Seagate Claims 2.5" SCSI Drive is World's Fastest

theraindog writes "Seagate has announced a 2.5" SCSI hard drive that spins at an astounding 15,000RPM. The Savvio 15K is the first 2.5" hard drive with a 15K-RPM spindle speed, but what's more interesting is that Seagate claims it's the fastest hard drive on the market. Indeed, the drive boasts an impressive 2.9ms seek time, which is more than half a millisecond quicker than that of comparable 3.5" SCSI drives. The Savvio 15K also features perpendicular recording technology and a claimed Mean Time Between Failures of 1.6 million hours."

Comments

laptop use?

[rootofevil]

would this work in a laptop, or would it just get too hot? has anyone seen the operating temp spec?

Re:laptop use?

[ColdWetDog]

How many laptops do you know of that use a SCSI interface?

Re:laptop use?

[morgan_greywolf]

Generally speaking, Seagate's Savvio line of HDDs are intended for server and enterprise storage (read: SAN/NAS) use, not for laptop use. 2.5" hard drives are particularly useful in some compact storage arrays or in blade servers. They probably consume wayyyy to much power for your average laptop. Also, most laptops don't feature SCSI storage. Most use IDE or SATA. It's possible that Seagate could, in the future, come out with a SATA version of this drive, but I don't think it's likely given the power consumption and heat characteristics of 15K RPM drives. Seagates laptop drives don't even break 7.2K.

Re:laptop use?

[spun]

It will get hot. It uses lots of power. It only comes in SCSI. It is for small form factor servers like blades with well engineered cooling systems. These are latop drives in size only.

I've also seen these 2.5" server drives used in cluster heads and RAID/SAN/NAS boxes as the OS boot disk. You can easily fit 16 regular 3.5 disks plus one of these, a slimline CD/DVD and floppy in a 4U case.

Re:laptop use?

[NSIM]

Two problems: 1. It will be considerably hotter and more power hungry than standard laptop drives that spin at roughly 1/3rd of the speed. 2. It has a SAS (Serial Attached SCSI) interface which you don't find in laptops. So, no, you can't stick one in a laptop

Re:laptop use?

[Beer_Smurf]

PowerBook Duo used SCSI.
http://en.wikipedia.org/wiki/PowerBook_Duo
http://www.everymac.com/systems/apple/powerbook_du o/stats/mac_powerbook_duo210.html

Re:laptop use?

[vought]

I'm not sure Serial Attached SCSI is going to work in your Duo/PowerBook 100 series. Kickass as they were, adding a disk drive designed and manufactured twelve years after the last Duo was already discontinued isn't going to help you put off that Mac Book purchase for any longer.

(And yes, I know about the PowerBook 150 and it's IDE drive. Shut up.)

Re:laptop use?

[Kadin2048]

I think the intended application is blade servers. Some blade designs put a disk on the blade itself, so they use 2.5" drives. They're usually designed with good cooling systems and power supplies, so the fact that you can probably cook eggs on it isn't so much of a concern.

It ought to be fairly simple for Seagate to produce the same drive in an IDE or SATA model, by replacing the controller, using the same physical structure and technology, if there's a demand for this in high end "desktop replacement" notebooks...but I don't see it happening.

Re:laptop use?

[dreamlax]

Most use IDE or SATA.

I think you mean PATA or SATA. IDE stands for Integrated Drive Electronics and simply means that the controller for the hard drive is on-board, and requires a suitable host. PATA and SATA are simply two different for the host to communicate with an IDE drive.

We are used to equating IDE and PATA because PATA was the only widespread method of connection between the host and the drive. So while we all understand what you imply by saying "IDE or SATA", it is more correct to say "PATA or SATA". In fact the term PATA was coined only after the invention of SATA.

Re:laptop use?

[jandrese]

That's not to say you won't find a company willing to put one of these (or two, in a RAID 0 array) in a "gamer laptop". Sure it sucks down 5w and is going to get mighty hot unless the ventilation is set up correctly, but that's far from an insurmountable problem. The bigger problem will probably be getting the SAS interface in there. Sure the laptop is going to be better classified as a luggable, but that's hardly unusual with "lan party laptops". If they can figure out how to cram an 8800GTX in there, they can figure out how to cool 5W of dissipation off of the drive.

Re:laptop use?

[NSIM]

Agreed, the issues are not insurmountable, though the SAS one is tricky, as you'd either need to put SAS on the motherboard or hope somebody will put a SAS interface on something like a PC-Card interface, neither of which is particularly likely given the small market for such devices.

Re:laptop use?

[Namlak]

It's possible that Seagate could, in the future, come out with a SATA version of this drive, but I don't think it's likely given the power consumption and heat characteristics of 15K RPM drives

How, exactly, does the power and heat of the drive rule out a certain interface? I run some extensive SATA(big)- and SCSI(fast)-based arrays and they're built on exactly the same platform, the only difference being the backplane PCB.

I'm with ya on the laptop power issue, tho.

Re:laptop use?

[morgan_greywolf]

My bad. What I really meant to say is a 'SATA version for laptop use', not 'a SATA version'. They may very well come out with a SATA version of this drive for other uses. :)

Re:laptop use?

[vought]

Do you see what's wrong here? :-)


Yes, and I feel quite stupid. Consider me chastised.

What can I say? I posted pre-coffee.

Re:laptop use?

[thogard]

A power book 165c and my sparcbook both use 2.5" scsi drives.
I am hunting to a cable to hook that size drive to something else.

Re:laptop use?

[TufelKinder]

I think you reserve the term "pedant" for those who insist
on thinking and writing more precisely than you. — Mike Godwin

Re:laptop use?

[bhiestand]

Pendantic jackass much? You even point out that PATA is a retroactively created term. I'm not sure what this has to do with pendants, but this is like saying we shouldn't differentiate between nuclear, diesel, steam, and wind powered ships because all we had a long time ago were "ships". If I suggested that I needed either "a sailboat or a ship" you would be wondering what I meant by the term "ship".

Re:laptop use?

[monsted]

Actually, 2.5" drives is for all high-performance servers to. Within the next few years, expect to see most (if not all) server hardware move to 2.5" SAS drives. They're cheaper to make, take up less space, use less power and are faster... What's not to like? EMC has already voiced their intention to make 2.5" drive shelves/systems.

SATA 3.5" drives will of course make up the bulk of archival data storage. Modular storage systems will let you mix and match 15-drive 3U SATA and 10-drive 1U SAS shelves.

Breaking the bottleneck

[cpearson]

They just keep chipping away at that Von Neumann bottleneck.

http://vistahelpforum.com/

Re:Breaking the bottleneck

[macadamia_harold]

The term "von Neumann bottleneck" was coined by John Backus in his 1977 ACM Turing award lecture. According to Backus: "Surely there must be a less primitive way of making big changes in the store than by pushing vast numbers of words back and forth through the von Neumann bottleneck. Not only is this tube a literal bottleneck for the data traffic a problem, but, more importantly, it is an intellectual bottleneck that has kept us tied to word-at-a-time thinking instead of encouraging us to think in terms of the larger conceptual units of the task at hand.

So that's where Ted Stephens got his analogy. I had no idea he was such a fan of the Turing awards.

Re:Breaking the bottleneck

[timeOday]

Not really. CPU's outpace storage more and more every year. Reducing seek time from 3.4 to 2.9 ms is only a 15% reduction, and the improvement in sustained transfer is less than that. In the CPU world 15% is nothing, it happens all the time. Even if they somehow repeat this speedup twice per year indefinitely (which they won't), they'd still be falling further and further behind CPUs. Not saying it's a bad product, it's probably the best out there, it just doesn't reverse the trend of storage speed falling further and further behind.

