Hamburglar

22k Rpm Sound Good?

Recommended Posts

Thanks balding_ape. I think you summarized my points very coherently. Perhaps I didn't explain myself adequately earlier.

Your attention to the marketing issue makes a good point that I certainly haven't really considered. It could certainly be possible, depending on the level of education of the market, to market 22K drives in such a way that they could be profitable. Afterall, people do pay a premium for 15K drives that isn't particularly proportional to the performance offered by such drives.

In my opinion the appearance of this new generation of 2.5" 10K RPM drives suggests that storage providers may be taking a new approach to improving disk speed. The seek times approach those of 15K drives and you can get a lot of spindles working on a problem more easily and cost effectively. Personally, I await their appearance with greater anticipation than I do the possible arrival of 22K disks.

Share this post


Link to post
Share on other sites

I don't think there is any misunderstanding over what he was trying to say Balding, the point is that years and years of progress in the multiple fields within the computer arena have shown the theories to be bogus. If it is faster, and not out of whack more expensive, a large enough market will purchase the product to make it profitable. Is a Radeon 9800XT more than incrementally faster than the previous generation 9800 Pro? No, does it cost a boat load more? Absolutely, and plenty of people still buy it. Are the top end CPU's cost proportionally higher than slower CPU's base on performance? Not even remotely close, do they still sell well? Absolutely. For your theory of additional cost not being worth it, you would have to have an exhorbitantly high cost for these new drives, with companies wouldn't bother making if they thought they couldn't keep costs down somewhat. I bought an original X15 18GB for something like $500, now the current generation are around $170 for 18GB. Back when the orginal Cheetah was released, highend SCSI was well into the 1000's of $'s range. As with everything else, the prices eventually come down. It's not like Seagate is going to try and sell a 36GB 22k RPM drive for $1500 or something. If they did, then I would agree, don't bother, because no one would buy it.

Most of us here understand at least the basics behind hard drive performance, so why you continue to rehash them is beyond my grasp. You can aruge all day long about how higher RPM's will only drop average latency a little, and nothing else will be affected. Again, history tells us, that other improvements and modifications are made to drives when a new rotational speed is introduced. If you don't agree, I'd like to see the arguement with actual factual data on previous generational advances, not your worst case scenario theories of HD makers doing the least possible work with the new unknown drives.

You assertion that higher RPM will lead to lower reliabilty is without merit. Using that theory, 7200RPM ATA drives should be the king of reliability. It's not how fast the drive spins, it's how well it is engineered. I bought an original X15 the week they hit the web and it's been powered on pretty much non-stop since I received it, and it's still running fine. I've killed probably half a dozen ATA drives over the same span. Is there any anecdotal evidence that 10k drives are more reliable than 15k drives? None that I'm aware of.

"a .6ms latency improvement can be offset by additional spindles"

Additional spindles eliminates any cost, heat, space and power saving advantages the lower RPM drives may have had.

"it would be significantly cheaper all around to go to smaller platters without increasing RPM"

True, but the platter shrink is already a foregone conclusion. A platter shrink and a spinrate increase will add additional performance without the above mentioned drawbacks of being forced to add additional spindles.

I wouldn't expect any miracles in performance out of first generation new drives. The X15 though ground breaking in performance didn't just crush everything in sight. That was left to the 2nd generation X15-36LP which really showed the potential of 15k drives. We haven't seen a major jump in 7200RPM drive performance in years, except for the introduction of 8MB cache drives, which has nothing to do with the mechanical workings of the drive. There comes a point when a technology begins to reach its limits and you need to move on to something better. You don't see it as necessary, or don't feel like paying for it? Not a problem, because someone else will.

Share this post


Link to post
Share on other sites

1: Improve either your reading comprehension or the care with which you read. Your response shows that you CLEARLY failed to comprehend a good chunk of my post, and your responses throughout this thread have indicated the same of Gilbo's arguments.

