SSD Price Drops Signaling End of Spinning Media? 646
gjt writes "When Intel and OCZ recently announced new 'affordable' Solid State Disk drives — offering a meager 32-40GB — we initially yawned. But, then we took a closer look at the press releases and the in-progress research and development in SSD technology and opened our eyes. While the new drives aren't affordable on a cost per gigabyte basis for everyone, it does set a precedent — and most importantly a barometer price of $100. And it really does start the death clock for hard drive technology."
...Or an arms race (Score:5, Insightful)
I think HDD will continue to stay enough ahead of SSD in raw capacity that it will stay relevant for a long time. When SSD is affordable at 200 GB then HDD will already be affordable at 2 TB, etc.
Re:In 5 years (Score:5, Insightful)
In 5 years, people will still be maintaining COBOL systems.
Price isn't everything (Score:5, Insightful)
Price is only the first hurdle for SSDs. There's also the issue of reliability, and reports from the field suggest that SSD reliability is highly variable, and in no case as good over the long term as hard drives. That will probably change in time, but they're not there yet.
Careful on Your Terminology There (Score:3, Insightful)
SSD Price Drops Signaling End of Spinning Media?
Blu Ray and CDs are still "spinning media" aren't they? I think I've seen many holographic storage disc products (touted to be THE FUTURE) that were spinning as well. I agree that our mechanical media may be just atop the apex or turning point but our non-mechanical disc based media is most likely set to be a some form of spinning disc [wikipedia.org] for at least a few years longer. If the article thinks that movies and albums will switch to SSD based distribution, I just don't see it happening real soon or even now.
Tiered Storage (Score:2, Insightful)
The clock is certainly ticking, but it's got a long time to wind down. The largest barrier to the death of mechanical storage is the looming halt in NAND geometry shrinks, as processes get so small that it goes from being merely crap to wholly unreliable.
Seeing as how we've got 2TB in single disks now, and that capacity will likely continue to rise, I suspect we'll see capacity increases for SSDs slow for a while as new NVM tech comes online. Instead, prices will simply fall and you'll (hopefully) see some more consumer-oriented hybrid solutions where frequently accessed bits are stored in NAND and large, infrequently files will be out on your (hopefully RAID-6 protected) mechanical storage.
This just in! (Score:5, Insightful)
Helicopters signal the end of automobiles, just as soon as their poor $$/mile traveled ratio reaches parity, but you can buy helicopters from Air Hog right now!
Solar panels signal the end of nuclear power AND the oil industry, just as soon as their poor $$/watt ratio reaches parity! But you can get a solar powered calculator RIGHT NOW!
Can I be a tech pundit yet?
Who really needs SSDs for Porn? (Score:5, Insightful)
I mean really, who needs an expensive big SSD for your porn collection? Unless you have 12 monitors running porn simulcasting...SSD speeds are really only needed for heavily accessed files. HDDs offer cheap storage for those not-so-often used files. The solution is relatively inexpensive, and here today
Interesting assumptions (Score:3, Insightful)
The article seems to assume that a typcial laptop user needs a 120Gig harddisk. I don't think that's true. I can most certainly live with a 20Gig to 40Gig harddisk in a laptop. As a matter of fact, my current laptop (3 year old AMD Turion with "120Gig" HD) has two parts: about 16Gig fro WinXP MCE and the remaining 100Gig for Ubuntu. The 16Gig has all the productivity apps I need + 1 game (Portal), which still leaves me 2Gig free for data. If I didn't have the game, I'd have ~8Gig free for data. For typcial data like word processing documents and the like that is more than enough. It is perfectly usable for day to day tasks. (The Ubuntu part is my playground, but it could live just as wel on a 16Gig partition)
If you enter digital pictures into the landscape, it does change a bit. Still, that's still a lot of pictures. Besides, you don't want all your pictures on the move. They're much safer at home on server and/or NAS.
Music you say? We're talking about "needing"... You don't "need" music on your laptop, unless that's your profession, but that doesn't make you a typcial user.
While I don't think I'm going to shell out 100€ for a 32Gig SDD, because I'm a cheap bastard and what I have works, I could most certainly live with a 32Gig disk in my laptop.
Re:Careful on Your Terminology There (Score:3, Insightful)
Blu Ray and CDs are still "spinning media" aren't they?
To be replaced with network-accessed or network-streamed material. Read-only rotary optical media will be a "way back" story our children will tell our grandchildren. (In other words, my 4-year-old daughter will tell HER 4-year old daughter "I used to watch Dora the Explorer on DVDs.")