Also, the storage capacity of these drives is small (36 or 73 GB). If they don't hurry up they could be overtaken by flash memory within the next few years.

wow

[mastershake_phd]

and a claimed Mean Time Between Failures of 1.6 million hours.

Thats 182 years.

Re:wow

[pe1chl]

Before you think that this means it has a lifetime of 182 years: this is not the case. The definition of MTBF is not related to lifetime.

Re:wow

[backwardMechanic]

For something that isn't repairable, surely they are related? Lifetime = alpha * MTBF, where alpha is some number less than one? Or are you thinking that the curve is rather broad?

Re:wow

[mike2R]

Before you think that this means it has a lifetime of 182 years: this is not the case. The definition of MTBF is not related to lifetime.

I don't suppose you'd care to explain that a little? I've always assumed Mean Time Between Failure to be what you got if you took a bunch of drives, ran them until they broke, added up the amount of time they worked for and divided by the number of drives. Which would equate to drive lifetime in my book. Am I missing something? 182 years does seem completely insane..

Re:wow

[Amazing Quantum Man]

That means if you have 200 of these bad boys in your data center, on the average one of them will fail every year.

Re:wow

[jimstapleton]

There's this process called extrapolation. It's not perfect, but it'll get the job done.

Basically, you test, say, 1000 hard drives for 2 years and you find:

1 fails in the first 8 months...
1 fails in the next 4 months...
1 fails in the next 2 months...
1 fails in the next 1 month...

even after the first two or three you can expect a mean failure time of 15.5 months. This however does take into assumption a bell shaped probability curve. With enough evidence they should be able to know the shape of the drive-failure-probability curve.

you can follow the pattern to determine

Re:wow

[jimicus]

Don't get too excited. They're being sold in relatively small capacities of 36 and 73GB. The reason for this is that any large business wants maximum throughput from the disks - and the way you do this is by spreading the data across as many disks as you can, usually in a RAID5. Who cares if you've only got 73GB/disk capacity when you'll probably stick a dozen of them in a server and get 730GB capacity (losing one as parity and reserving one as a hotspare)?

The upshot of this is it's quite reasonable to expect a large server room to have hundreds of disks. 182 years MTBF and 360 disks means you can expect about two failures per year.

Re:wow

[mike2R]

Ah, some light begins to dawn. So it's run a number of drives for a set period of time, after this period of time add up the number of hours all the drives have run for and divide by the number of failed drives?

Makes sense, and the comment MTBF != drive lifetime becomes comprehensible, thanks!

Re:wow

[norton_I]

MTBF is only defined within the drives expected life (something like 3 or 5 years). So, if you take 182 drives, you expect about 5 of them to die within 5 years, even if all of them die within 10 years.

Re:wow

[mike2R]

Thanks. I've been wondering for a while why MTBF never figures much in consumer marketing/reviews of drives, it now makes a lot more sense.

Re:wow

[pe1chl]

As also explained by others, it is an average failure rate figure. It does not tell you how long it will take for one drive to fail, but it tells you how many of your drives will fail when you have a large number of them in use.

Re:wow

[whoever57]

For something that isn't repairable, surely they are related? Lifetime = alpha * MTBF, where alpha is some number less than one?

It's a common misconception that MTBF is related to lifetime. In fact they are separate parameters.

Failure rates of electronic and other components are usually modelled as a "bathtub". In this model, there is an initial high rate of failures, which rapidly drops off. This is the "infant mortality" period. Then there is a period where there is a low rate of random failures. MTBF refers to the failure rate during this period. Finally, the failure rate increases as the device reaches end of life. In other words, wearout.

So, as you can see, end of life is related to wear functions, while MTBF is related only to random failures during the device's normal life expectancy.

Re:wow

[jandrese]

Er, if you're throwing spindles at the problem RAID 5 might not be the best solution to the problem. A lot of times it makes more sense to configure it as RAID1_0, even though it's less effective storage space the speed is a lot higher. The complexity is lower too, which is a good thing because even on high end devices it's depressingly common to find bugs in the RAID5 support on controllers, especially once you start pushing them hard for months on end and have hundreds to manage. Every storage engineer dreads having to go to each machine (usually in the middle of the night) and upgrade the firmware yet again on their high priced controllers.

Re:wow

[profplump]

I think I'd much rather know the median time between failures than the mean. It's not that the MTBF is unrelated to lifetime, it's that I don't have enough information to know *how* it's related to lifetime.

Re:wow

[jimicus]

All hardware sucks.

All software sucks.

If you haven't yet learned this, you've not spent enough time in IT.

Re:wow

[backwardMechanic]

Okay, bare with me, I'm thinking about this. My bicycle has a MTBF which isn't related to its lifetime - it breaks and I repair it, until at some point it breaks really badly and I replace it. But if my hard drive dies, it's dead, end of story - in other words I only care about the first failure. The 'infant mortality' stage makes sense, but let's assume that is covered by the guarantee (I know, my data's gone, but I'm bound to have backups ;-) Once we've reached the 'low failure rate' stage, isn't the mean time-to-first-failure related to the MTBF?

Re:wow

[whoever57]

Firstly, you have to understand that these are statistical measures, which don't apply to individual cases.

Mean time to first failure is much more likely to be associated with end of life -- it would be a combination of lifetime measurements and MTBF, but MTBF is typically tiny component in that calculation.

For disk drives, unless you are running a data centre, the MTBF is an irrelevant piece of information. What is much more important is the expected lifetime, which HDD manufacturers do not normally quote. One can get an indication of this from their warranty length -- 5 years in Seagate's case.

Think of it this way: MTBF represents failure rate only during the normal lifetime. Once the normal lifetime has passed, MTBF has no meaning.

Re:wow

[Xabraxas]

Okay, bare with me, I'm thinking about this. My bicycle has a MTBF which isn't related to its lifetime - it breaks and I repair it, until at some point it breaks really badly and I replace it. But if my hard drive dies, it's dead, end of story - in other words I only care about the first failure. The 'infant mortality' stage makes sense, but let's assume that is covered by the guarantee (I know, my data's gone, but I'm bound to have backups ;-) Once we've reached the 'low failure rate' stage, isn't the mean time-to-first-failure related to the MTBF?

I don't think you understand. MTBF doesn't have anything to do with multiple failures when talking about hard drives. It really just means that if you have an MTBF of 100 hours and you have 100 drives, on average one drive will die per hour. MTBF is the inverse of failure rate. MTBF, when describing hard drives, says nothing about shock, heat, or electrical damage outside of the hdd itself. So basically what MTBF tells you about the reliability of a specific drive is NOTHING.

Re:wow

[jlarocco]

Okay, bare with me, I'm thinking about this. My bicycle has a MTBF which isn't related to its lifetime - it breaks and I repair it, until at some point it breaks really badly and I replace it. But if my hard drive dies, it's dead, end of story - in other words I only care about the first failure. The 'infant mortality' stage makes sense, but let's assume that is covered by the guarantee (I know, my data's gone, but I'm bound to have backups ;-) Once we've reached the 'low failure rate' stage, isn't the mean time-to-first-failure related to the MTBF?

Think of it this way: If you buy 100 drives, you can expect 50 of the drives to die before the MTBF, and the other 50 to last some amount of time after the MTBF. Pondering the implications for a single drive is meaningless.

Re:wow

[drsmithy]

Don't get too excited. They're being sold in relatively small capacities of 36 and 73GB. The reason for this is that any large business wants maximum throughput from the disks - and the way you do this is by spreading the data across as many disks as you can, usually in a RAID5. Who cares if you've only got 73GB/disk capacity when you'll probably stick a dozen of them in a server and get 730GB capacity (losing one as parity and reserving one as a hotspare)?

