Re: Seems a little apples vs oranges
Except regular hard drives do seem to show wear-out. After 4 years or so, failure rates tend to climb, sometimes steeply. So much for HDDs wearing better than flash!
247 posts • joined 16 Apr 2011
Except regular hard drives do seem to show wear-out. After 4 years or so, failure rates tend to climb, sometimes steeply. So much for HDDs wearing better than flash!
Execept regular hard drives do seem to wear out after around 4 to 5 years. The failure rate tends to climb, sometimes steeply. So much for being better than flash!
Flash isn't more expensive than 15K hard drives. It's most definitely the other way round. A 1TB flash drive is around $360 today. Sure, it isn't the fastest "enterprise" flash drive, but it's still 1000x faster on random IOPS and 5x on sequential.
A 500GB 15K HDD is around $650.
Puts things in a different perspective!
Chris must be feeling a need to defend the drive vendors after the huge drop in enterprise disk demand this last quarter. 15K RPM drives are dead already and 10K seems destined for the same fate perhaps by the end of this year, though we'll always find people with nostalgic desires to use spinning rust. Violin is essentially right for primary tier storage!
That leaves the thorny question of what happens to the bulk storage tier, The "media evolution" comment above is close to the mark...economics will decide, and if there is a sure profit, the market will fund the foundries. With prices closing fast, the economics start to become compelling even before price parity is reached..
But will we need Chris' trillions. Data compression and deduplication will reduce space demand by as much as 5x. And, once 3D NAND gets out of the lab properly, the incremental cost to add 2 or 4x the layer count will be small. New error correction will move the sweet spot to TLC.
1 Trillion now looks like $40 Billion and possibly less! That's not much given market size.
4x the sequential perfomance isn't much?
Micron's new crosspoint memory may have a place in the sun here, at 1000x faster than flash. Package it in Hybrid Memory cube modules close to the DRAM/CPU/GPU complex and that is a screamingly fast solution - timeframe to able to do that is about 3 years which is when real money starts getting spent on the new super.
Note that HMC could be touching 1 TB/s bandwidth for DRAM in that timeframe, and that's per CPU/GPU module.
NVMe may form the second tier memory, but it really isn't enough on its own. The data has to go to networked storage, so compression accelerators are needed to get the data rate down, assuming there is reasonable compressibility in the data. Then we'll need some rally fast networks to move data out.
Let's get real! The traditional large storage vendors get their gear built by ODMs in China. There is NO secret sauce in the hardware, though quality varies depending on supplier. The same is true of drives. The large vendors do quality control (which costs maybe 1%) and then mark up the drive price 10X.
Appliance based redundancy has made much of the storage mantra redundant. We don't need enterprise drives with 2 interfaces. We don't need RAID.
The mega CSPs figured this out 5 years ago and they buy storage platforms direct from China. They roll there own code, but that's beyond most companies.
Traditional vendors have a lot of code, but it was designed in a different era with proprietary architectures and RAID as the focus. By starting from a blank sheet, Nexenta and others are offering a modern code base with today's design focus, and ultimately that should be better than the traditional code sets.
If you look at the numbers, you'd question that reliability statement. The facts support SSD as being reliable, while HDD show high early life failures and also batch-related or model-related failure.
Now SSD do wear out, but a bit of care in getting the right class SSD to match your write rates will give 8 years useful life, if you need that much.
Any hard drive over 8 years old is wearing out fast, too. In fact failure rates seem to increase after 4 to 6 years of operation and then rise rapidly after a couple more years. IT's a wash on wearout, and late this year we'll see improvements in error-correction in SSD that will increase wear life by as much as 100x. That will put SSD well in front of HDD on reliability!
I think the workstations argument is wrong. Even with the pitifully slow Internet that US Telcos provide, it's a wash to use a cloud-based compute cluster for a job and just have a display on a tablet...there are many articles on that issue - just google Adobe!.
With faster Internet, the balance swings in favor of using cloud clusters instead of workstations, especially as that solves collaboration and parallel working on a job.
Even the gamers are going to the cloud for that reason!
There's no room to remove cost from a drive. We are not going to see any major component disappear form the parts list. The only play for the drive makers is density, but here they are slowing down as the technical barriers for each step get greater.
Hard drives can't get faster, and they are already way slow compared with SSD. Add to that, SSD are lower power, silent and robust.
The demise of the hard drive is inevitable, just as PCs will be replaced by mobiles and tablets. It's happening faster than WD or Seagate would like.
There must be some real dinosaurs out there. Who in their right mind would buy 15K drives? They are 2x the price of top-end SSD and 1000x slower in random IO, which is their only reason for existence.