Again, history tells us, that other improvements and modifications are made to drives when a new rotational speed is introduced.

It's hard to see into the future when you walk though life backwards. Look, Gilbo was arguing that this has changed now, precisely because of the diminshed returns from decreased rotational latency. You can still make all those other improvements, just skip the expensive, power hungry, high-heat increase in RPM that will bring a relatively small gain.

If you don't agree, I'd like to see the arguement with actual factual data on previous generational advances, not your worst case scenario theories of HD makers doing the least possible work with the new unknown drives.

See above.

You assertion that higher RPM will lead to lower reliabilty is without merit.

If I had made that assertion, then maybe you would be correct. However, I did not make that assertion, and even if I had, there are some valid grounds suggesting that a cooler drive is less prone to failure than a warmer one. At the very least, it would allow HDD manufacturers to decrease the cost of drives by spec'ing components to lower tolerances. You are straw-manning me. I didn't say that rotation speed makes a drive more or less reliable.

Additional spindles eliminates any cost, heat, space and power saving advantages the lower RPM drives may have had.

For someone who complains (incorrectly) about assertions without basis, you're quite good at making them. On what basis do you make this statement?

There comes a point when a technology begins to reach its limits and you need to move on to something better.

That is precisely the argument that Gilbo advanced. The gains to be had in increasing RPMs have reached their practical limits, and it's time to move on.

Share this post


Link to post
Share on other sites
Sounds as exciting as bumping up the P4 from 3.2 to 3.4 GHz :P

Agreed. Call me when the 0RPM drive (solid state) of atleast 4GB is 'inexpensive'.

Its nice to have faster (and larger capacity) drives, but as long as they're mechanical devices, thier wont be a real exciting shift in the storage industry.

Share this post


Link to post
Share on other sites
I don't think there is any misunderstanding over what he was trying to say Balding, the point is that years and years of progress in the multiple fields within the computer arena have shown the theories to be bogus.  If it is faster, and not out of whack more expensive, a large enough market will purchase the product to make it profitable.  Is a Radeon 9800XT more than incrementally faster than the previous generation 9800 Pro?  No, does it cost a boat load more?  Absolutely, and plenty of people still buy it.  Are the top end CPU's cost proportionally higher than slower CPU's base on performance?  Not even remotely close, do they still sell well?  Absolutely.

While I agree that if someone will buy it, they will make it; your example completely fails to address the issue.

Both high-end video and CPU sales are supported by computer enthusiasts. There are enough of them and they spend unnecessarily.

Generally, however, these people do not pop-off, en masse, and buy the best harddrive out there. Storage is not very sexy, and it doesn't give more frames in Counterstrike. While these same enthusiasts salivate over Western Digital's Raptor drives; two years ago, they wouldn't have gone and bought a SCSI drive with similar performance.

I simply don't see a mass-market 22k RPM drive coming soon. There's not enough market and, as mentioned multiple times, better ways to improve performance.

Share this post


Link to post
Share on other sites

Perhaps I am missing something here, but - forgetting the various notions of whether it would be "worth" it - the two "offset" conditions specified for higher RPM drives, if applied to the RPM drives themselves (more spindles, smaller platters) would push the higher RPM drive further still.

Correct?

I think the debate of "worth it" is different from, and perhaps in some ways confused by, the technical analysis. Well, perhaps not confused by, but perhaps things are being considered in an order that is not so clear.

I think the usual order is more along the lines of: Consider a technology, in this case probably not in a vacuum (other technologies might follow along), consider its advantages, consider what is required to realize that technology (including cost of manufacture), consider if there is a market.

I think there is little disagreement here until we get to the end of that process.

The naysay side has yet to establish an argument that, for example, a 48 spindle small platter 22K array would not have a market. The "yeahsay" side has yet to establish that such would be a viable commercial product. This seems like a manufacturing/marketing question, one on which I may speculate by will admit that I clearly lack the knowledge to address.