If it's performance you're interested in, you use RAID10, not RAID5.

Indeed, with dual parity schemes (RAID6, RAIDZ2, RAID-DP) rapidly becoming du jour, it's hard to see a purpose for RAID5 in any array of meaningful size, for people who are interested in reliability (and if you're not, you may as use RAID0 and take the additional performance).

Personally, I think anyone using RAID5 on an array bigger than 6 or so disks - or a couple of terabytes - is insane.

(Then there's ZFS, which in one fell swoop has damn near obseleted the whole idea of hardware RAID arrays and volume management. Awesome work, Sun.)

Back to these little drives, it's hard to see a real need for them. People prioritising performance and/or capacity are almost certainly better off with 3.5" drives in pretty much any given situation. Blades (the most obvious application) generally use a SAN or NAS for their data storage needs and only use local disk to boot from (if they even have local disk at all). The use of such drives in 1U and 2U servers (eg: Sun's Opteron boxes) is a bit silly, IMHO, because it restricts their usefulness as standalone machines (due to limited usable space) without really providing any significant benefit (further exacerbated, in the case of Sun, by their onboard "RAID" controller not supporting RAID5 or RAID10 - dumb, dumb, dumb, DUMB).

Re:wow

[WuphonsReach]

As others have said, bite the bullet and go for RAID 10 (RAID0 across RAID1 pairs). Some advantages:

- Less chance of data loss in a 2-disk failure compared to RAID5.

- Rebuild rate for a RAID10 array is based on the size of the mirror pair, not the entire array size. Rebuilding a RAID5 array with more then a handful of disks is *slow*. Rebuilding a RAID10 array typically takes 3 hours... max. Maybe faster depending on the size of your mirror pairs. Which means you have a smaller window where a double-drive failure can kill your array.

- You can go paranoid (although it's a bit wasteful). Three drives in each RAID1 array (all actively mirrored) with RAID0 laid over the top allows you to handle a double-disk failure without worry. You would have to lose all 3 disks in the RAID1 set before losing the entire array. Downside is that your net capacity is going to be below 33% (including hot-spares).

- Performance of RAID10 is nice and predictable. Additional RAID1 elements in the RAID10 array increase read/write rates in a linear fashion until you hit a bottleneck. So if a RAID1 pair gives you 50MB/s read/write, a 10-disk array should hit 250 MB/s read/write (should... not always).

(I like the idea of RAID5 for large amounts of storage. But the rebuild times and risk of a 2nd drive failure during the rebuild window make me hesitant to recommend them anymore. RAID6 is pretty new and for smaller arrays of less then 8 disks, it's simpler to do RAID10. All of this is, of course, my own opinion and YMMV.)

Re:wow

[backwardMechanic]

Another thank you. That's clarified everything.

1.6 million hours?

[huckda]

divide that by 24...and you get 66666.66667....

definitely looks fishy to me

Re:1.6 million hours?

[fatphil]

There's only one important question - what's the manufacturer's warrantee?
That's them putting their money where their mouth is - everything else is just lies, damned lies, and manufacturer-selected statistics.

FatPhil

Re:1.6 million hours?

[drinkypoo]

It's seizegate, so the warranty is five years.

Re:1.6 million hours?

[fatphil]

Thank you, sir. I believe 5 years is the maximum any manufacturer is offering currently.
I've never had a problem with any seagates, nor have I heard of any huge scandal surroundiung them.

Re:1.6 million hours?

[drinkypoo]

I haven't had a problem with Seagates since I was a teenager, back in the days when all my hard disks had ST-506 interfaces - hence the name "Seizegate". I actually had a 40MB RLL disk I opened up with a jeweler's screwdriver (the lid was held on with Torx screws, but back then you couldn't just get a set of torx bits for $7 at kragen's) and unstuck on two separate occasions when I couldn't unstick it by just whacking one corner with the handle of a screwdriver - which always worked for my ST-225 21MB MFMs. Anyway this drive finally burned a trace right off its PCB while in operation one day, so I soldered a jumper across that and it actually worked just fine for a few weeks, then one day it did it again, this time melting the solder holding the wire on. I fixed it one more time but eventually it cooked itself somehow rather than burning off my jumper lead. High-tech shit!

Re:1.6 million hours?

[pe1chl]

Incredible that you are so enthousiastic about Seagate when you have experience from those days.
Most users back then have seen an ST-225 fail. The ST-238R was even worst (same drive but used with RLL instead of MFM).

Re:1.6 million hours?

[drinkypoo]

Incredible that you are so enthousiastic about Seagate when you have experience from those days.

Well, those days were a long time ago. Since then Seagate has had horrible reliability problems, solved them, and then backed up their confidence in them with a five year warranty. Since that time Seagate has apparently become one of the most reliable hard disk vendors.

I mean, to put it in perspective, IBM's policy used to be that any software you wrote on an IBM mainframe became the property of IBM. Today, they're one of the biggest supporters of Open Source...

SAS is a little disappointing

[robosmurf]

I know that this drive is supposed to be a server one, but I'm still disappointed that the SAS standard is not properly compatible with SATA.

SAS is pretty similar to SATA in physical connections, and most SAS cards support having SATA drives plugged into them. Sadly it doesn't work the other way around: you can't plug a SAS drive into a SATA connector.

It's a pity that they didn't sort this out, as drives like this would be nice for workstation users looking for a little speed boost.

Of course, it looks like these kind of low-capacity / high-speed drives are about to be overtaken by the even faster flash based drives coming out.

Re:SAS is a little disappointing

[PopeRatzo]

I'm using a pair of those 10k rpm SATA drives on my audio/video workstation and they're pretty quick. I tried a RAID array of regular SCSI 15k drives and there wasn't enough difference for me to notice. I saw it on the benchmarks, but it wasn't enough to make me want to switch.

Re:SAS is a little disappointing

[ErMaC]

SAS is not designed to be used by a SATA controller. If you wanted your cheapo SATA controller to work with SAS drives, it wouldn't be a cheapo controller. The difference between SAS and SATA is that SAS uses SCSI as its command language, which requires a whole different set of logic on the controller end.
If you're a workstation user looking for a speed boost, then you use SCSI or SAS drives with a proper controller like workstations have since 1990.

And Flash drives have almost no chance of penetration in the server market, which is where this drive is being targeted (not at Laptop or Workstation users). Don't let the 2.5" form factor make you think it's for laptops, it's for high density servers or blades.

Re:SAS is a little disappointing

[evilbessie]

Um SATA is forward compatible to SAS, so you can use a SATA drive on a SAS connection. This implies that the SAS controller can 'speak' SATA which should be fairly easy to implement. Whilst having a SAS drive (SCSI instruction set drive with a SATA type connector) speak both SAS and SATA would put too much electronics on these drives (you are only going to buy SAS drives if you really need them as they are much more expensive, but you may have SAS connectors you wish to use on your motherboard).

If you really want to use SAS drives on a SATA controller you are a complete weirdo. You lose all the benifits of the SCSI instruction set whilst paying for the complete lack of any features not already available on cheaper SATA drives. Maybe a slightly better manufacturing quality, but then if you don't back up your data you will be sorry sooner or later.

So the added expense is making the SAS controller on the motherboard understand SATA command set; (and the SCSI command set) so that you can use cheap drives on a motherboard with SAS drives, useful for non-mission critical machines; is not much compared with the overall cost of the motherboard.

Rather than have SAS drives support SATA as you are only likely to buy these (you'd be really dumb otherwise) if you need the benifits of SAS, ie. the SCSI command set.