Reports of HDDs' demise greatly exaggerated (Score:2, Insightful)
As TFA says, For $125 you get a 40GB SSD. Today on Newegg I can pay 110 for a 1500GB hard disk drive - that's about 40 times more storage, for LESS!
Unless SSD suddenly becomes 40 times cheaper, it's unlikely to wipe out regular HDDs. And it has to cope with the fact HDDs get better every year too.
There has always been a sliding scale in computing with "faster, less storage" on one end and "slower, more storage" on the other.
Cache RAM -- RAM -- Flash RAM -- SSD -- HDDs -- tape.
As time goes on, everything gets faster and everything grows in storage capacity - but they all stay the same relative to each other on the list. Anybody who tells you otherwise is selling you something.
Re:...Or an arms race (Score:4, Insightful)
Re:...Or an arms race (Score:3, Insightful)
Essentially: I expect my next PC to have an SSD for important program files and data, and HDD(s) for big data files which don't need fast random access (e.g. video files). Or I'll offload them to an OpenSolaris server with a bunch of HDDs in a RAIDZ.
The idea that cheaper SSDs will kill HDD is silly when most peoples' storage needs expand to meet whatever they can afford to buy. Certainly they are likely to kill HDDs in simple home and office systems, but for everything else HDDs will continue to be vastly cheaper for at least the next decade or so.
When the latest game I installed on my PC wanted 20GB of disk space, a $100 200GB SSD won't last long.
Re:In 5 years (Score:2, Insightful)
What makes you think that spinning drives don't have a limit on total number of writes?
Re:...Or an arms race (Score:5, Insightful)
SSDs will replace all the small hard drives.
When you get down to small enough drive SSDs will be cheaper per Gig than HDs.
Right now you can buy a 1TB drive for right around $90.
But you can not buy a 5ooGB drive for $45 or a 250GB drive for $22.50. There is a limit to how cheap you can make a harddrive.
At some point SSDs in the 120Gb range will be cheaper than spinning platters. It is probably close right now.
When that happens you will see SSDs replace HDs in that range. That range will keep creeping up and up.
So HDDs will be what you get when you need a lot of storage. Maybe they will eventually be used only for externals and NASs.
Eventually 1 TB SSDs will be cheaper than HDDs but for all I know we will have 100TB HDs for $90.
BTW as someone that paid several hundred dollars for a 30MB HD in 1984 the idea of a sub hundred dollar 100TB HDD just seems like a matter of time.
Re:In 5 years (Score:3, Insightful)
Re:Price isn't everything (Score:1, Insightful)
SSD reliability is highly variable, and in no case as good over the long term as hard drives
SSDs are much like anything else, you get what you pay for.
Buy the cheapest hard drive you can find; it won't last 5 years. Same for CD-R media and SSDs.
It's a pretty safe bet that under normal workloads, a good SSD will outlive just about any HDD.
Re:Reports of HDDs' demise greatly exaggerated (Score:3, Insightful)
Oh, and don't you have those MP3s, Videos, Documents et al of yours?
Re:In 5 years (Score:5, Insightful)
Pity the lesson of Y2K went unheeded - where every COBOL programmer was paid whatever they asked to fix their code, but after should have all been taken out to a field and shot in the head.
Why shoot the programmers? Why not shoot the managers too ignorant to modernize their code base?
To get back on topic, I see spinning drives as the new backup or large file storage medium. You boot off your SSD and keep most of your files there, but anything you want a backup copy of or anything large enough to not need fast access, like movies, pictures, and music get stored on the HDD.
Re:...Or an arms race (Score:4, Insightful)
I think HDD will continue to stay enough ahead of SSD in raw capacity that it will stay relevant for a long time. When SSD is affordable at 200 GB then HDD will already be affordable at 2 TB, etc.
Ah, but when 200 GB of storage is $20, no hard drive will ever be able to be that cheap. There is a fixed minimum cost for building a hard drive. Spindle, motor, etc. It's about $70. When "enough storage" for the average user, let's say 200 GB costs less than that base cost, almost all new storage sold will be SSD devices due to their overall advantages, especially in a battery-powered machine (which are the majority of all computers sold today).
This will completely gut the market for hard drives and R&D into them will cease. All money will move to SSDs and they will improve even more rapidly.
Re:In 5 years (Score:4, Insightful)
I've never seen a consumer hard drive last even 3 years
Maybe you're doing something wrong in that case, because all but one of the five consumer drives in my Windows PC are over three years old and it's still working about as well as a Windows PC ever does.