Share this post


Link to post
Share on other sites

One point that seems to have been neglected thus far in the debate, by me and others, is that of "sufficient improvement". Customers expect a sufficient performance increase to justify any change on their part. Thus the solidification of market segment based on spindle speed. While it's true that spindle speed is only one part of the performance equation, the other parts scale approximately in sync with that measure. The combination of the two produces steps, if you will, in performance. 5400RPM drives are one market, where heat and noise are a major concern. 7200RPM drives are mainstream and compete primarily on performance. 10K and 15K drives likewise have their markets.

While I readily concede that it is possible to improve the mechanical performance of a drive without increasing its spindle speed, I question the conclusion that this is a sound business strategy. While not precisely analagous, I think the fate of (Fujitsu's??) Pegasus drive, a 12K RPM drive, illustrates that customers are resistant to the idea of moving to a new class of drive without a good increase in speed to go with it. This, I believe, is one of the major factors in the pairing of actuator performance with spindle speed, to compound the advantages of moving from one speed class to the next.

You're correct that an increase to 22k RPM probably necessitates a low areal density and smaller platter diameter. You haven't approached it from the other side though. If one is going to gear up for production of a new smaller-platter form factor, the argument that high spindle speeds necessitates this change becomes moot: the change is being made already. The question then becomes "is increasing the spindle speed at the same time worth it?". My evaluation of the available information suggests the answer is yes. Yours seems to suggest no; I'm comfortable with that.

In any event, presuming the rumors we base our discussion on are true, we will have our answer in some months.

Share this post


Link to post
Share on other sites
Both high-end video and CPU sales are supported by computer enthusiasts.  There are enough of them and they spend unnecessarily.

Are you aware of any data to support this? I would be curious to know if this were true. Based on the ads for the high-end workstations I see, most with the TRUE high end video cards for example - many $Ks, and many of which will not even run games or common apps - they do not seem to be marketed toward "enthusiasts" - whatever they are.

How many "enthusiasts" are running Itantium 2 systems?

Did "enthusiasts" build the reputation of the HP Kayak series?

Are $30K LCD displays for this "enthusiast" market?

Sorry, but I have a great deal of doubt about that notion, frankly. Care to put the annual expenditure by "enthusiasts" against say, the pharmacology industry modeling drug effects and interactions? Against telcos? Etc.

I am open to data that proves this otherwise, but I suspect that if a drive manufacturer is considering a 22K drive - unless it is some sort of hacked SATA drive for $500 - I doubt that the notion of a "PC enthusiast" ever enters the equation.

Share this post


Link to post
Share on other sites

Bald, you didn't directly address any point I made, and seemingly are more interested in degrading me than discussing the topic at hand. I have no desire to participate in such a debate and will refrain from conversing with you any further.

Both high-end video and CPU sales are supported by computer enthusiasts. There are enough of them and they spend unnecessarily.

Bad example by me. Xeons, Opterons, Quadros, and FireGL's are not supported by the enthusiast, they are purely corporate targetted products, that cost even more than the highend enthusiasts products. SCSI is not targetted at the enthusiast, so it makes no difference whether they buy any or not. The corporate world is what SCSI is targetted at, and they will spend more, even a lot more, if they determine the performance increase will make them more money since for them, increased performance is money, not a hobby. The corporate market is significantly larger than the enthusiast market. I'm sure Seagate and others survey the market and get customer feedback about whether or not there is demand for such a product before they begin work on it. If they didn't think they could sell it for a profit, they wouldn't waste the money on R&D to develop it.