Example: 500GB SATA 300 £125; 146GB SAS £435 or 36GB SAS £126.

You are basically saying you should use >5x more expensive disks should be used when you lose all benifits of paying so much.

And anyway most of the modern workstations i've seen recently have contained at least some SAS connectors, if you don't have them and you have the money to waste on SAS drives go buy yourself one.

And anyway a SAS controller card is only about £200 so it's no worse than getting one more SAS drive.

Re:SAS is a little disappointing

[FlyingGuy]

Some enlightenment for you.

If you look inside an SCSI,SAS,SATA or even plain old ATA you will not notice anything different physicaly at all. They all have the same haeds and platters.

ATA drives are far simpler to implement because everything is at the interface layer. There is no command queueing, there is not 16 levels of drive negotiation or anything like that happening. Remember the SCSI spec is for anything, hard drives, CD-ROMS, Tape drives, printers, you name it!

The ATA controler was extremely simple AND inexpensive to implement because it could only have TWO devides MAX and one was always the master, and if it died your slave drive was nowhere to be seen!

In terms of raw performance, there is no longer a big difference between drives, bascialy the data bus is the bottle neck. Where the difference begins to really show is when you put these things in a server and when you do that, SCSI or SAS just rips ATA or SATA a new one. The SCSI command set and interface logic was designed from the ground up to handle massive amounts of simultanious read and write requests. When your server is handling file and print requests for say, 300 people and hosting a database engine the system really needs to offload things to the SCSI controller. So the write commend is sent, along with the data block to the controller and the NOS can simply forget about it and move on to service the next disk request, since SCSI will start sending back error reports if something goes wrong with a particualr command.

That is the big difference. That is why SCSI or SAS is so much more expensive then ATA. They have tried to make ATA raids but not many people use them in servers, because they are more complex the SCSI raids and less robust.

Can you hook a SATA drive to a SAS controller, I have never seen it but i will take your word for it. You can buy a dual ported SATA controller at CompUSA for pretty cheep. The adaptec web site lists the Adaptec RAID 1220SA SATA controller for $75.00, 2 SATA ports, 2 drives, RAID levels 0 and 1.

SCSI or SAS RAID controllers start at around $400. bucks.

Re:SAS is a little disappointing

[goarilla]

The adaptec web site lists the Adaptec RAID 1220SA SATA controller for $75.00, 2 SATA ports, 2 drives, RAID levels 0 and 1.

sorry but i don't think anyone will buy that product
if people buy cheap e-ide controllers or SATA controllers they usually
do that for the intent of running lots of cheap big harddrives
they don't care about redundancy, they care about lots of disks
and room to still expand the total storage capacity
2 drives just isn't enough!

Re:SAS is a little disappointing

[drmerope]

SAS is not designed to be used by a SATA controller. If you wanted your cheapo SATA controller to work with SAS drives, it wouldn't be a cheapo controller. The difference between SAS and SATA is that SAS uses SCSI as its command language, which requires a whole different set of logic on the controller end.

Just so. Except one detail: This isn't the 1980s any more. Buying or designing the IP for a SCSI aware controller is simple&cheap. The fab costs are not likely to be more than for SATA. So, this isn't a cost issue per se.

The meat of it is the same market dynamics that lead to the widespread adoption of x86. SCSI might be better in the abstract but there is a non-SCSI market. IDE survived from its initial status as being cheaper. This happened because drive manufacturers mirrored the market. So by the time we get to SATA, SCSI is in the market as the expensive product and with the big cable. Yes they could have skipped SATA and gone straight for SAS but that suffered a perception of risk.

And it would mean (ultimately) that the high margins on SCSI drives would come down.

Re:SAS is a little disappointing

[dfghjk]

SAS stole the entire physical interface from SATA and was deliberately implemented to allow combination SAS/SATA controllers. Saying that SAS isn't designed to be used by a SATA controller shows a total lack of understanding in the matter.

SCSI doesn't offer any "speed boost" over ATA either and SAS is certainly not faster than SATA. It's the devices that may or may not be faster.

Finally, solid state storage has been used to accelerate server apps for decades.

This is apparently not your area of expertise.

Re:SAS is a little disappointing

[dfghjk]

"Yes they could have skipped SATA and gone straight for SAS but that suffered a perception of risk."

I'm not sure who you think "they" is. SATA was developed first by a consortium of manufacturers that desired to replace parallel ATA. They did not want nor did they have any need for a SCSI protocol. What they needed was a serial interface that worked with existing software.

When SATA came out, it became clear to the SCSI people that SATA was a grave threat to SCSI's artificially inflated margins. Coming out with high-margin server drives with SATA interfaces would cause customers to question if the high prices were justified but SCSI knew that the future was in serial interfaces. In the end, the SCSI group decided to adopt the entire SATA physical interface; performance was fine, production costs lower, time to market quicker, and interoperability possible. In the end, we all benefit because products can transparently support SATA or SAS if the manufacturers desire it. Drive manufacturers still get to pretend that SAS, being SCSI, commands a higher price because it's better.

Re:SAS is a little disappointing

[mnemotronic]

The major end user benefit (as I see it) of SCSI is command queueing and re-ordering. This allows a device to queue up commands from the host and execute them in a order more optimal to it's specific disk layout and current head position, as long as the re-ordered commands don't create data integrity problems. SAS can also do this. Another "benefit" of SCSI (from the OEM's point of view) is all the wonderful MODE SENSE/MODE SELECT crap. I work for (ahem) a disk drive company, and I've written SCSI drive code, and let me tell you, MODE SENSE/SELECT sucks. It is an enormous amount of work to implement (and test, and version control) especially all the customer-specific pages. It adds cost and time to the product, and doesn't really offer any benefit, other than the customer can get all kinds of obscure drive diagnostic info, and change the drive's configuration to make it perform worse. The diagnostic info doesn't do them any good because they don't know diddly about the subtleties of magnetic storage devices, and the drive is already configured for optimal performance. Why would we ship a non-optimal performing product given the availability of SE Asian alternatives?

The major benefit of SCSI from the drive vendor's point of view is the word "$C$I" itself. Anything within a stone's throw of the word "SCSI" suddenly becomes 2-3 times as expensive - it's like an Ebola virus for pricing. Higher prices lead to higher profit margins. This allows for whopping stock options and obscene bonuses for upper management, along with no-interest loans to those same execs to buy the options. And then the company goes and "forgives" the loans or misplaces the paperwork.

'Scuse me - the guy with the whip is coming back. I gotta go write code.

What's so astounding about 15k rpm?

[Pegasus]

I have 15k rpm disks in production since ... 2002 I think. The poster should mention data per actuator figure from TFA, because that is what really matters.

Re:What's so astounding about 15k rpm?

[drinkypoo]

You have 2.5" 15k RPM disks in production since 2002? Who are you? And how were you able to make such bitchin' hard drives in your mother's basement?

Re:What's so astounding about 15k rpm?

[TeknoHog]

I guess what's new is the 2.5'' form factor. Smaller drives should be generally faster due to increased density, but they get a bad reputation from laptop drives with really low RPM.

Re:What's so astounding about 15k rpm?

[Jeff DeMaagd]

I think the main idea is that you can hypothetically install more drives per rack or greater flexibility in the design of devices that need high performance drives. The 3.5" high RPM drives basically use smaller platters anyway, so it's not too much of a stretch to put them in a smaller enclosure, but there may have been concerns about miniaturizing other parts of the drive and still maintaining the enterprise-level reliability.

I think the show-stopper here is that the drive stated capacities are still small, there do exist 150GB 15k drives but these are half that.