There are still people who would opt for a KIA even if a Ferrari were the same price!
Barnum was right!
Hilary may have decided to keep her email on her own server so she could "edit" her legacy, or perhaps it was just laziness or typical arrogance, but the real question is whether it was secure.
If she applied all the safeguards big governrment secure operations use it was probably safe, but if, as is more likely, email@example.com had a password like "Chelsea" or "I8Monica" that never got changed, the Chinese probably read her email before she did.
That's the real extent of what she did - not just being a bit naughty, but exposing America's privates to the world!
I've worked extensively with NVRAM architectures. Any of the byte-addressable tape open Pandora's box on software changes (all for the better, I might add).
Just using it as RAMDisk doesn't cut it. The OS virtual file system and SCSI stack are too clunky to even try, and they don't provide granularity or atomicity. Moreover, apps need to understand the NV nature of NVRAM memory to take advantage of it. That means compiler and link editor changes galore.
Then there's the issue of data integrity and a minimum of RAID.
There's a lot more, but you get the point!
This is my hypothesis.
Google cycles hardware on a much faster rate than most corporations. Hardware has a typical 4 year life. what do you do with the old stuff...it's starting to increase failure rates, right. Meanwhile the market for old computers is slow, so selling them isn't to good an idea.
The answer is to use older gear for bulk cold storage where it is rarely powered on. This extends drive life and reduces power drastically. And the nice thing is there's no installation costs!
The economics are compelling. There is no acquisition cost since the gear is depreciated.
The big question is what happens when they run out of old gear!
Is it true the Portuguese have a low-methane donkey now, and have stopped using cars?
"A few times" - that sums up the renewables problem. They are unreliable and that means traditional generation is always running at idle to save the day(light). This is why Europe is paying as much as 5x the price for electricity.
The concept of wind/solar renewables is stupid. The technology just isn't ready for mainstream.
If we want to get CO2 out of the air, we need to accept clean nuclear as the alternative
Violin is calling the changes in the market correctly...and they are pretty unique in doing it. Disk is dying...two or so more years and it's over.
They can deliver what they boast about, too, so as a way of getting the story out this campaign is fine...and a bit of good fun, too.
Why there are so many admins who deny the benefits of flash eludes me. I suppose they are the people who would still buy a KIA if it were the same price as a Ferrari!
With 3D NAND hitting the market and bringing flash and SSD pricing down fast, we can expect capacity parity early next year and price parity by the end of 2016...and that's parity with bulk SSD. In fact, SanDisk projects having 16 TB SSD next year
There are already SSD cheaper than "enterprise" HDD at perhaps 60 percent of the price!
As the spinning drive makers say: "Winter is coming!"
I'm pretty sure it was motion sickness. I felt disoriented on the rollover (which was the standard game entry video. The image quality emphasized the motion and the headset reacted to head tilt etc. Typically, viewers looked sideways to watch the tunnel walls, which definitely contributed to the problem, since they were caught by surprise at the rollover.
As a bit of releif from all the porn experts:
I led a team developing a high-performance headset like Oculus some years ago. Our test bed was a copy of Descent, with very high-res graphics and really fast GPU processing. We found that 1 user out of 3 couldn't take the "roll-over" at the start of Descent, where the user dives down a mine shaft.
We kept a bucket by the test setup!
When you buy a new car, you balance your budget against features an performance, but if I offered you a BMW for the price of a budget family car, which would you buy? There wouldn't be many people saying it's too fast!
That's where we are going with SSD. Next year, at some point, we'll see SDD get as cheap as the cheapest hard drive. With lifetimes well beyond any realistic wear-out, and better reliability, would you buy another hard drive in 2017? You'd be nuts...or you'd swallowed all that FUD hook, line and sinker.
Reality is that we are already making a dumb comparison. SSD are cheaper today than "enterprise" hard drives., and come in much larger capacities. The enterprise HDD is already dead!
I think the math was a bit off.
There are two sorts of problem with loss of data. One is a random failure, and there the BER is near irrelevant since the random failure mechanism that would cause loss of data has to occur if the SAME block is corrupted on two drives. The probability of this is the UBE rate multiplied by the number of blocks on a drive, which is roughly. This multiplies the UBE rate by another 2.5E+08, which means its really unlikely this would happen.
The other problem is a drive failure. Now the question is will a UBE occur during the rebuild time, if there is no parity left. 1E14 is an awful large number of bits, and no consumer operation gets even close to that per day (1E9 is more likely!). (The probability of a second killer BER is actually lower than this on real-world consumer drives, since only a failure in used space should be counted.)
