Hamburglar

22k Rpm Sound Good?

Recommended Posts

balding_ape,

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.

This is the crucial point which has been ignored. We are talking about spindle speed and the advantage derived from accelerating it.

KingGremling,

I already did the math earlier. Your calculations are entirely incorrect. You do not isolate spindle speed. Such a simple mistake.. I can only assume you are intentionally delivering misleading statistics to try and bolster your argument. Harmburglar, you offer no data, no math... only arbitrary numbers that you could easily have made up.

Observe, one of my first posts, where I did all this math, the conclusion:

This [the transition from 15K to 22K spindles] requires an 11% performance advantage in situations which benefit the most from positional performance. 11% at a tremendous cost. 22K rotational speeds are significantly more difficult to design for relative to 15K than 15K drives were relative to 10K. Imagine the price premium.

11% more IOPs is a linear extrapolation from existing data. This is all you get from switching a platter from 15K to 22K rotational speeds. This is a fact. It has been explained. Existing data has been gathered and analyzed. I have extrapolated it. The math is correct. Rotational speed, through access time, has a directly proportional affect on Random I/O. There is no give in this math. My calculations are accurate. Yours are not.

In fact due to the exponential decrease in the amount of data/unit of linear seek distance (given by the relationship: pie*r2 : r) 11% is actually too high since a platter shrink will be required. While you lower access time, you also reduce the amount of data available under a given stroke distance. This information is secondary. I believe the hard data above is enough to make my point.

Imagine the competitive advantage a company has in delivering a 15K product that offers 90% (not rounding, exactly 100/111) of the performance of a 22K product at significantly less cost, with significantly less heat and power consumption. I'm not saying 22K drives won't appear, I am simply noting a fact: that, at this point, increasing spindle speed no longer offers a competive advantage. In a capitalist market place a lack of a competitive advantage is not a good thing.

I am disregarding marketing and the lack of education of purchasers that would be required for the effective marketing of a 22K product.

As for the blatant fallacy that KingGremlin keeps tossing out:

"If you shrink the platters you may as well increase the spindles speed."

I don't think a reply to this is even necessary, especially as regards an 11% increase in Random I/O. No matter what the platter size is there are disadvantages to accelerating the spindle speed. Additionally, one encounters another barrier of diminishing returns, that of the exponential decrease in data/"unit of seek distance" this, necessarily ruins the benefits of faster seeking. As platter densities increase this disadvantage of smaller platter drives will be increased. Wow, it's not looking good for 22K drives when one actually breaks the math down.

Hamburglar,

Your density comments demonstrate an apprehension of some of the issues of the target market however, heat and power consumption are the primary concerns in high density applications. It is heat and power consumption that constitute the largest cost. Heat increases cost dramatically by necessitating expensive air conditioning to facilitate its removal. Most people don't understand how much money enterprises spend cooling their computer racks; it is enermous. I am going to google for some articles that appeared about a year ago that quantified air conditioning as a percentage of TCO. Air conditioning actually constitutes a significantly larger fraction of the bill than rack space does. This is the most important reality for most enterprises racks.

ddrueding's graph is right,

by the way, as are your data points E_dawg. Your both right.. how? The labels on the graph lie directly over the data point, not to the right of it as I believe you have read it. If you read it in the manner I describe it appears correct. The graph is correct just easily misinterpreted. However, his point was the curve of the line --and it is unmistakeable.

I think all the important points have been clearly and, where possible, quantitatively addressed now. If you doubt... if you don't like math then look at the picture ddrueding drew.

Do well.

Jonathan Guilbault.

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.

I see that you've dropped the video card example -- because it doesn't work for your argument.

I'm sorry, but even Microsoft Office doesn't need a 4.0GHz P4 to run. And file servers don't need Xeons, because P3s work fine.

Furthurmore, how do you know that the Opteron isn't really an Athlon 64 with Hyperlink and dual channel memory? Or that the P4EE isn't just a regular P4 with more cache?

Let me know when you come around. I'll be waiting.

Share this post


Link to post
Share on other sites
Imagine the competitive advantage a company has in delivering a 15K product that offers 90% (not rounding, exactly 100/111) of the performance of a 22K product at significantly less cost, with significantly less heat and power consumption.  I'm not saying 22K drives won't appear, I am simply noting a fact: that, at this point, increasing spindle speed no longer offers a competive advantage.  In a capitalist market place a lack of a competitive advantage is not a good thing. 