Re:What's so astounding about 15k rpm?

[evilviper]

Smaller drives should be generally faster due to increased density,

Only if EVERYTHING ELSE remains equal, which it NEVER does.

In this case, they've got many times lower capacity than even their 10k RPM 2.5" HDD, never mind their 3.5" HDDs.

Plus, you can accomplish high capacity by having more platters (rather than denser info on fewer platters) which will obliviate the benefits of smaller drives. No mention of this in the article, but worth consideration.

Personally, I'll wait for 3.5" HDDs with dual servos instead (basically, internal RAID), which will completely smoke this, and everything else out there.

Re:What's so astounding about 15k rpm?

[Agripa]

In this case, they've got many times lower capacity than even their 10k RPM 2.5" HDD, never mind their 3.5" HDDs.

One of the applications for these drives are systems that are performance limited by access time and not capacity that can not yet use solid state storage. In a lot of very large storage installations, the existing arrays are already capacity underutilized because excess spindles and actuators have to be added to lower the average access time for multiple requests. It is not uncommon to not even utilize the inner area of 3.5 inch drives because the extra capacity is not needed and doing so marginally lowers the access time for systems where this is of primary importance.

Personally, I'll wait for 3.5" HDDs with dual servos instead (basically, internal RAID), which will completely smoke this, and everything else out there.

Dual actuator drives would indeed help significantly and it has been tried however the price premium over using twice as many standard drives would seem to make it too expensive. I suspect solid state storage will become cost effective before multiple actuator drives do.

Re:What's so astounding about 15k rpm?

[evilviper]

In a lot of very large storage installations, the existing arrays are already capacity underutilized because excess spindles and actuators have to be added to lower the average access time for multiple requests.

I think you missed the context, here. I was explaining that the much lower capacity means that this drive probably doesn't have higher data density than a 3.5" drive, so it won't (necessarily) have higher throughput.

Re:What's so astounding about 15k rpm?

[TeknoHog]

So, bigger drives are better? I remember having a 5.25'' HDD, but I don't see them around any more for some reason.

Re:What's so astounding about 15k rpm?

[Agripa]

I think you missed the context, here.

This is likely enough. The subject has been on my mind recently do to a similar discussion over at www.realworldtech.com.

I was explaining that the much lower capacity means that this drive probably doesn't have higher data density than a 3.5" drive, so it won't (necessarily) have higher throughput.

Something which I considered adding in my post was that if the drive's internal data rate is limited by the read channel electronics (there was another post here about this) then it should not be significantly different from lower speed drives. The linear bit density would be lower of course. Before the use of magneto resistive heads this was not necessarily the case since increased linear bit density also translated into a larger signal.

I wonder if there is a significant difference in the zoning between 7200 RPM and 15000 RPM drives giving the later more consistent sustained transfer rates.

Re:What's so astounding about 15k rpm?

[evilviper]

So, bigger drives are better?

No, but smaller isn't necessarily any better, either.

Nice, but not big news.

[Anonymous Coward]

They've had 15K RPM SCSI drives for years and years. This is no big deal.

By only using a 2.5" drive rather than 3.5 of course the average seek time is lower, because the read head doesn't have the extra 1" to cover. This is at the expense of all that extra storage area.

You could get just about as high an average seek if you partitioned up a 3.5" 15K drive and only kept data on the inner partition.

It's nice that they have these, but it's really not that super special. Why is this front page news?

BTW, your laptop is going to need some serious cooling to use this, as 15K drives do get rather warm.

Re:Nice, but not big news.

[fred fleenblat]

>> By only using a 2.5" drive rather than 3.5 of course the average
>> seek time is lower, because the read head doesn't have the
>> extra 1" to cover.

it's even more trivial than you paint. The 2.5 and 3.5 numbers
represent diameter, but the head only travels on one side of
the disk so to it the difference is only 0.5 inch as far as it
is concerned.

Re:Nice, but not big news.

[Fweeky]

It's news because it's supposedly the fastest platter based disk yet, and because it's the first major development in 2.5" disks in several years; in that time they've grown rather popular in servers, as seen in Sun's range of Opterons for example.

This now makes the form factor even more competitive in IO-sensitive applications, and I dare say Slashdot has enough users interested in such a thing to warrant a FPP.

Re:Nice, but not big news.

[dextromulous]

It's nice that they have these, but it's really not that super special. Why is this front page news?

Maybe it's not super special... but check this paper out. Figure 1 in particular (yeah... they are comparing 10k 3.5" to 10k 2.5"... it's an old paper, but the theory is the same.)

Who would want faster 2.5" drives? People who want 6 drives in 1U instead of in 2U. People who want faster drives in their blade servers.

Re:Nice, but not big news.

[TopSpin]

This is at the expense of all that extra storage area.

The people for whom these high end disks are intended aren't concerned with the "storage area" of individual devices. They care about the ratio of storage to spindles and arms. They buy things like this.

Why is this front page news?

Because it's a site about stuff geeks want to read. It's actually rather nice to hit the page and find some news about the latest incremental change in storage, as opposed to more move-slash, dot-on politics.

Re:Nice, but not big news.

[noidentity]

"You could get just about as high an average seek if you partitioned up a 3.5" 15K drive and only kept data on the inner partition."

Wouldn't it be better to put the partition at the outer edge of the disk, where you get higher data rates and more data per cylinder (and thus less head movement to get from beginning to end of the partition)?

Re:Nice, but not big news.

[johu]

Actually 3,5" 15krpm drives have very small platters inside, not much bigger than those used in regular 2,5" drives. I don't know if 2,5" 15000 rpm drives have even smaller ones.

Here's first hit from google image search that shows how small they are. http://www.starthomepage.com/shp_upload/board/win_ user_tip/15,000RPM%20HDD%20Cheetah%2015k.3.jpg

Re:Nice, but not big news.

[Anonymous Coward]

You could get just about as high an average seek if you partitioned up a 3.5" 15K drive and only kept data on the inner partition. Perhaps, but the 2.5" drives use less power and you can squeeze more of them into the same space. This gives you more performance for a fixed amount of space or power consumption.

Re:Nice, but not big news.

[Nyph2]

By only using a 2.5" drive rather than 3.5 of course the average seek time is lower, because the read head doesn't have the extra 1" to cover. This is at the expense of all that extra storage area.

You could get just about as high an average seek if you partitioned up a 3.5" 15K drive and only kept data on the inner partition.

Actually, you'd want to keep it only on the outside partition. The edges of the platter have the most data and still make rotations at the same speed, hence more data per rotation. This is why you should make your swap partition the last last on the dive.

I must have ordered from the future!!!!

[CaymanIslandCarpedie]

... becuase last week I ordered a server from HP with 2.5" 15k drives HP.

Re:I must have ordered from the future!!!!

[CaymanIslandCarpedie]

Opps, ignore me. I didn't realize they were talking about old SCSI. Mine are SAS (serial attached SCSI).

How many seek/ECC errors does it give??

[madhatter256]

I don't know about you, but every single Seagate HDD I've tested, both brand new and used give a lot of seek errors way above the SMART margin if you run SpinRite 6.0. I've experienced Seagate HDDs simply failing because of too many logged seek/ECC errors and Windows will freeze as it initially loads. I have never seen this type of perfomance with Samsung, WD, Fujitsu (SCSI) and Hitachi HDDs. Sure, not all hard drives are perfect but in my experience, Seagates have always given me problems to the point where I simply don't recommend them anymore.

Re:How many seek/ECC errors does it give??

[mungtor]

Everybody has stories like this. I have no problems with my Seagate drives, but I wouldn't put anything on a WD drive. Sure, it will be fast for 3 months until you lose it all. With most manufacturers it comes down to a particular model being a bit flaky, although all WD drives suck.

