 for a quantity purchase - yes we get preferential rates but we supposedly get the same rates from SG/WD too).
Current mechanical HDD prices are still higher than they were before the Thai floods, consumer warranties are laughable and reliability has gone out the window (I paid $65 for 2TB drives which have just hit EOL at 48k hours. Their exact replacements are $90 and warranty is 12 months vs the previous generation's 5 years (consumer drives)
On that basis the extra cost of buying enterprise SSD over enterprise spinning media (about 2.2:1 factor currently at 4TB 2.5") is justified by simply looking at labour costs associated with replacing drives which have failed within the warranty period (_EVERY_ Seagate Constellation in our fleet has been replaced twice in that period, all sizes and WD's drives haven't been much better. Now that HGST is officially folding into WD and Toshiba into Seagate, there's even less reason to feel confident about mechanical drive reliability.)
tl;dr: SSDs surpassed HDD _volumetric_ density(*) a while back. the only advantage they still enjoy is upfront price but the higher backend costs and unreliability wipe that out over the typical server drive lifespan.
WD and SG are on their last legs. They're frantically trying to reposition themselves as vertically integrated vendors but enterprise (and consumer) buyer memories are long enough to remember being screwed over and most will happily give money to silicon-based manufacturers for long enough that there will be one last merger before WestSeaDigitalGate crashes and burns. (there will be much rejoicing, but expect one of the SSD makers to buy and keep the brands)
(*) The amount of storage per platter or areal density is immaterial. What counts is the amount of storage in the 2.5" or 3.5" form factor(**). One of the reasons HAMR got delayed is that the larger vertical size of the heads (and the extra heat generated inside the casing during writing) means you can't get as many platters in the case, which more than wipes out any density advantage over shingling, for a lot of extra cost/complexity (someone remind me again how reliable Tahiti drives used to be?)
(**) Those 4TB SSDs in 2.5" cases only occupy 60-70% of the internal space. The only reason Sammy aren't selling larger is that they don't feel there's enough of a market and my purchasing 20 or so wouldn't skew that projection. They've already demonstrated 16TB in 3.5" format and that was the generation prior to what's now on the market. Can't happen soon enough for me.
Only for certain applications. Otherwise you are completely wrong. I work with both SDD/Flash manufacturers as well as HDD/head manufacturers on leading edge technology development especially reliability so I have to specific insight on the limitations of each. (Yes this includes HAMR but also 3D Flash).
If I need to store more data than will fit onto an SSD, I cannot just wait for the process to finish no matter how much faster the SSD is.
Most of my time these days is spent waiting for Java or Javascript garbage collection. I use a 4 drive RAID controlled by an Areca 1210 and most processing throughput is limited be either CPU or sometime the PCIex4 connection to the Areca.