Share this post


Link to post
Share on other sites

I'm with those who see no point in going to 22K RPM. For starters, the decrease in rotational latency from 15K RPM is a scant 0.64 ms. Surely it would cost less from both a manufacturing and R&D standpoint to simply make an actuator on a 15K RPM drive faster by that much than to try to develop one that works as fast or faster on a drive with higher rotational speed. Second, you run into noise, heat, and vibration problems, and these all increase in proportional to the square of the rotational speed, meaning they are over twice as bad at 22K RPM versus 15K RPM. Third, the market for such a drive is rather small considering the high price point such a drive would sell at. Fourth, and most importantly, any R&D put into increasing spindle speeds is better spent developing solid state storage. It just makes no sense at all to put that much money into a technology that is approaching obsolescense within a decade, if not sooner. For example, look how little areal densities have gone up in the last year. We had 80 GB platters, and it took over a year to increase density by a scant 25% to 100 GB platters, with only one manufacturer thus far (Seagate) actually bringing such drives to market.

A similar line of reasoning applies to any efforts to decrease access time. I doubt you'll ever make an actuator faster than about 1 ms. The laws of physics prevent it. I also doubt it will be possibly to raise rotational speeds past about 30k RPM (rotational latency = 1 ms). This translates into an access time of what, 2ms? Current 15K RPM drives access a bit faster than 4 ms on average, so you're talking about an ultimate possible less than doubling of current performance regardless of how much money is pouring into R&D. This to me is throwing good money into bad. I for one would rather see a real solid-state drive with an access time measured in nanoseconds, and with no noise, no moving parts, and little heat. Face it, the end is near for magnetic storage. Perhaps it will continue to coexist with solid-state storage until the price per GB of solid state is on par, and then it will disappear. I think the first manufacturer to make a 10 or 20 GB solid-state drive for $100 or less will corner a good portion of the storage market. That's all you need for now-enough space for the O/S and apps on the solid-state boot drive, and a magnetic hard disk for bulk storage. A few years after that, when the price of multi-hundred GB solid-state storage approaches that of the same size magnetic storage, the traditional magnetic hard drive will have passed into the dust bin of history. Given all that, why bother to spend heaven knows how much money to develop a drive that will at most only be 20 or 30% faster than today's best? No, 22K RPM makes no sense at all for the same reasons CD burners stopped getting faster after about 52X. The incremental gains just don't justify the engineering and manufacturing costs.

Share this post


Link to post
Share on other sites

I'll stand with Gilbo, BA, and jtr on this one. The law of diminishing returns makes going much over 15k RPM totally unfeasible from a cost perspective. I'm sure this would be best illustrated by a graph comparing rotational latency to RPM. Unfortunatly, I'm not familiar enough with the equations involved to do so in such a highly-charged topic.

And just to point out some common ground, I'm sure we're all looking forward to higher aerial densities and smaller platters/form factors.

Share this post


Link to post
Share on other sites
Unfortunatly, I'm not familiar enough with the equations involved to do so in such a highly-charged topic.

Approximate it by:

rotational latency = (1 / (SpindleSpeed / 60)) * 0.5 * 1000  
                   =    30000 / SpindleSpeed

Share this post


Link to post
Share on other sites
Bald, you didn't directly address any point I made, and seemingly are more interested in degrading me than discussing the topic at hand.  I have no desire to participate in such a debate and will refrain from conversing with you any further.

My "denigration" of you was no more than telling you what I felt were problems with your posts. It was not designed to make you feel bad or make you look bad. It was designed to point out a problem with your posts. I was more abrupt about it than I should have been, and for that I apologize.

Which point did I fail to address? You made the point about the history of hardware showing us that Gilbo's argument of diminished returns was invalid, but later you contradicted yourself by stating that eventually hardware reaches its limit and a new paradigm must be implemented.

You asked me for evidence, if I don't agree with you. 1) I don't agree or disagree; I merely point out that there is validity to the idea that .6ms improvement in rotational latency will not yield much improvement in performance, and that it is possible that additional spindles may be any or all of cheaper, faster, or cooler. You offer no evidence to contradict that. 2) There is evidence in the thread already about how limited .6ms advantage might be.