Re:How many seek/ECC errors does it give??

[D4rk Fx]

I've only had 2 of more than 12 WD drives die; one was because it fell while running from more than 8 feet off the ground, the other was insufficient cooling. I've had 5 of 6 Maxtors die, and I'm 4/4 with IBM drives deaths. 0/4 for Seagate, but they are my most recent acquisition.

You're right, everyone has stories. I have 2 4 drive WD arrays that have been around for 3 and 2 years, no failures there. But I wouldn't trust any data to an IBM or a Maxtor drive.

Re:How many seek/ECC errors does it give??

[Slaimus]

I have seen very high seek error rates reported by SMART as well for Seagate drives. However, I have not seem them affect anything in terms of reliability.

Maybe Seagate just uses uses a different metric for measuring seek error rate that results in higher raw numbers.

Re:How many seek/ECC errors does it give??

[WuphonsReach]

And the flip side, I've owned close to 2 dozen IBM Deskstar drives (mostly 72-80GB). No more then a handful died before their warranty period expired.

Most of those deaths were directly related to heat issues (poor cooling or poor airflow). Some were undetermined cause.

From my experience over the past decade, heat is the #1 killer. Some makes / models are better at dealing with 50C+ temperatures then others. Maxtors seemed to be a bit sensitive to anything above 50C (and Maxtor drives were a real PITA to RMA, IBM RMAs were a simple click-click-click on a web form prior to send it back).

Nowadays, I simply plan for failure (RAID1 across 3 drives or RAID10 w/ hot-spare) along with backups. I try to keep drives at or below 40C and I keep enough airflow across them that their operating them doesn't change by more then 5C between idle/active.

Re:How many seek/ECC errors does it give??

[UnknowingFool]

I've had a few Seagate die on me for various reasons like cooling, etc, but I would have trusted them in the past. I've had no issues with WD. I still have one that has been running since 2000 without an issue. Maxtors have always crapped out on me. I've had to return the same model twice now.

With Seagate buying Maxtor, the line will be blurred. I would hope that Seagate would bring Maxtor's quality up but I'm afraid it is the other way around.

Re:How many seek/ECC errors does it give??

[fatphil]

Name a time when Seagate was ever involved in a scandel like the Hitachi deathstar/troublestar one.

I will never buy a Hitachi or Hitachi rebrand. My data is just too valuable to risk it to a company that can produce things with a ~30% failure rate. (Though from personal experience, it feels more like 50-60%, as I've had 3 fail on me.)

Re:How many seek/ECC errors does it give??

[D4rk Fx]

I completely agree that heat is the #1 killer. Yes, drives will run hot, but they will last a lot longer if they run cool. Last I checked, none of the drives I run now were hotter than 30C. I haven't had any significant drive deaths in a few years. I had one that seemed like it had firmware issues, as it would just stop responding on occasion, but would be fine when the power was re-applied.
On a side note, the hard disk in my laptop thinks that the Min/Max temps it's seen while operating is 52C/65528C. Now why the manufacturer would have used an unsigned 16 bit integer to track temperature escapes me...
KEEP YOUR DRIVES COOL, PEOPLE!

Re:How many seek/ECC errors does it give??

[Yewbert]

My personal and professional experience tend to align more with this. I've personally had at least three Maxtors die *very* prematurely (the first time, a SATA, losing me a fair amount of data in the process) out of maybe four or five that I've ever bought. One WD death out of maybe half a dozen, and so far, 0 Seagate deaths out of what must be approaching 20. I tentatively think Maxtors may be more sensitive to overheating than other brands, 'cos the circumstances in most of these drive deaths included sub-optimal ventilation.

At work (one of my several hats is 'workstation support' in a department with around 300 specialized workstations [this dept only - company-wide, there are probably tens of thousands of desktops/workstations/laptops with hardware equivalent to store-bought builds, not to mention hundreds of monstrous servers]), it's become a running joke to the point that when a workstation in the field has a hard drive failure, we practically write the trouble ticket up as "Maxtor failure."

Re:How many seek/ECC errors does it give??

[Emetophobe]

And here is some anecdotal evidence to counter your claim that Western Digital drives suck.

I have 4 250gig WD SATA drives (all model WD2500KS). I've had 2 of them for a year and not a single issue. Recently I bought two more and I've had them set up in a RAID0 array for the past 3 months without any problems. I use Acronis True Image just in case, but I haven't had to restore any images yet... IMO, these western digital drives are great, they are fast and quiet, and they cost less than $90 a piece.

One major problem I've seen is people with crappy power supplies causing hard drive corruption. If your power supply isn't able to keep your hard drive(s) powered correctly, you can have corruption, blue screens of death, etc.. It's pretty important to have a good PSU that delivers stable power across all rails. A bad power supply won't be able to keep your drives powered at all times, causing data loss, etc..

Re:How many seek/ECC errors does it give??

[springbox]

All of the Seagate drives I have have been excellent. No unusual readings from SMART or anything. Cooling helps. Although, I've had a lot of bad luck with Maxtor drives dying in the past.

Re:How many seek/ECC errors does it give??

[evilviper]

I've had 2 of them for a year and not a single issue.

Well you're just reaching the point where you can expect problems.

My last WD HDD (100GBs) turned to crap exactly ONE DAY after the end of the 1 year warranty. To WD's credit, they agreed to allow a replacement under warrant, with only minimal complaints and little hassle.

Re:How many seek/ECC errors does it give??

[Anonymous Coward]

Wow, you're correct.  Here's the output from smartctl from one of 90 new Seagate 750 GByte drives we have (w/ the middle unimportant columns deleted for clarity):

ID# ATTRIBUTE_NAME         RAW_VALUE
  1 Raw_Read_Error_Rate    214270477
  4 Start_Stop_Count       21
  7 Seek_Error_Rate        10434919
  9 Power_On_Hours         163
194 Temperature_Celsius    42
195 Hardware_ECC_Recovered 2701

Notice it has only run 163 hours, but has 214,270,477 read errors!  That's around 365 errors per second.  Something definitely isn't right about the way Seagate counts errors.

It also claims the drive is 148 degrees F.  It's in a case with very good airflow in a 65 degree F computer room.  That number is also bogus.

The other three drives I looked at had similar numbers.  Out of the 90, two have quit so far.  I'm going to have fun trying to keep that 67TByte storage cluster running.

Re:How many seek/ECC errors does it give??

[adolf]

It's interesting that you seem to think that your experience is in some way predictive of what others will encounter. But that's just not so.

Pick a manufacturer of ANYTHING more complicated than a soupcan, and you'll always be able to find someone writing sky-is-falling stories about how the thing died within hours of the warranty expiring, right after it kicked the dog and set the house on fire.

It happens with any product. The thing of it is, though, that except for a few particularly egregious and well-publicized examples (eg, IBM Deskstar 75GXP, Sony batteries, Firestone tires), stuff generally lasts a good long while after the warranty is gone.

So for good measure, I'll now proceed with writing a bit about a few things I've noticed:

I recently discarded a Plextor burner which had seen hard duty for 9 solid years, because it (finally) started to get a bit flaky and no longer produced perfect audio CDs. It lived 4 years past its 5-year warranty.

I'm still running old 4 and 6 gig drives in a couple of machines here that don't need much for storage. 5+ years past warranty and counting. (I'm contemplating replacing these with flash memory, but more for general principal and noise abatement than any particular fear of instability.)

The Uniden cordless phone I purchased 10 years ago works like new, though the battery has been replaced a few times.