Actually, at this point you have deviated from fact and moved to opinion. You have admitted a "10%" performance advantage, then opined that this would not be a competitive advantage. Obviously, if this is correct, then the drives are not likely to be manufactured.

The critical language here appears to be "significantly less cost" when referring to the 15K option. What is the basis of this "significance"? Are you asserting that there is no chance that this 10% performance boost would make a drive option on an enterprise level viable over an even more expensive storage technology? What is you specific cost formula for making this assessment?

The usual (or at least common) trade off for performance is to pay exponentially for linear (or worse) improvements in performance. This fact tends to support a small, but critical market for high-performance products. Moreover, it seems that this market permits inroads in technology that ultimately leads to more mainstream performance gains.

I think that you can discount the "uneducated user" factor. The market for these hypothetical devices would be highly educated, large scale enterprise systems folks. If they deploy them, it will be because it makes sense.

If curves are to be drawn to make it possible to draw any meaningful conclusion here, they have to consider cost. Specific, numeric, cost, at the same level of accuracy as the performance numbers being bandied about here. Assuming that ddrueding's plot is accurate it is still not possible to draw any conclusion - if there is some (purely imaginary) way to produce the 22K drive below the cost of a 15K drive, well then what? If the 22K drives are 10% more and give back 10% more performance, then what? If the 22K drives are 30% more for the 10%...? Etc.

And I would make the same challenge to those folks who seem fairly certain that the drives will happen.

Share this post


Link to post
Share on other sites

KingGremlin, Hamburglar, and anyone else interested,

Now, to increase signal to noise ratio further. I have done some more research relating to performance/watt. As I have indicated this consideration is primary in the enterprise market. I said it is more important than space savings. Once again there is data and elementary mathematics which confirm my assertions.

Perhaps I should have examined this earlier, but I initially thought my arguments on performance were enough to demonstrate the disadvantages of the pursuit of higher spindles speeds. Additionally, it is only in the past couple threads that those arguing the advantages of 22K spindle speeds have begun to argue on the basis of efficiency and suitability in the Enterprise market --leaving behind an exclusively performance-oriented examination of the technology.

Now,

This whitepaper supports my declaration that power and power-related costs are the considerations of greatest import to data centers. Power related costs constitute 63% of the TCO of a datacenter. Space constitutes only 9% of the cost.

Power: 63% TCO (Power equipment 36%, Cooling Equipment 8%, Electricity 19%)

Space: 9% TCO

Difference (factor): 7x advantage for power (approaching an entire order of magnitude disparity!).

What factor is of more concern to the enterprise market? (Rhetorical question).

Now, let us examine the power requirements of increasing spindle speed, while spindle diameter remains unchanged.

73GB 10K (Savvio): Idle power 5.1W.

73GB 15K (MAS): Idle power 11.3W.

Difference (factor): >2.0 advantage for lower spindle speed.

It is worth noting that this disparity is likely to be further exaggerated in a move from 15K to 22K. 22K drives by extrapolation will require more than twice the electricity of a 15K drive. I would argue that such power dissipation is ridiculous in the context of the enterprise market.

Henceforth I have examined in detail:

1. Performance.

2. Suitability to Enterprise relative to TCO.

I welcome more suggestions for further debate on this topic.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

He - he - we were writing our posts at the same time...I for one at least know who I'd give full marks to on "doing his homework", no matter who ultimately ends up being right....

Share this post


Link to post
Share on other sites

btb4,

The critical language here appears to be "significantly less cost" when referring to the 15K option.

Initial purchase price is irrelevant in the enterprise market. By costs didn't mean simply ASP, I meant all the other costs that I have enumerated many times (heat, power, etc), inclusively. See my last post for details on what are likely to be most significant deal-breaking factors for 22K RPM drives.

As for the argument "it must make sense, because they are doing it." I enumerate the simple facts regarding products relevant to this discussion.

1. There are 10K products with 15K platter sizes announced.

2. There are no 22K products announced.

You will note that the current plans of hard drive manufacturers actually support the converse to KingGremlin, Haversians, and Hamburglar's arguments. The current product picture supports the arguments of those who acknowledge the benefits of 22K drives relative to 15K drives as minimal.

It seems ridiculous to me that anyone could argue "drive manufcaturers intend to bring 22K drives to market" based on Hamburglar's post. I'm not saying that they don't intend to. I am just pointing out that the basis for any argument that drive manufacturers do indeed intend to bring 22K drives to market is nonexistent. There is absolutely no suggestion that this is the case.

