Hamburglar

22k Rpm Sound Good?

Recommended Posts

These aren't cosumer targetted products. SCSI hard drives of any speed never have been. In the enterprise market where time and performance = money, there will always be buyers for bigger and better. Higher RPM drives are not some totally out of left field new technology that we require ridiculous prices to be viable economically.

Share this post


Link to post
Share on other sites

By consumer I mean an entity(a business or individual) who buys a product or service.

Share this post


Link to post
Share on other sites

I'm saying they are going to shrink the platters.

I'm saying that the performance increase derived from moving from a 15K rotational speed to 22K rotational speed is inconsequential.

And I'm saying that, because of the fact that the performance increase is inconsequential, it is probably not worth implementing, and most relevantly, that it is almost certainly not worth implementing from an economic point of view when competitors can achieve virtually identical performance with existing technology.

This is all very simple, and exactly accurate. It is a "No Brainer."

KingGremlin,

Why just shrink the drives, when you can shrink them and speed them up?

Firstly, I have personally posted a clear and exact answer to this question twice already. A detailed quantification of the relative benefits of increasing the spindle speed to 22 000 RPM formed the core of my answer. I don't think a more accurate answer could be derived.

To reiterate, again,

The only way this would be remotely worthwhile was if the cost of designing drives to 22K RPM tolerances was nearly exactly equivalent to the cost of using existing 15K drive. I believe you'd have to use very small platters for the relative costs to be remotely inline.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites
This is all very simple, and exactly accurate.

So you're telling me, that you already know with 100% assurance how these new non-existent drives are going to perform. On top of that you've already done the market research that these companies have done, to know that the increase won't be worth it to anyone, and no one will buy. And even further you are intimately knowledgeable about the internal financials of these companies to know this makes bad financial sense for them. And lastly you know that these companies have weighed all their options of how to improve performance and if theses 22k drives are the future, that they either all made the wrong decision, or intentionally chose a more expensive route to get the same peformance that could have been done with a simple platter shrink? Amazing. I can't argue with that. You win.

Share this post


Link to post
Share on other sites

KingGremlin has a good point--we should give the companies a chance to demo an actual working product before passing judgement.

Sometimes the predictive math and the real-world result to not entirely match, as there may be variables that are not immediately obvious.

Were everything so easy to predict, I would not find the computer industry to be as interesting as I do.

Share this post


Link to post
Share on other sites

I would not care to try to evaluate the validity of the "22K in 24 months" prediction, but it seems to me that 22K could be an element of a successful enterprise design strategy - think 24 - 36 - 48 - + spindle arrays. Obviously there are companies that still find economic sense in producing $750-1,000+ drives, so it seems there is a market for some of these extremes. If the reduction in latency is sufficent to take a notch out of the SSD high-transaction server market, then it may well be worth it. If this proves to be the case, I suspect that it will drive other more pedestrian drives to better performance/price points as well.

I recall when 640K was considered "ridiculous" "exorbitant" "unrealistic" etc....

Share this post


Link to post
Share on other sites
22k sure is a big step. If I was a bettin man, I would've counted on 18k - 20k as the next progression.

Well, all the speed bumps have been ~50%.

5400<7200<10000<15000<22000

So it makes sense.

The speed improvement of 22k over 15k will be comparable to the improvement of 15k over 10k. I wouldn't classify this with the trivial boost from P4 3.2 to 3.4.

Certainly, upcoming cutting edge HDs will all have 2.5" platters. I haven't heard any talk of a new enterprise form factor, so I doubt there will be a move away from 3.5" hh. Then again any 22k drive is at least a couple years away... who knows.

Eh?

72 / 54 = 33%

100 / 72 = 38%

That's close to 1/3 than 1/2.

Share this post


Link to post
Share on other sites
I second Eugene.  The advantage going from 15K to 22K imparts to the rotational latency of a disk is minimal.  I don't think a lot of thought has gone into your excitement about these 'on their way' disks.

You only shave 0.6 ms off the average rotational latency.  That's only 60% of the gain achieved from going from 10K to 15K.  Compare the performance differences between the best 10K drives and the best 15K drives.  In IOMeter where the only advantage is derived, it's only 36%.  Remember that a good portion of that is due to seeking advantages not latency advantages.  Now contrast this performance boost with the usual performance increase between generations within a drive family.  Great...

Just trying to be sensible.

SAN_Guy,

22KRpm is coming but it will only be available in the new 2.5 inch enterprise form factor as the platter diameter needs to be quite small so packing in a 3.5" form factor doesn't make sense.

Of course. 15000 RPM drives are already confined to 2.5" platters. There's nothing new here. As for "packing in a 3.5" form factor doesn't makes sense," they'll pack it in a 3.5" form factor if it is still a standard, just as they did for 15000 RPM drives. As for 2.5" if it becomes a standard, as I hope it does, then, of course, we'll se drives in that form factor as well.