The toaster in my kitchen. Its warranty, if there were any, expired decades ago. I dump the crumbs out of it from time to time, and it keeps toasting bread just fine.

I'd go on, but it'd be beleaguering the point. Look around you, and you'll find a bunch of stuff which is currently well past its warranty period.

Nevertheless, if you start Googling a few of the more recent ones, I'm sure you'll find sad, sad stories from people whose otherwise-identical stuff died, like clockwork, precisely 20 minutes after the 90-day warranty expired. But that, my friend, is more an indication that the world is a very big place than of any grand trend toward generally-faulty products.

Relax.

Re:How many seek/ECC errors does it give??

[evilviper]

It's interesting that you seem to think that your experience is in some way predictive of what others will encounter. But that's just not so.

Oh, but it is. It's a fact that WD has higher power consumption than other hard drive manufacturers.

I recently discarded a Plextor burner which had seen hard duty for 9 solid years,

My Lite-on CD burner is going strong as well.

I'm still running old 4 and 6 gig drives in a couple of machines here that don't need much for storage.

Same here. I've got a couple 1 and 2 GB hard drives going still. A 20MB MFM drive with controller around here somewhere, too, which still works fine.

My 6.4GB hard drive was destroyed by a PSU short years ago, unfortantely. Through no fault of Maxtor, of course (at least, AFAIK).

My 2x 27GB hard drives are still running fine as well.

With modern hard drives, however, the situation hasn't been as plesant. I have 4 Western Digital HDDs... 2 are older 40GB HDDs, and 2 are 100GB HDDs.

One of the 40GB HDDs hasn't had any problems, though it's been a backup drive most of it's life, and never been heavily used.

The other 40GB drive was in my main desktop machine, and under heavy use, though not always-on. I believe it lasted 9 months. Incidentally, the refurb drive WD sent was unfortunately extremely noisy, while the original was nice and quiet. That drive made it for 2 months, and failed. The (second) replacement drive has been working fine ever since (about a year now).

The first 100GB HDD was in a workstation, that was always-on, for about 10 months. When it had to be replaced, I bought a new 100GB drive, and used the refurb only for backups... Despite its light-duty, it has twice developed read errors, once in the MBR, losing all the (backup) data, and requiring clearing the drive and repartitioning.

The (new) 100GB HDD is the one in question which failed exactly 1 year and 1 day after I bought it. The refurb has worked for a few months now with no problems yet.

I'd say that's a bit stronger than just one-off anecdotal evidence, wouldn't you?

Omission from TFA

[Timesprout]

The laptop holding the drive was itself spinning at 5000 RPM to achieve this figure, which makes it slightly difficult to use.

Re:Omission from TFA

[Hao Wu]

Spinning your computer equipments that fast would cause serious damage to components. It would not work anymore, and using it would be virtually impossible.

I think it is implausible that it was really spinning as fast as you say.

Re:Omission from TFA

[ovideon]

Spinning your computer equipments that fast would cause serious damage to components. It would not work anymore, and using it would be virtually impossible. Not if you're spinning at the same speed! Try holding onto the cambelt of your car while sitting in an office chair.

Re:Omission from TFA

[Hoi Polloi]

They spun the laptop at 15000 rpms opposite the drive so the platters never actually moved.

the edge of the plate spins 50 meters a second!

[dgerman]

This is insane. The edge of the plate travels 3km a minute:

2.5 inches diameter => ~20cm perimeter at 15k RPMs => 3km/Minute => 50m/s => 180 km/hr.

Re:the edge of the plate spins 50 meters a second!

[DarkSarin]

Somehow, this is the most interesting and unique response to the article I've seen.

That is really fast. So, now use your mad math skills, how fast would it travel if it was 3.5 inches in diameter?

Re:the edge of the plate spins 50 meters a second!

[Radon360]

What about the hard drives in the laptops of passengers on commercial airplanes? I would think that some of those travel in upwards of 800km/hr during flight.

Re:the edge of the plate spins 50 meters a second!

[Qzukk]

For the metrically challenged among us, 180km/hr is 12025769.5 rods per fortnight, or really, really fast.

Re:the edge of the plate spins 50 meters a second!

[Spezzer]

Math may be slightly wrong on my part but...

2.5in is about 6.35 cm, so it's actually closer to 32cm perimeter, which roughly is about 4.8 km/m or 80m/s.

But also keep in mind that the outer edge of a 3.5in plate spinning at 7200rpm is about 4.4km/m or 73m/s, so it's not that surprising.

In fact, the 3.5in 10000RPM raptors edge spins at about 110m/s.

Re:the edge of the plate spins 50 meters a second!

[KonoWatakushi]

That's nothing; in terms of rotating things, flywheel batteries are much more interesting. They have achieved a velocity of 2km/s at the edge. (about Mach 6)

Take a look at http://www.llnl.gov/str/pdfs/04_96.2.pdf

Re:the edge of the plate spins 50 meters a second!

[CODiNE]

It also doubles as a cockpit anti-terrorism unit. When the pilot is attacked from behind he ducks and pops off the side of these babies... WHAMMO!! Flying blades of death!! Actually who cares that they can store data ... wiping out terrorists is the important thing!

Re:the edge of the plate spins 50 meters a second!

[dcw3]

This is insane. The edge of the plate travels 3km a minute:

Well, it certainly is fast for such a small platter, but I recall servicing the old HP7906 removable platter disks back in the early '80s. During one of our moments of boredom, we did the math, and came up with a figure somewhere around 170mph on the outer edge of the disk. Granted, those platters were huge in comparison. I tried to find some specs, but had no luck in my five min. search.

Re:the edge of the plate spins 50 meters a second!

[jo42]

Don't the magnetic bits get really really dizzy spinning that fast? What keeps them from flying off or chucking their cookies?

Re:the edge of the plate spins 50 meters a second!

[J.R. Random]

At the rim of the platter there's nearly 8,000 g's of centripedal acceleration. I wonder if the engineers have to take into account the gradual stretching out of the disk.

The math: 1.25" radius = .03175 m.
15,000 rpm = 250 cps = 1,570.796 radians / sec.
acceleration = r * omega^2 = .03175 * (1570.796)^2 = 78,339.98 m / s^2.
1 g = 9.8 m/s^2, so acceleration = 7994 g's.

Re:the edge of the plate spins 50 meters a second!

[A_Non_Moose]

This is insane. The edge of the plate travels 3km a minute:

Is that with or without a spoiler and type R stickers?

Re:the edge of the plate spins 50 meters a second!

[hurfy]

hmm, good question.

I posted about the speed of my vintage minicomputer. I have platters that i can see the g-forces trying to pull the alloy platter apart. The magnetic section is darker at the outside. It did pull even more than these little ones and it ran for 15 years. However it used a platter for tracking that would have migrated as fast as the data tracks so maybe it compensated for it. Adding a NEW removable platter may prove unstable now, if there were such a thing ;)

Don't know how these little ones do tracking tho.

Re:the edge of the plate spins 50 meters a second!

[frank_adrian314159]

I wonder if the engineers have to take into account the gradual stretching out of the disk.

Quick answer, yes, but only to the extent that the structural form of the disk has to resist it up to a given tolerance. The rest of the variation is handled in the same way that other variations are handled (e.g., differences in electronic/magnetic component response times, motor speed variations, plate wobble, etc.) which are usually larger. And how do you handle it? Slow it down a bit. If they can mass produce 15K RPM 2.5" drives, they've either gotten up to ~20K RPM in development, or have a buttload of 12.5K RPM 2.5" drives to shove onto the market.

Re:the edge of the plate spins 50 meters a second!