And personally I've never bought a drive which failed in less than three years (for that matter I've only ever bought one drive which failed before I swapped it out because it had become too small).
Re:Interesting assumptions (Score:5, Insightful)
Music you say? We're talking about "needing"... You don't "need" music on your laptop, unless that's your profession, but that doesn't make you a typcial user.
Fail.
Damned fast, worth the price (Score:4, Insightful)
I received a 128 Gb Kingston SSDnow as a gift from a friend, to put in my laptop. The laptop had a 320 Gb hard drive, so I've had to not lug 2 years of photos around, but it's well worth it because this this is damned fast. Things that had 10 second times now are sub-second. The thing boots Windows 7 in less than 10 seconds.
Capacity is nice, but once you get past 40Gb or so, you only need it to store images and things in bulk. It's like having the speed of a SAN in a laptop. SSD is an order of magnitude faster as far as the user experience goes, and if you can get one for less than $200, it's well worth doing, IMHO.
Once the end users see this in action, the price/Gb won't matter to them, because responsiveness is the name of the game.
Re:In 5 years (Score:3, Insightful)
(not 3.3V fed, 12V at a third the amperage, less heat)
P=IV
So I'm having a hard time reconciling how raising the voltage by 3x (roughly), and using a third less current changes power consumption at all. I'm pretty sure transmission distances and losses are pretty low inside a computer case.
Disks are dying -- AGAIN... (Score:5, Insightful)
But somehow, the rotating storage business manages to innovate its way back to relevance -- Winchester technology, thin film heads, headerless architectures, increased spindle speeds, bigger caches, perpendicular recording, 4k sectors, continuing advances in encoding and ECC, continuing advances in media -- the advances keep coming.
And whatever happened to bubble memory, anyway? Wasn't that supposed to save the day and obsolete rotating storage once and for all? Isn't that what Intel promised us?
Re:This just in! (Score:5, Insightful)
Helicopters signal the end of automobiles, just as soon as their poor $$/mile traveled ratio reaches parity, but you can buy helicopters from Air Hog right now!
Solar panels signal the end of nuclear power AND the oil industry, just as soon as their poor $$/watt ratio reaches parity! But you can get a solar powered calculator RIGHT NOW!
Can I be a tech pundit yet?
Yeah, and LCD's signal the end for CRT's...
Oh wait.
Re:Nikon F6 and FM10 (Score:1, Insightful)
Oh please, Get over your aesthetic snobbery, luddite.
Velvia and Kodachrome produced exposure styles that are fairly trivial to reproduce in both Gimp and Photoshop- Film locks you in to an entire set of aesthetic preferences by brand, Digital gives you a raw capture and leaves it to the photographer to apply the aesthetics after the fact (or Batch Apply, if you find a style you like).
If anything, these days, the 'consumer aesthetic style' of film is being replaced by the camera bodies themselves: Measurebators spend endless hours debating and posting samples from each camera, pointing out the tiniest minutae of style differences so that other people can choose cameras that handle particular situations (specular highlight clipping being the most important to me).
I've been doing artistic and semi-professional photography for a decade and a half, starting by manually developing film from nikon bodies in the 90s, and then moving in to digital photojournalism early this decade. I know a lot of photographers, I am a photographer, and I will absolutely assert that "(digital cameras) compared to film are simply outmatched" is blatantly and patently false. The simple fact is that modern digital sensors capture a technically much larger envelope of information across the spectrum of imaging: More resolution, Larger colorspace, Higer signal to less noise with equivalent, or even Higher Dynamic Range, etc (not even touching the far cleaner workflow, the added advantage of immediate image review for stylistic learning, etc). The simple, linear exposure model of modern digital cameras is far superior for capturing the information actually in front of the cameras lens. With Digital, the color balance and dynamic range of your photograph does not depend on the precise temperature of the film manufacture, and/or developing, something you probably had no control over). Name a SINGLE axis on which Velvia or Kodachrome produce more information about the exposure than your modern full-frame digital sensor. Go on, try, I dare you.
Shooting Film is like shooting JPG, you lock in all of your style information and throw out a huge portion of your exposure data the moment the shutter closes. Maybe if your a luddite who lacks basic post-processing skills, you won't understand that this is all information you can trivially throw out later with the right curve, color balance and exposure profiles to produce the _exact_ same image you would have gotten with film, but at least you have the option to 'undo', or take the image in a completely different direction after the fact if necessary.