You mistakenly thought I had asserted any claim of reliability, and I disputed that I had made any such assertion.

You made the assertion that increasing the number of spindles will overcome the decrease in cost, heat, and power. You offered zero eplanation of this, let alone evidence. I asked you to back up your own assertions -- no higher a burden than you asked for previously, and we attempted to meet.

Share this post


Link to post
Share on other sites

Interesting little debate we have going here. Good show, Gilbo, as usual. I have to say, however, that I don't view the 22k drives with as much impending disappointment as you do in enterprise server applications. In desktop usage, I would agree with you, but just like how 15k drives represented a significant and necessary improvement in IOps and tpm in OLTP applications in 2001, I do believe that 22k drives will do the same in 2006. Add to that the packaging/integration improvements, and you have a winner.

Perhaps the incremental improvement may not be as earth shattering as one would like, but let's look at things holistically. The increase in IOps cannot continue unabated without increasing spindle speed as part of the package. Sure, it is cheaper and more efficient to make increasing spindle speed one of the last resorts, but it is likely that the economics are such that increasing spindle speed will be justified from a cost/performance standpoint in the near future. Just like when gas prices rise above a certain level and causes exploration and drilling of certain areas to become cost effective, the same is likely true about the economics of increasing spindle speed.

The industry feels it is time to make the jump from 15-22k, and they wouldn't do so unless the economics were sound... or at least in the ballpark. Admittedly, however, performance is not the only issue here, as the different form factor is part of the deal as well. But taken as a package deal, 22k/2.5 does makes sense economically and techincally.

... SSA? Yeah, wake me up in a couple years.

Share this post


Link to post
Share on other sites

Most of you are looking at the performance increase from the wrong perspective and in the wrong situations which is why you aren't seeing the reason for continued improvments in spindle speed. You're looking at the absolute time decrease rather than the relative work capability increase. For a single user, the time improvement in loading Quake or surfing the internet will probably be negligible and less than the improvement there was going from 10k to 15k. This, however is not the situation for corporate customers, who don't measure peformance in load times but in transactions per specified time.

Using the previous numbers from Maxtor, the Atlas 15k was 29.9% (rounded to 30%) faster than the 10k IV in SR server benchmarks. That's 30% more I/O's per second. Using the 10k as a baseline will assign its performance 100 I/O's a second, and relabel I/O's as transactions and thus customers. Going to 15k drives allows 30% more transactions (customers) per second which means now they can handle 130 per second. Now lets hypothesize that Maxtor can drag the same 30% improvement when moving to imaginary 22k drives. A 30% increase over 130 transactions is 169 transactions/customers per second. That's an increase of an additional 9 transactions per second over the 30 transaction increase going from 10k to 15k. So while the same percentage increase in performance will yield a smaller absolute decrease in load times, the same percentage increase results in a larger increase in work potential. So depending on usage, it's not a case of diminishing returns, it's a case of increasing returns.

Share this post


Link to post
Share on other sites
Both high-end video and CPU sales are supported by computer enthusiasts. There are enough of them and they spend unnecessarily.

Bad example by me. Xeons, Opterons, Quadros, and FireGL's are not supported by the enthusiast, they are purely corporate targetted products, that cost even more than the highend enthusiasts products. SCSI is not targetted at the enthusiast, so it makes no difference whether they buy any or not.

Amazing that you would push this example, when it's still wrong. Xeons, Opterons, Quadros, and FireGLs are just slightly modified versions of the same high-end video cards and CPUs which are pushed by the enthusiast market. To put it another way, progress is still supported by enthusiasts.

SCSI, as you have pointed out, is not. As shown previously, there are more cost-efficient ways to improve performance. And, dollars being dollars, those ways will be pursued.