Do well.

Jonathan Guilbault.

I disagree - the thermal design criteria for the 2.5 22K series make it not feasible to do in a 3.5 form factor without modifying the 3.5 standard significantly. The 22K is a 2.5 only product, as are a lot of other drives in the design phase.

The 2.5 standard offers much better cooling as forced-air cooling is required as well as aluminum drive carriers, etc, etc. The 2.5 standard for enterprise fixes all the problems with trying to pack desktop drives into high density enclosures.

SG

Share this post


Link to post
Share on other sites

Sivar,

we should give the companies a chance to demo an actual working product before passing judgement.

Of course. However the benefits of decreasing rotational latency are a known quantity. They are predictable because they are necessary --increasing the spindle speed lowers the rotational latency in a simple inversely proportional relationship.

KingGremlin,

I can't argue with that.

The math is, indeed, very simple.

Finally,

Bear in mind I am simply pointing out that the move to 22K lacks a competitive advantage in terms of cost relative to performance.

Also bear in mind that your entire last post is based on an assumption that companies have actually made plans to release 22K drives. This assumption is based on significantly less information than I have drawn upon to generate my arguments in this discussion. In fact your assumption that companies have concluded contrary to what I have argued is based entirely on the following post by 'Hamburglar':

New higher speed drives in the works. They aren't coming soon, so cool off, hotrod. They'll probably come with serial SCSI, as well as fibre channel and legacy SCSI.

I got this info from an unnamed source at EMC. Sure it's not specific, but it's enough to get the speculation mill a-turnin.

Just inspires confidence doesn't it... It's practically trolling.

KinGremlin, and from this you are arguing KingGremlin, all of this:

So you're telling me, that you already know with 100% assurance how these new non-existent drives are going to perform. On top of that you've already done the market research that these companies have done, to know that the increase won't be worth it to anyone, and no one will buy. And even further you are intimately knowledgeable about the internal financials of these companies to know this makes bad financial sense for them. And lastly you know that these companies have weighed all their options of how to improve performance and if theses 22k drives are the future, that they either all made the wrong decision, or intentionally chose a more expensive route to get the same peformance that could have been done with a simple platter shrink?

I am thinking about the issue. You are not. You are simply talking. In fact I think you are arguing a baseless argument simply for the sake of arguing. You certainly haven't given us any insight into the information on which your chosen position is based.

I am drawning on existing data and very elementary mathematics to derive as much insight as I can. Your arguments however are based on absolutely nothing. Oh wait --you do have Hamburglar's very specific, informed data to back you up... Seagate must have a 22K drive on the way :blink: . Come on. I am just asking you to sit down and think about the simple, quantifiable benefit of changing a drive's rotational speed from 15K to 22K. Think.

And on this,

So you're telling me, that you already know with 100% assurance how these new non-existent drives are going to perform

Spindle speed math is simple. I've already explained it with hard, incontestable data several times. Several times. Stop arguing for the sake of arguing and pay attention.

Do well.

Jonathan Guilbault.

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

This is exactly why the 18K drives that were in prototype never came to market. The performance gain over 15K was very small, the heat issue became a big problem with the current 3.5 form factor, and the platter size had to be cut back to a point where capacity was an issue.

With the 2.5 format being designed for better cooling in high density enclosures and the higher areal densities now available they decided to jump straight to 22K.

I also do know that at least one manufacture has 1.4 inch platter drives running at 30K right now in testing. Performance is supposed to be quite impressive, heat and capacity are not....

SG

Share this post


Link to post
Share on other sites

In a desktop environment, it's certainly true that cache algorithms seem to be more important than outright mechanical performance. As has been mentioned, witness Raptor v. 15k SCSI.

On the other hand, for multiuser workloads, optimized firmware exists, and mechanical performance becomes a major differentiating factor. It's been said that the server performance of top-rated 10K drives lags top-rated 15K drives by roughly 30%. That's 30% more users served, or users served 30% faster. That's non-trivial. There are many applications where a move to SSD is unwarranted, yet where disk is limited by IO performance rather than capacity, and in this market reduced access time is of immediate and pressing concern.

The figure of 0.6ms rotational latency reduction has been tossed around. While that's insignificant on a desktop drive sporting 10ms average latency, it's a significant improvement on a server drive pushing 5ms. If it also comes with smaller platters and further reduced seek time, I don't see why a 20% improvement in average latency is at all unreasonable. For workloads where that 20% improvement really shines through, it will undoubtedly be worth it. For the rest of us, buying more RAM will probably remain the better option.

-Haversian

Share this post


Link to post
Share on other sites

Haversian,

My examples above have used Server data because that is exactly the data that derives the most benefit. It was my intent to demonstrate the minimal advantage that is present in the best case scenarios.