Film will remain a luxury niche product for well heeled aesthetic snobs stuck with their highschool dark-room days' photography expertise whom are willing to pay a premium so that they don't have to bother learning how to get the same results from modern technology, but the rest of the world has left yall behind.
Re:In 5 years (Score:5, Insightful)
The initial assumption here is that there was a design flaw in their code. It wasn't a design flaw; the code was simply never designed to be running for this long. In some cases of very old code, it wasn't practical to use a 4 digit date when the code was written. In some cases the programmers warned well in advance that it would need to be fixed but that costs money and business don't willy nilly spend money unless they have to spend money.
Re:In 5 years (Score:2, Insightful)
Just wait till HD pr0n becomes commonplace.
Re:Nikon F6 and FM10 (Score:1, Insightful)
> only two advantages SSD has over spinning media at this time: Access speed and Durability
and NOISE !!
Re:...Or an arms race (Score:3, Insightful)
I thought the same thing, which is why I bought a used 10-tape Exabyte library for my home network. And then I found CrashPlan and its ilk (MozyBackup, etc), and realized:
Statistically, yes, tapes are more reliable than disk - but not perfectly reliable. Which means you need to check them every once in a while to make sure you still have the backups you think you do. And if you're going to fetch the tapes once in a while, it's not any harder to fetch the disks. And if you're fetching the disks every once in a while, it's not any harder (once you've written the software) to maintain the equivalent of a nearline RAID array; any one platter can fail without you losing a byte of data. Now, disks are MORE reliable than your never-verified tapes, and cheaper too.
Oh, and once you're storing your archives on disk, you can do things like automatic x-deltas, versioning, pruning to a grandfather-father-son history, etc. as part of automatic maintenance. And you can maintain eventual consistency with off-site storage. And the list goes on.
Good spinning media is in high demand. (Score:2, Insightful)
Re:In 5 years (Score:3, Insightful)
However lower clock frequencies also require less voltage to keep a stable signal (and correspondingly, high frequencies need to be driven by a higher voltage). Taking this into account, power use does drop much more than linearly when clock frequency drops.
End of HDDs? Not likely soon. (Score:2, Insightful)
Q: "SSD Price Drops Signaling End of Spinning Media?" A: No
Let's face it, a hard drive to hard drive is currently the backup method of choice. Anyone who denies it can be pointed to a plethora of, "Ask Slashdot: How do I store my data?" discussions. Just like when tape drives could store more than the systems hard disk, a hard disk offers to hold more than the average SSD. Never mind the fact that when an SSD fails, it's more than likely end-game for your data. But when a HDD fails, there's any number of data recovery companies at hand to restore it.
The introduction of SSDs will add pep to the computers we use, but hard drives will continue to be the workhorse for storing the bulk of our data for a long while to come.
Re:In 5 years (Score:4, Insightful)
Actually, back in the day, all we booted from was SSD (a few kB of ROM) because spinning media (floppy's and 'hard' drives) were freakin' expensive, not to mention gigantic and slow. This made stuff like instant-boot very normal to have back in the day. In the mean time we decayed to using ever faster spinning media until the hardware couldn't go any faster (15k drives since 1997) and the capacity couldn't increase (perpendicular recording a couple of years ago) and we waited minutes for our OS'es to boot. Now we're back at SSD's which don't really scale very well for large amounts of data (smaller chips means more expensive and more potential errors) until somebody finds a better format for storing large amounts of data cheaply (probably in the realm of 3D optical storage) which will slow us down again a bit.
Re:In 5 years (Score:2, Insightful)
Yeah, if you really want to compare apples to apples, measure MTBF. Oh, and let's not forget the SSD's far superior ability to decay gracefully.
SSDs have a better MTBF, but I think you have the graceful decay backwards. Good SSDs do wear leveling and use SMART to tell you when your ten-thousandth write is approaching. But once they die, they're dead. Solid-state failures are a lot less predictable and more unforgiving than mechanical failures. (For reference, read up on the Poisson Process as it relates to solid-state failures [weibull.com].)
Re:HDDs are Done When Google Says They Are Done (Score:3, Insightful)
A few test installations does not constitute a wholesale change in direction, but they do serve as portents of the future.
Re:In 5 years (Score:5, Insightful)
You are an idiot.
At the time most of that code was written, 32K (words) was a large computer. You SQUEEZED the bits into words tightly. TIGHTLY. People recommended tricks like XORING two pointers together to save space, at the cost of additional computation. And mainframe computer time was in the neighborhood of $700/hour. And that was before several rounds of 12% inflation. At that time a paperback book cost between $0.50 and $0.75, to help you calibrate what that meant.