Share this post


Link to post
Share on other sites

Uh, no. Those 4 products lines are in no way influenced by the enthusiast market. They use some of the same technologies as consumer products to reduce overall development costs, but the P4 does not drive Xeon development. The P4 Extreme is a repackaged Xeon, and the Athlon 64 FX is a rebadged Opteron 1xx, not the other way around. With or without the enthusiast market, the advancment of the corporate products would still continue.

Share this post


Link to post
Share on other sites

I can't say if 22000 is where the wall is on a mechanical drive, but there definitely is one, so I tend to agree that returns are diminishing in an absolute sense, but it still may be worth it.

One point lightly addressed is that of array performance. With more spindles, you can certainly have more THROUGHPUT, even if the absolute minimum time a transaction is completed in is greater. (This is like running Sun's SPARC crap. Lots of processors which are slower than everyone else's :P ) However, for a GIVEN number of spindles in an array (or a chassis), the higher spindle speed will increase the throughput of the solution, if only by 1020% ballpark. So, if the budget can afford one frame or one drive enclosure with the new expensive drives, but not TWO, then it's still valuable. This example could be scaled, of course.

Also, lower latency can help other situations, like paging, where there's a serious degradation in overall system performance while waiting on disk (although in this case, more RAM is a better answer).

The corporate and enthusiast markets are synergistic. Corporate desires fuel much technology-- enthusiasts demand it at a low cost. Quadros and FireGLs are, however, modified desktop parts, mainly created for games. Far more video processing power is sold to gamers than corps. The opposite is true for storage.

Share this post


Link to post
Share on other sites

A 10K drive with 4.6ms avg. seek, 3ms avg. latentcy you get ~131 IO's per sec.

A 15K drive with 3.6ms avg. seek, 2ms avg. latentcy you get ~178 IO's per sec.

A 22K drive with 3ms avg. seek, 1.4ms avg. latentcy you get ~227 IO's per sec.

It's more feasible to use double the amount of 10K 2.5" drives then a single extreame 22K 3.5" drive.

2x10K drive = 262 IO's per sec.

You can fint 4x 2.5" hot-swap drives into a half height 5.25" (1.75" high) bay.

A case that normally has 6x 3.5" bay can probably fit 12x 2.5" accross with more space behind (eg. for an extra CPU or two on the motherboard). You may yet see the fabled HD with independant heads made of two 2.5" drives placed back to back in a 3.5" case (SAS with internal bridge or 2+ external interfaces).

Two $300 2.5" 10K drives will be better then one 3.5" <25K disk.

When you try to run a 3.5" drive at twice the speed it will likley use 4x the power.

Doubling the use of 2.5" drives will only double the power, but equal the power of a same speed 3.5" drive.

Share this post


Link to post
Share on other sites
A 10K drive with 4.6ms avg. seek, 3ms avg. latentcy you get ~131 IO's per sec.

A 15K drive with 3.6ms avg. seek, 2ms avg. latentcy you get ~178 IO's per sec.

A 22K drive with 3ms avg. seek, 1.4ms avg. latentcy you get ~227 IO's per sec.

It's more feasible to use double the amount of 10K 2.5" drives then a single extreame 22K 3.5" drive.

2x10K drive = 262 IO's per sec.

You can fint 4x 2.5" hot-swap drives into a half height 5.25" (1.75" high) bay.

A case that normally has 6x 3.5" bay can probably fit 12x 2.5" accross with more space behind (eg. for an extra CPU or two on the motherboard). You may yet see the fabled HD with independant heads made of two 2.5" drives placed back to back in a 3.5" case (SAS with internal bridge or 2+ external interfaces).

Two $300 2.5" 10K drives will be better then one 3.5" <25K disk.

When you try to run a 3.5" drive at twice the speed it will likley use 4x the power.

Doubling the use of 2.5" drives will only double the power, but equal the power of a same speed 3.5" drive.

All of this only equates to value, rather than absolute performance. Also there is no way it would use 4x the power-- maybe on startup, but not in normal operation.

Besides, we were talking about 22K 2.5" drives, anyway.