I have examined the mechanical advantage in Server situations quite thoroughly above. Let me summarise it again:

1. 10K rotational latency: 3.0ms.

2. 15K rotational latency: 2.0ms.

3. Difference in rotational latency from 10K to 15K drives: 1.0ms.

4. Difference in seek time from 10K to 15K drives ~1.0ms.

5. Access time advantage of a 15K drive over a 10K drive ~2.0ms

The difference in spindle speed only accounts for 50% of the mechanical advantage you emphasize. A 10K drive with an otherwise identical configuration offers ~87% the positional performance of the 15K drive. The improvement that 22K offers over 15K is even smaller. What does this suggest about the performance advantage of 22K drives over 15K drives? (see my previous posts).

Hopefully the horse is dead now. :blink:

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites
I disagree - the thermal design criteria for the 2.5 22K series make it not feasible to do in a 3.5 form factor without modifying the 3.5 standard significantly.  The 22K is a 2.5 only product, as are a lot of other drives in the design phase.

The 2.5 standard offers much better cooling as forced-air cooling is required as well as aluminum drive carriers, etc, etc.  The 2.5 standard for enterprise fixes all the problems with trying to pack desktop drives into high density enclosures.

What is this 2.5 enterprise form factor? Is there anything online? What timeframe?

Is the Savvio part of this? Why didn't they make it 15K?

Share this post


Link to post
Share on other sites

ericq,

Your questions were discussed a little while ago in this thread.

I think you'll find that it answers most of your questions.

Do well.

Share this post


Link to post
Share on other sites

As for not making the Savvio 15K,

I would argue that their decision was founded, at least in part, on the issues I have argued in this thread.

The positional performance of 2.5" 10K drives is very close to that of 15K drives. It is not advantageous to introduce all the disadvantages of a 15K spindle when the 10K spindles offer 90% of the performance at significantly lower cost, with significantly less heat and power consumption.

In short they likely determined that the tiny advantage of 15K spindle speeds in the performance arena was relatively inconsequential.

This conclusion is especially likely to be derived from consideration of the needs of the high-performance enterprise market. In this market arrays of drives are used to increase Random I/O performance. Using more, less expensive, cooler 10K Savvios achieves a given level of performance more economically than using hotter, more expensive 15K drives.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

Gilbo,

Reiterating your comments on rotational latency adds nothing to the discussion and takes up space. We heard you the first time, and the second time. We're all familiar with diminishing returns, having experienced the same thing with everything from CD burners to RAM.

The main thrust of your argument seems to be that faster spindle speeds are not worth the effort because of the minimal return they provide.

Then you state that the spindle speed accounts for ~50% of the speed advantage 15K drives hold over 10K drives.

All of which suggests that the amount of speed afforded by smaller platters is not worth the effort either, leading to one of two conclusions. Either that further R&D is pointless, or that it only makes sense to combine the two into one package with sufficient gains to justify the effort.

You seem to be suggesting that the latter is foolish, and must therefore be advocating the former, which is rediculous.

Perhaps what confuses me is your apparent, yet unstated, assumption that a 20-30% performance improvement is unworthy of interest, let alone engineering effort. Is that what's going on here?

-Haversian

Share this post


Link to post
Share on other sites
Using more, less expensive, cooler 10K Savvios achieves a given level of performance more economically than using hotter, more expensive 15K drives.

That is an unsupported and not trivially true assertion.

More cooler drives quite possibly generate as much or more heat than fewer hot drives.

Likewise, using more 10K drives uses up expensive rack space and RAID card ports as compared to fewer 15K drives.

Which is cheaper depends on the ratio of various costs, many of which differ from organization to organization. I find your blanket judgement suspect for these reasons.

Share this post


Link to post
Share on other sites

Haversian,

All of which suggests that the amount of speed afforded by smaller platters is not worth the effort either,

I have stated the contrary earlier. Apparently I do need to repeat myself.

Increasing positional performance by shrinking platters offers no disadvantages except in STR, which is a relatively inconsequential consideration in situations where positional performance is useful (Random I/O), and in capacity, which is mitigated by the tendency of the enterprise market to use disks in large arrays. This, of course, makes it a more suitable route for improving positional performance of drives. Earlier I declared:

Virtually equal positional performance could be achieved at such a dramatically lower cost from simply shrinking the platters, without increasing the spindle speed, that I doubt anyone will bother to create such a drive [a 22K drive]. If a hard drive manufacturer did, competitors would be able to offer approximately equal performance at such a price advantage by using equal platters at 15K that the drive would almost certainly be a failure from a profitability stand point.

Please observe the access times of 10K drives with 2.5" platters for an example of the effectiveness of this alternate design philosophy.

It is simply a cost issue. Shrinking the platter costs virtually nothing while improving positonal performance. Increasing the rotational speed costs a great deal while accomplishing nothing more than a platter shrink.