Also turnaround for many programmers was once per day via courier.
At that time two digit years were the appropriate choice. Four digit years didn't become reasonable, by and large, until the 1980's or even later. (Remember when we moved from mainframes to CP/M computers, our disk storage was trimmed to around 70KB. And our RAM was limited at 64KB. It wasn't until personal computers got hard disks that this limit was lifted. (Networked hard disks came later for most people.)
So for anything written after 1990, you might well have a point, but that's not the code you're dissing. Idiot.
The other respondent who said you should have blamed the managers was more reasonable. Unfortunately current management theory claims that managers don't need to know anything about what they're managing. So the individual managers, themselves, probably aren't to blame. I'd put the blame on the general managers, who should know better than to accept that theory. (Though at their level it becomes true. But a part of their job is to know how the job requirements change as the degree of separation form the actual work increases, and they generally fall down on that. Badly.)
Re:Disks are dying -- AGAIN... (Score:3, Insightful)
Pundits have been tolling the death knell of rotating storage for ... decades?
But somehow, the rotating storage business manages to innovate its way back to relevance -- Winchester technology, thin film heads, headerless architectures, increased spindle speeds, bigger caches, perpendicular recording, 4k sectors, continuing advances in encoding and ECC, continuing advances in media -- the advances keep coming.
And whatever happened to bubble memory, anyway? Wasn't that supposed to save the day and obsolete rotating storage once and for all? Isn't that what Intel promised us?
But despite all of those improvement, over decades, the minimum and average access speeds of drives hasn't improved significantly. A random read still takes milliseconds, which is a simply enormous time for a modern computer to wait. This in turn reduces the throughput in comparison to the streaming speed, making even the 'fastest' hard drives quite slow in practice for most workloads. I've seen benchmarks for 'enterprise' drives doing only a few hundred KB/sec for random reads. This is 2010! Computers can process data just a *tad* faster than that!
In comparison, most SSDs have random read times under a millisecond already, and some PCI-E versions are in the dozens of microseconds range. The random read throughput can be as high as 50% of the streaming throughput, and some drives are closer to 90%.
That's a night & day difference. It's like going from VHS to DVD.
And it's still not fast enough. A modern CPU can process IO data at rates of up to gigabytes per second, easily*. Physical drives have fallen woefully behind, and even SSDs aren't quite there yet unless you RAID a bunch of them.
*) For example, a SUN Thumper, which is just two average quad-core AMD Opterons, can send 1GB/sec (not gigabit, gigabyte) of iSCSI traffic down the wire. That includes reading from disk, decoding the ZFS structures, verifying the hashes, processing the iSCSI commands, and talking to the network card at 10Gbps. In comparison, the best consumer SSDs are still around the 250MB/sec mark.
Re:Not so fast. (Score:3, Insightful)
If I hadn't already spammed this thread with posts I'd have modded you up - other than price/GB, most of the problems in the wiki article are exaggerated IMHO.
I don't get the issue with the difference between read and write speeds though. Hard drives are slower ar writing too, usually about 2/3 the speed of a read for sequential access - all the SSD's I've used have a similar discrepancy (thanks to clever controllers and TRIM making writes far faster).
As to the longevity issue, I guess the jury is still out on that as these things have only been in the wild for a couple of years. But IIRC Intel said you could write 1/3 of the capacity to their drives every day for five years before you start to run out of writable blocks, and even then it's meant to fail gracefully.
I don't understand the point about SATA-based SSD's being slower for writes either; even the crappiest SSD I've used (an OCZ Agility) has random 4k writes somewhere in the order of 8-10 times faster than a 10k SAS drive. Smells like baldercrap to me but I'll go give the sources a read - here's looking forward to your edit!
Re:In 5 years (Score:3, Insightful)
Re:Child pornographers. (Score:1, Insightful)
So how do you know the data was corrupted if the computer wouldn't recognize the drive??
Re:Not so fast. (Score:3, Insightful)
- Limited lifetimes
Funny, that's considered the major downside to platter drives. Anyway, this is the same as the first point.
NO. It’s NOT the first point. The first point is a symptom of this.
Funny? Not so funny when you notice that your music, movies, project backups etc, are all corrupted and lost. Also not so funny that you will notice it way too late, and all your backups will already contain partially corrupted files.
Actually there were viruses who did that exact thing: Write a bit of corruption here, a bit there, slowly, so it ends up on the backup media. If you ever had something like that: It’s the worst possible thing that can happen to your data.