The rules are different on servers
It's this glib, arrogant, "We've already won" attitude that sinks empires.
Cisco is already in trouble....AWS, Google and others use those whiteboxes in huge volumes, and their ODM makers do a great job of quality (heck, they make switches for Cisco too!)
We just need a year for the software explosion that's coming, and it will be Linux versus the mainframe again! We know who won that one!
3D flash significantly raises the bar for replacement technologies. ReRAM looks like the only one, right now, that can easily reach the same data density. ReRAM may be able to mount a challenge on power use and speed, but the Samsung success probably delays introduction even more.
A few years back a comparison of drone crashes showed a big disparity between the Air Force and the army. Turned out the Air Force used pilots, who flew the drones, while the Army used youngsters fresh from the XBox. The pilots had to land the drones themselves, while the kids used a push-button auto-land feature.
The rules haven't changed in 5 years, and highly trained, highly-paid pilots still crash drones. Oh, and they get flight pay while droning! And Army privates earn less than 30 percent of a pilot's base pay.
Does that sound stupid? Never underestimate the pushback old aviators can exert. It needs politicians to apply some serious common-sense.
Once the wandering planet idea went south, why do we need a collision at all? The moon could just be Earth's twin, created by accretion just like earth was, and captured gravitationally
The flash technology being used by Opportunity must be at least 9 years old...we've come a long way since then. Was the wear-out issue so overblown?
It really doesn't matter that shingled drives are only good for archiving. That's the role of all spinning drives in the future, as solid-state gets cheap and very dense.
Intel and others are predicting equal capacities next year, and price parity in 2017. when this happens, the spinners are doomed!
With typical technical vision. the US Federal Communication Commission has just upped US broadband speeds to 24 Mbit/s. The telcos have complained that this is too fast and that no-one really wants such speed!
Oh well! Bring on the lobbyists and open the checkbooks!
Yes indeed, but it's been cold in a lot of other places, Try Europe, Siberia, China.... and polar sea ice is expanding rapidly past record levels.
But wait! The Oracle at NOAA says its the warmest year ever, by 1/100th degree (with a margin of error of +/-50/100ths!), and blithely ignored the fact that the WORLD AVERAGE temperature hasn't really changed for 20 years, refuting their models. They also failed to mention that the "WORLD AVERAGE" is actually not real, since they apply computer corrections that appear to systematically increase the number! And a growing number of scientists cite bias in the measurements towards locations that ARE being warmed by human activity, such as concrete parking lots.
PT Barnum said it all..."You can fool some of the people all of the time!" Clearly, this Floridian didn't see the headline "Coldest decade in US recorded history...and it's only half over!" Of course IPCC would do anything to keep that news from us! The people in Massachusetts would tell you it's true, though!
This would have been an interesting topic 15 years ago, when RAID was our only data integrity option. Today, with erasure codes and replication, the thesis of the article is badly off base (except for the idea that drive replacement is passe) and essentially irrelevant.
We can get the benefit of no-repair storage arrays using erasure codes. This spreads the drives over a set of appliances. Typical configurations protect against a 6-drive loss, which will allow plenty of time for rebuilds, which can be anywhere in the storage pool (with say Ceph or other modern storage software). There is no need for huge numbers of dedicated spares up front, since adding new boxes of drives is the solution for sparing. Replication is not quite as good, typically providing 2-drive failure protection, but again recovery is to spread the data over available space or onto empty drives in a new appliance.
Object storage doesn't require disk-level recovery - objects can be spread over existing free space on many drives.
As a former OEM customer of Supermicro, I can tell you they were a pleasure to work with. Products were reliable, and support excellent.
Mr. Cameron obviously felt there was value in showing his electorate how tough on terror he is ( I understand an election is coming soon!). Sadly, he is clueless as to the encryption game. One solution that has defied interception is the one-time cipher pad, favored in the Cold War by three generations of Russian spies. This still works today, and no amount of legislating or Cameronian bag-piping will fix that.
One might argue that just seeing a string of unrelated numbers in a message is a red flag to intelligence agencies, but the art of steganography makes hiding these very easy, and the explosion of online books makes the selction of source-texts easy too. Cameron should acknowledge that the only people he'd catch would be common citizens not professional terrorists!
The software looks to be around $300 per student. That seems excessive given the huge number of students involved.
Maybe LAUSD should get some of their teachers to write courseware...it would be cheaper by far and also more useful. Of course, they could partner with other school districts that are doing this and save even more! And they could sell on the courses and create a good revenue stream to offset the cost!