Secondarily,

Additionally your insinuation that "pointless R&D" is not ridiculous is a logical fallacy. Diminishing returns necessitate new approaches. Apprehending this fact is in no way ridiculous. In fact it is the converse.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites
ericq,

Your questions were discussed a little while ago in this thread.

I think you'll find that it answers most of your questions.

Nope, it does not answer my questions at all. I want details of the 2.5 vs 3.5 form factors.

Many of us are fed up with your trolling in this thread and others. If you do not stop repeating yourself, I will file a complaint and encourage others to do so.

You have simply lost the argument. It is obvious 15K has a 35% better access time over 10K. In the absence of more data, 22K should have the same advantage.

Share this post


Link to post
Share on other sites
When almost everything in a computer is electronic and measured in microseconds, 1 ms is an eternity.

I agree, 1 ms is an eternity and this is the area where hard drives are not keeping up with the other components of a pc.

Share this post


Link to post
Share on other sites

ericg,

Nope, it does not answer my questions at all. I want details of the 2.5 vs 3.5 form factors.

What sort of details? There are links to the product announcements. I'm sorry if it wasn't useful to you. If you want height or depth information I would suggest a simply google search.

Many of us are fed up with your trolling in this thread and others. If you do not stop repeating yourself, I will file a complaint and encourage others to do so.

Ericg, I have thoughtfully discussed the issue at hand. I have provided evidence and data to bolster my arguments. I have taken care to actually think about the various elements of the subject. There is nothing improper about that.

I always make an effort to express any point I might have in a fair and carefully considered manner. I don't believe that my posts testify to anything that could be or should be considered as trolling. I usually try and ensure that both sides of any argument are considered.

I'm sorry if any of my posts offend you, but you're being overly sensitive. Quite frankly, this is an intelligent debate, and if debates offend you, you may want to reconsider your participation in a forum like this. Attacking my participation in this thread and others is certainly not particularly constructive.

I offer no insults, no personal criticisms of any kind, simply carefully considered arguments and information. Notably, you're the only person who has offered rudeness and personal insults.

As for this,

It is obvious 15K has a 35% better access time over 10K.

I don't think you appreciate several elements of the issue, but I will not repeat myself --all the pertinent information has already been discussed.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

Oh, Ericg,

This might help assuage your curiousity on the 2.5" hard drive form factor.

It's more related to laptop drives. I don't think there is a lot of information on the details of the enterprise variation. It's not really a standard yet, after all.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

Surprisingly, a 2.5" hard drive is actually 2.75" in width; it's named after the size of the platters it uses. I thought they were actually 2.5".

Share this post


Link to post
Share on other sites

It seems that some folks are misunderstanding Gilbo (at least, there is misunderstanding somewhere). I'm going to break down what I think is Glibo's argument (correct me if I misinterpret), and hopefully I can distill it into something simple to understand -- thus far the thread has been difficult for me to follow. I am in DEU mode right now, so please pardon what may seem to be condescension.

Hard drives' access time is more than just their RPM. In terms of access time, several issues affect the figure:

  • platter size
  • actuators
  • RPM

That list is probably not exhaustive, but hits the big points, as I understand them.

The access time speed difference between 10K RPM and 15K RPM is about 2ms, correct? Rotational latency of 15K drives is 1ms better than 10K drives, so 1ms of the 2ms difference comes from, presumably, either actuators or platter size (probably the reduction in platter size).

A shift from 15K to 22K only brings .6ms rotational latency advantage, and comes at a significant cost monetarily. It would also necessitate a reduction in platter size (to account for issues like flutter of the platters at such high speeds), and possibly data density on the platters, etc. The shift down in platter size might bring another 1ms (I really have no way of estimating this) advantage. Heat and power requirements would probably remain about the same.

Gilbo argues that a competitor of a company that pioneers the 22K RPM drive would merely shrink the size of their current drives, which wouldn't be significantly difficult or expensive, so they would gain the maybe 1ms just as the 22K drives did. And it would lower heat and power requirements, allowing more to be used in the same environment as a 22K RPM drive. The additional spindles in most multiuser situations would make up for the extra .6 ms rotational latency, and would be overall cheaper and possibly more reliable than 22K parts, and depending on myriad factors, may even outperform a 22K drive in implementation. And all the capital expense of R&D is saved.

The underlying assumptions of this argument are:

  • a .6ms latency improvement can be offset by additional spindles
  • it would be significantly cheaper all around to go to smaller platters without increasing RPM
  • marketing doesn't have that much affect WRT higher numbers like RPMs

I do not yet take any particular stance on this issue; I'm undecided. IMHO, the third assumption (which I infer independent from Gilbo's posts) is the weakest. It is quite possible that the demand for "faster" drives (via higher RPMs) might drive a large price bump upwards, even if actual performance doesn't.

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