For any given finite space in which to store drives or enclosures, higher spindle speed drives will yield more IOPS. They will ALWAYS have lower minimum service time (except for cache hits). They will probably have more IOs/power used, but that may not be the case, and that would be a strong argument against using them in the enterprise if not. Since power is directly related to heat, then the price in power & heat per extra IOPS may make it less desirable, but there will still be applications due to high IOPS requirements in a finite space, or a requirement for absolute minimum service times. This is regardless of the fact that they may be more expensive to buy and to run. Some applications ignore value and only require performance. Certainly not most, however.

Plenty of corporate buyers just buy whatever their vendors tell them to, and will buy for this reason as well.

Share this post


Link to post
Share on other sites
Unfortunatly, I'm not familiar enough with the equations involved to do so in such a highly-charged topic.

Approximate it by:

rotational latency = (1 / (SpindleSpeed / 60)) * 0.5 * 1000  
                   =    30000 / SpindleSpeed

rotational-latency.png

Just in case anyone was as curious as I...

Share this post


Link to post
Share on other sites

Wow, that graph is so wrong.

4200 rpm = 7.14 ms latency

5400 rpm = 5.55 ms "

7200 rpm = 4.17 ms "

10 k rpm = 3.00 ms "

15 k rpm = 2.00 ms "

22 k rpm = 1.36 ms "

Share this post


Link to post
Share on other sites

Interesting points raised on both sides.

Let me add some business numbers, since that's what we are talking about after all...

The expectation is 170 IOPS per spindle from 22k products. Accepted (general) values for 10k and 15k are 100 and 130, respectively. KingGremlin was right on in this respect. Of course, the final numbers will look different, but I started this thread assuming that speculation would run rampant, and you all did not disappoint.

I propose a hypothetical case that might clarify the business decisions driving 22k:

You are tasked with setting up a new Exchange server for 500 users. You determine that usage patterns for your average Exchange user is .5 IOPS during the week, and peak at 1.0 IOPS on Monday morning. You budget in an extra 60% for future growth making 1.6 IOPS per user. That's 800 IOPS. You decide to use RAID3, so you need 9 10k drives, or 7 15k drives, or 6 22k drives.

Your server will have a 2U chassis with 6 hotswap bays. You would like to avoid an external array cage, saving the rackspace for other projects. If the cost difference between a 10k and 22k solution is <$3000 on a $15000 server, cost is not an issue.

Yes, in a couple years 10k and 15k drives might be a little faster, but that only mitigates my example a bit and maybe not at all. I'm just pointing out the reality of business storage decisions.

Stop trying to apply consumer marketing principles to an obviously business oriented product.

Share this post


Link to post
Share on other sites

You're all still missing the point Gilbo is making: if you reduce the size of the platters of 10K (or 15K) drives (and incorporate any actuator or other improvements), you gain all the benefits of 22K except for the rotational latency improvement. Hence, the difference in iops will be far less than any of the examples you've used. It will be the difference of .64ms for 15K vs. 22K, or 1.64ms for 10K vs. 22K. There will probably be an STR advantage for 22K as well, but it is my understanding that STR is almost irrelevant to multi-user, random i/o.

All that without the power, heat, or (potentially) cost of 22K. So...what is the performance benefit of 1.64ms or .64ms? And what would be the value of being able to add another spindle or two?

Personally, I'm not arguing that 22K is not "worthwhille," nor am I suggesting that it will have no market. I'm merely pointing out that it is possible that 22K is the point at which the gains of increased RPM are no longer economical. Perhaps it is at 30K. But no one so far has made a case that 22K is significantly better than 10K/15K with all the improvements of 22K. And I'm not sure anyone can unless we actually see 22K drives, and see 10K or 15K drives with all the 22K side benefits. LidlessEye has an excellent point about density, of course. 22K allows for more density with equal performance, or should I say for the same volume, more performance.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now