Hamburglar,

I don't mean to offend you Hamburglar, I understand that you may have real information, and I also understand that if you did you would not be able to share it with us. I am required simply by virtue of possessing a rational mind to clarify the veracity and usefulness of any information which is being used to support arguments in this debate.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

btb4,

Thanks. I think the initial clarification I give on 'costs' in my last post may be unnecessary. You certainly seem to appreciate what 'costs' I meant.

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.

I see that you've dropped the video card example -- because it doesn't work for your argument.

I'm sorry, but even Microsoft Office doesn't need a 4.0GHz P4 to run. And file servers don't need Xeons, because P3s work fine.

Furthurmore, how do you know that the Opteron isn't really an Athlon 64 with Hyperlink and dual channel memory? Or that the P4EE isn't just a regular P4 with more cache?

Let me know when you come around. I'll be waiting.

FireGL and Quadros are base around the same GPU's, but they are not modified 9800XT's or FX5950's. FireGL isn't even originally an ATi product line, they bought it from someone else. Obviously it makes sense from a financial standpoint to develop a chip that can be used for multiple purposes. That in no way implies that one side is driving the other and there is no evidence you can provide to prove that point. As for the Opteron vs A64. Lets see, the Opteron was released about 8 months before the A64, and the FX was not in the original release schedule. The FX works only with Opteron board and requires registered memory which in no way benefits the user. And right from the start AMD has said it will be moving to another platform (socket 939). How much more obvious does it need to be? The P4EE is a repackaged XeonMP, if you can't see that, then you're one of the few. Look around the web, everyone knows it.

http://www.xbitlabs.com/news/cpu/display/20040312121737.html

"Intel Corporation announced changes in the design of the Intel Xeon and Intel Pentium 4 Extreme Edition processors based on 0.13 micron core known as Gallatin."

Back to the original topic. You're still looking at it from a time saved rather than work gained veiwpoint. As long as the 2 sides are debating from two completely different perspectives there isn't any point in continuing. We'll just have to disagree. Time will tell if these drives ever become a reality and who was right.

Share this post


Link to post
Share on other sites

KingGremlin,

As long as the 2 sides are debating from two completely different perspectives there isn't any point in continuing.

I, and several others, have directly addressed all your arguments with hard data, and careful analysis.

Time will tell if these drives ever become a reality and who was right

All the information we need to evaluate the transition from 15K drives to 22K drives is available to us right now.

Spindle speed is not a complicated issue. It has direct, measurable, predictable effects on all the factors to which it is related.

This is true, not simply of rotational latency, the manner in which spindle speeds affect access time, and therefore Random I/O performance, but also of power consumption. The efficiency of electric motors has been constant for more than a decade (at least... I wouldn't be surprised if someone told me it was 50 years).

Platter size is not a complicated issue. It also has direct, measurable, predictable effects on all the factors to which it is related.

Quite frankly, I believe that your, "time will tell" retort demonstrates more clearly than any of your previous posts that you are simply incapable of acknowledging the very valid arguments that your opponents have put forth.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

On ddrueding's graph. I missed the 4200 RPM disk. That point is a little ( :blink: ) off. The rest seem to be in line though.

The asymptotic progression is there though. And that's what counts.

Share this post


Link to post
Share on other sites

Funny Gilbo,

You could make the same argument about increasing RPM from 10K to 15K. Drive manufacturers could have simply used 2.5 inch platters and high speed heads in 10K designs instead of designing 15K drives, with only about a 1ms difference in absolute performance.

However, they did design 15K drives ... so there must be a good reason for it.

As usual, I will reserve judgement until there are actually 22k drives to test.

Share this post


Link to post
Share on other sites

With all this talk about the next HD drive RPM, it may not be 22K. 20K is a nice round figure so is 22, 24, 25 ... 30. It doesn't realy matter. The manufactures will make their drives performance destictive. They aim the 10K SCSI at a slighlty different market to the 15K SCSI. Each speed has a different set of priorities. They wil make sure that the lower cost drive is significantly slower then the premium drive.

15K, or fastest available

* faster head movement

* smaller platters

* max# platters restricted

* data density restricted by rotational speed and limits in head speed

* less storage

* more heat/power

* less volume manufatured and higher cost components (price premiums on components and manufaturing). tighter tolerances.

* make sure drive is significantly faster (eg. >30%) then the next speed family down.