And other than ZFS’s constant scrubbing ECC with checksums and a 3-disk mirror, nothing protects you from it. No RAID, no backups, no checksums alone.
- Performance of SSDs degrades with use.
Solved. See TRIM.
No. TRIM does not solve that problem. Do you even know what TRIM is?
Note that this is also a problem on platter drives.
I notice a pattern in your arguments: You say that there is the same problem with hard drives, and use the false dichotomy that there would only be the two states of “problem” or “no problem”. When in reality, the question is how much of a problem it is. And the answer is, that it’s much worse for SSDs.
- SATA-based SSDs generally exhibit much slower write speeds.
This one doesn't even make sense.
SATA limits the speed. Not only the bus. But also the internal controller. Simple as that.
My irrelevant thoughts (Score:3, Insightful)
From TFA:
This has always been the argument against SSDs, and it's always been wrong. Pundits are under the impression that it should be possible to get speed, capacity, and affordability all in one go. To use a car analogy, they're asking for a 12-cylinder sports car that gets 40mpg and costs under $30k.
For as long as there have been SSDs, consumers have been waiting for lower-capacity versions that were affordable. It's not that hard to do: just take your "low-end" 160GB version, leave a few chips off the PCB, and voila. The manufacturers so far have been hitting the overclocker and enthusiast crowd who will pay any amount of money for the latest and greatest and the companies are just now realizing that hey, average Joes might buy these things too if we can meet a reasonable price point.
I've personally been waiting for an affordable SSD for my laptop and desktop machines but so far the options have been:
1) A mini-PCIe thing that barely holds an OS and doesn't perform any better than a mechanical disk
2) A fast 2.5" SATA SSD with about 4x more space than I need and costs 2x more than I'm willing to pay
I'm one of those people who doesn't store a crap-ton of data on my computers. 95% of the data on my computers is OS and applications. All of my important or bulk data goes on a file server which is accessed over the network and hence does not need to be available within less than a milisecond.
The day someone sells a fast, reliable, low-capacity SSD (20GB is fine) for under $100 is the day I'll buy three.
Re:I think so. (Score:3, Insightful)
LTO-4 are 800 GB uncompressed, 1600 GB compressed, I get a compression ratio of about 1:1.2 - 1:1.4 So I fit between 1 and 1.2 TB on each tape, some data compresses up to 1:1.8. But that's not the advantage of tapes.
Hard drives are extremely fragile and not portable. Drop, mishandle, submerge, freeze or run over a harddrive and then recover data from it. With a tape you can destroy the casing, even cut the actual tape and you can still recover from it. Maintaining a large RAID array on-site doesn't help if your building burns down and maintaining one off site is quite a bit more expensive then a 24 LTO-4 tape loader.
You fail to take into account the cost of actually buying and maintaining a redundant RAID array. An EMC SAN will cost upwards of US10K, an IBM ServerRAID 8i PCI-X card will cost about US$800, even software raid requires a box with an OS to run it with warranty and more hours spent on maintenance. Even if you're mad enough to run a critical business system on consumer hardware a decent PC will set you back A$1500 for a 6 disk system, A$2000 for a 7 disk system as most consumer boards only have 6 SATA slots. Aside from the risk of running consumer disks (3yr warranty) compared to enterprise level tapes (lifetime warranty from IBM and Tandberg) there is little difference in the set up costs and a big difference in the cost of maintenance and life expectancy.
On-line backups are difficult when your data upload rate is limited, I back up 6 TB weekly plus daily differentials which are about 1 TB per week, one set of tapes goes off-site each month. unlimited 10 Mbit Fiber costs A$1500 a month in Australia, 8 LT0-4's cost A$560 and copy the full 6 TB in under 18 hours, that 10 Mbit link would take 52 days to copy that data at it's theoretical maximum transfer rate. Yes we wont have to transfer the entire lot each week if we use a differential or incremental backup but we backup 1 TB a week just in differentials so that's still 8 days. 75 MB/min max theoretical transfer rate for 10 Mbit fibre to the 3.6 GB/min real world I get out of a LTO-4 tape. Granted I get close to the max on the 2 Mbit fiber line we have (A$400 per month).
Tape is cheap, fast, reliable and high capacity. Other backup solutions are cheaper or faster or have higher capacity or more reliable but never manage to combine all these qualities like tape. Optimality I'd like to combine differnt backup solutions so they are redundant and I don't have a single point of failure, but I'd also like a pony.