Using an alternative thorium-based nuclear cycle, radioactive waste can be reduced to a fraction of the waste from the current uranium cycle. The thorium approach also prevents Fukushima type accidents.
We have enough thorium for 3500 years ready to be mined, and could probably increase that 10-fold with less rich sources.
LED bulbs suffer from the decades of having light bulbs last around a year. No-one is willing to spend 15x the price on the promise of a very long life.
The idea is very powerful,since LEDs are less than 50 percent the power of CF bulbs for twice the light output, but the economics and perception will be a killer until price drops near to the CF range, and legislation/price of electricity pushes the LED...that's going to be a while!
Look for Europe, with their incredibly pricey "green" power, to lead the growth of LEDs.
EMC's logistics and cost structure make competing in the cloud impossible. With big-iron under pressure from flash products and the move to low cost mid range arrays, the hardware business is on the downward slope. This leaves VMware as the core of a potential software empire in a few years time.
If VMware is divested, what is left behind is not too attractive, especially with the ODMs, WD and Seagate moving into the midrange array/appliance space.
Gigabit fibre costs $35 a month in Seoul and is readily available - because there is a fierce competition among several providers.
It's clear that lack of competition means least cost service and high prices.
Well, they can fight over the number of angels that can dance on a sensors head, but the VHS/Betamax fight set the stage for the DVD to wipe them both out!
Will they never learn?
This is clearly a major trend, and unstoppable. It doesn't work for all-flash arrays, since the flash modules are proprietary to save cost and space, and to increase performance.
Note that most servers have just a few drive slots. This doesn't work well for today's HDDs, since the idea is to have a lot (60 drives???) of bulk SATA as the secondary storage behind a flash primary tier.
More detail is needed or this whole idea will rebound!
Commoditization is already moving at a fast pace, but it will accelerate as the ODMs who service the big CSPs start selling branded product and whiteboxes in the US. The barrier to entry on storage is reaching the Linux level, and we all know what happened to mainframes!
I can imagine the debates at EMC. On the one hand you have guys feeling the hot breath of the ODMs on their necks and worrying the price of hardware will drop like a rock, and on the other you have guys trying to embrace the inevitable, and build software to control it.
It looks like the typical reaction of the status quo. Every salesman in EMC is motivated NOT to sell ViPR, since it's cheaper overall and it loses them most of their account control. Result is dismal sales for the product. ViPR will quietly be buried, and everyone will heave a sigh of relief.
Of course, the ODM issue won't go away, and the ViPR folks will be vindicated, but much good it will do them! Better for EMC that ViPR be transferred to VMWare. Now EMC is without a unified storage approach or a good Software-Defined Storage story....Oh wait!, ViPR isn't dead yet!!!!
This seems a bit FUDDY. Internet retail prices of MLC flash drives have already fallen below SAS HDD OEM pricing.I don't know why anyone would use enterprise HDDs.
As to the supply issue. Pricing tells the story. If there were a supply problem, pricing would be going up not down. I agree that flash won't replace bulk disk drives with 5TB capacities any time soon, but enterprise "fast" drives are dinosaurs. There's plenty of flash production capacity for that segment to be satisfied. Note that the replacement rate of flash for enterprise HDD is significantly lower than 1:1 for both drives and total capacity. The SSD are much faster, and don't need short-stroking.
US cities are not much different to say South Korean cities. The rural excuse for lousy broadband doesn't cut it. It's true for Kansas, but not for New York. With most subscribers in cities, the point is even more obvious.
The real driver for the Telcos is getting as much return on their investment...that means no upgrades to the physical plant. Competition is the only force that can drive the issue.
In many countries, usage of this tool requires a warrant. If it is being used illegally, will that invalidate convictions?
There is a partial solution to congestion sitting at hand, and it isn't this crazy electric car thing. Most commuting is done solo, so why not require 1 or 2 person sized vehicles. These would be super-compact, gas efficient, but most importantly, you can park at least 50 percent more cars in the same spaces, and they will effectively reduce traffic by around 30 percent, just by taking up less space on the road. They'd be cheap to manufacture, too.
Apart from giving field engineers a hernia, you have to wonder what this box is for. It can't hold much SSD because there aren't enough interfaces. This means it is a secondary bulk storage system, which begs the question of why SAS redundant interfaces were provided for the drives. SATA would have been the right choice. Front to back depth is going to cause problems, as is the floor load.
This looks a bit like a solution looking for a rapidly vanishing problem. The trend to compact storage boxes with fewer drives, such as FaceBook's 1U 12-drive module, makes much more sense, both use-wise and economically.