10K, or lowest SCSI available

* step down from fastest head (reduces power/heat and size of head coil/magnet)

* designed for higher data density and more max. capacity

* lower power, less heat

* cheaper to make in volume and wider tolerances (doesn't use the price premium A1 hand picked parts).

* designed to be drop in replacement for most existing systems.

Share this post


Link to post
Share on other sites

Sound and reliability will be important issues as well. 22K is a damn speed demon.

Share this post


Link to post
Share on other sites
Sound and reliability will be important issues as well. 22K is a damn speed demon.

Actually, I don't think it will be much of a problem.

The 15K drives were designed with only ball bearing motor technology. Since then fluid bearings have been developed. The new 22K drives will probably be designed from the ground up to use fluid bearings. Since fluid bearings have lower flutter, lower noise and probably better reliability then the ball bearing motors, it shouldn't be any worse of a problem then when the first 15K drives were designed.

Share this post


Link to post
Share on other sites

I just hope you're right on this one. At least reliability is paramount on that segment.

Share this post


Link to post
Share on other sites

At what point does it become more economical to just add an extra armature 180 degrees separated from the first one in order to cut rotational latency in half and double STR without the incredibly high RPMs?

Share this post


Link to post
Share on other sites
At what point does it become more economical to just add an extra armature 180 degrees separated from the first one in order to cut rotational latency in half and double STR without the incredibly high RPMs?

When drive makers no longer have to follow standard form factor size constraints. :)

Seriously, while FDB motors have better performance than BB motors, particularly when it comes to things like runout, acoustics, and non-op shock resistance, they apparently take more run current. No doubt further refinements can improve this, but I would think a first-gen FDB design for whatever the next speed bump is would likely have this problem.

Since going with a faster spin speed would likely require more current to begin with, perhaps the first-gen 22K (or whatever) design would start out with ceramic ball bearings and then be refined to FDB?

Share this post


Link to post
Share on other sites
Obviously it makes sense from a financial standpoint to develop a chip that can be used for multiple purposes.

So you see my point. When are you going to either prove yours or come around?

Furthurmore, you've proven that this has little or no bearing on the move to 22k RPMs.

Share this post


Link to post
Share on other sites

It's great you're all using logic and numberic data to backup your arguments but you all forgot something: common sense.

Companies won't shoot themselves in the foot by clashing their 22k and 15k drives in the same market.

They'll improve their 22k drives over their 15k to make the 22ks more attractive.

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 "

How come I never saw a 72k drive with 4.17ms latency :(? Are you just doing the math for roational latency or have I been missing out of nice fast hard drives?

Share this post


Link to post
Share on other sites

qawsedrftgzxcvb,

Latency (rotational latency that is) is different from access time. E_dawg quoted latency, not access time.

They'll improve their 22k drives over their 15k to make the 22ks more attractive.

Or their competitors will improve the competing 15K drives and make them more attractive... Those 10K 2.5" are pretty close to 15Ks in positional performance.

That argument is rather circular.

Share this post


Link to post
Share on other sites
Guest Eugene
Or their competitors will improve the competing 15K drives and make them more attractive... Those 10K 2.5" are pretty close to 15Ks in positional performance.

This is, incidentally, what seemed to happen when Seagate intro'ed the X15. They seemed to keep the 10k cheetahs released around then at a 5.2 ms seek time to maximize the difference... meanwhile, Quantum/Maxtor and Fujitsu forged ahead with faster 10k actuators.

Share this post


Link to post
Share on other sites

So considering my attempt at a graph was so innacurate, would someone who actually knows please post one? I am actually curious on this point.

Share this post


Link to post
Share on other sites

ddrueding: your graph seems accurate, but the RPM labels / tics are misplaced.

E.g.: the 30 ms data point (for 1000 RPM) appears to be labeled as 4200 RPM while going by the tics on the abscissa it looks like being at ~500 RPM.

glug/Mickey: ~8 years ago Seagate actually did manufacture a 3.5" drive with two heads per surface and almost twice the STR of a "normal" drive. Although technically impressive it wasn't a big commercial success.

22K drives are ready and will be release when the market demands them - and is willing to pay the premium. Smaller platters are not really antagonistic - the "sweet spot" for the enterprise market is in the 36 - 72 GB range.

Share this post


Link to post
Share on other sites
Although technically impressive it wasn't a big commercial success.

True. However, I wonder if it's possible to fit two actuators typical of 15K drives (with those huge beefy magnets) in a standard 3.5" form factor. There really isn't much room. Now, if you could use a larger form factor, that would be a different story.

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