[KingKiki217]

(pi)d=circumference (pi)3.5=11 inches 11 in x 2.5cm/in = 27.5 cm 27.5 cm x 15k rpm = 412500 cm/min 4.125 km/min x 60 min/hour = 247.5 kph (times .62 to get mph if you're American like me gives you 153.5 mph)

Re:the edge of the plate spins 50 meters a second!

[trjonescp]

This rods per fortnight conversion is amazingly understood by Google. Bravo, Google.

Re:the edge of the plate spins 50 meters a second!

[KingKiki217]

Do I have to use html to keep a break in the comment? *sigh* It was so nicely formatted before I hit submit.

Re:the edge of the plate spins 50 meters a second!

[Wolvie MkM]

Good call... don't forget it adds 50 more horse if it's yellow

Re:the edge of the plate spins 50 meters a second!

[rew]

That's why they don't have any 3.5" 15kRPM drives. Sure, 15k drives exist, and they come with the "metal" so that they fit into your standard 3.5" drive bays, but the disks are barely larger than what would fit into a 2.5" drive.

Re:the edge of the plate spins 50 meters a second!

[Hoi Polloi]

As a person with an astronomy background I prefer to use parsecs/hr (0.000000000005833401930 pec/hr)

Re:the edge of the plate spins 50 meters a second!

[Hoi Polloi]

30 hp more if it has a sewer pipe exhaust. Nothing excites the ladies more than a car that sounds like the muffler fell off of it.

Re:the edge of the plate spins 50 meters a second!

[canavan]

Revolutions per second, i.e. Hz is the proper unit to fill in this formula ofr omega to get accelration in m/s^2. With a= r* omega^2 that just gives us more reasonable 0.03175m * (250/s)^2 = 0.03175m*62500 / s^2 = 1984.375 m / s^2

That's just 202.5g. Nothing to sneer at,

Re:the edge of the plate spins 50 meters a second!

[PayPaI]

Here's an image of the inside of a 15K 3.5" drive.
I'd guess the platter is about 2 and 1/3 inches across.

Re:the edge of the plate spins 50 meters a second!

[Raenex]

Preview is your friend.

Re:the edge of the plate spins 50 meters a second!

[J.R. Random]

Wrong. Radians per second, not cycles per second, is the appropriate measure.

To put it another way, the acceleration is v^2 / r, where v is the velocity at the circumference. But v = circumferance * cycles per second = 2 pi r * cycles per second = r * radians per second. So acceleration = r * omega^2 where omega is radians per second.

But, about that noise?

[LibertineR]

2.5 and 15K?

That sucker must screech like your ex-wife one day after your alimony payment was due.

Re:But, about that noise?

[Intocabile]

I would imagine a drive with such a high mean time before failure is quieter by design.

Re:But, about that noise?

[Anonymous Coward]

Right now I'm sitting in a room with 40 Seagate 3.5" 15k RPM drives. The noise isn't that bad. I can still hear the heads move over the rotation noise. Of course in a few months when the bearings start going, then they'll start screetching like hell. I'm dreading that because when the bearings in the 7,200 RPM Seagates we replace with the 15k ones started failing I couldn't hear to talk on the phone. I'm sure it's going to be worse with the 15k ones.

Re:But, about that noise?

[keilun]

If you goto the link in the article, which leads you to their review of the Savvio 15k, you'll notice a noise comparison on page 12:

http://techreport.com/reviews/2004q4/seagate-savvi o/index.x?pg=12

Re:But, about that noise?

[J.R. Random]

Actually, that review is dated October 15, 2004 and is for the 10k Savvio. So I don't think it tells us anything about the noise level of the 15k Savvio.

Re:Moving disks are old SSD is in

[KonoWatakushi]

You do realize that the SSD you reference is based on flash, right? If you look carefully, you will find that no vendors list write seek times or write IOPS for such devices. The reason is that the performance is just plain awful.

RAM based SSD is nice, but flash based SSD won't touch a decent 15k drive for any write heavy application.

Why the low capacity?

[TheRaven64]

The drives are 36GB or 73GB. This seems to be a standard size for SCSI, but SATA 2.5" drives have capacities in excess of twice that. Can anyone explain to me why SCSI drives always seem to be lagging IDE in terms of capacity? Does the increased rotational speed make them unable to discern smaller features on the disk?

Re:Why the low capacity?

[D4rk Fx]

The increased rotational speeds dictate that they must use smaller diameter platters, or risk the platters exploding because of the increased centripetal forces exerted.

Re:Why the low capacity?

[WuphonsReach]

Can anyone explain to me why SCSI drives always seem to be lagging IDE in terms of capacity? Does the increased rotational speed make them unable to discern smaller features on the disk?

SCSI drives, while using a 3.5" form factor, use smaller platters inside so that they can spin at the higher rotational speeds. Thus, lower capacity. AFAIK, SCSI drives use the same bit density per square unit of linear measure as SATA/PATA drives.

Re:Why the low capacity?

[tppublic]

Because SCSI and SAS are not about density. See the Seagate Research paper in my other post.

Re:Why the low capacity?

[jimicus]

The people buying SCSI drives are going to be attaching them as part of a honking great array. The biggest concern is speed, and you get that by spreading the data across as many drives as you can.

73GB doesn't sound so bad when you multiply it by 12 for the number of disks you plan to use.

Re:Why the low capacity?

[rrohbeck]

Can anyone explain to me why SCSI drives always seem to be lagging IDE in terms of capacity? The main limitation for bit density on a high speed drive is the channel data rate (since you can't use anything but standard CMOS in a low power, high volume, low margin product.) If you spin faster, at a given maximum bit rate, you lose bit density. Also, for faster seeks, you have to put down more servo information (otherwise you may not see any servo bursts for some time while the head is crossing only data.)
You can generally stuff more data on a platter by spinning it slower. That's why basic 2.5" drives usually spin at 5400 or even 4500 rpm.
Of course the interface has nothing to do with it. SCSI=>high end=>faster=>lower capacity. This may actually change with the convergence between SATA and SAS.

Re:Why the low capacity?

[Emetophobe]

Higher capacity = lower speed. Most SCSI drives sacrifice capacity for higher speed.

They probably have smaller platters, and also, less total platters in the drive. This means the drive has a smaller capacity, but the read head has the advantage of not having to move as far across the disk to read data, reducing latency and increasing performance (AFAIK).

Re:Finally, an upgrade for my beloved powerbook

[vought]

Now if only I can get around the 14MB RAM limit, I'll be set up.

Well, there's also the 40-minute battery life to be contend with...but nothing beats that 8.4" Toshiba active matrix display!

15k rpm -- old, OLD news

[smellsofbikes]

Putting it in a 2.5" package is pretty cool, but there have been 15k rpm 3.5" drives since the early '90's, as far as I recall. My desktop Dell has one. Here's a review of three popular ones. And, for the record, the edge velocity on a 3.5" is considerably higher than a 2.5" for the same rpm.
Correct me if I'm wrong here: 3.5" x 3.14 = 11 cm circumference, *15,000 = 1.6E5 cm/min, /100 = 1.6E3 m/min, *60 = 98910 km/hour.

Re:15k rpm -- old, OLD news

[tppublic]

Well, you're half right. The chassis is 2.5" or 3.5"... the platters are not. SCSI drives generally use smaller platters than SATA drives. Seagate's Cheetah X15 15,000 RPM drive is a 2.5" platter in a 3.5" chassis.

Re:15k rpm -- old, OLD news

[lagfest]

I hope you don't work for NASA.

In addition the units, note that 3.5" is the size of the drive, and not the disk itself. and for 15k drives, the difference is substantial.

Re:15k rpm -- old, OLD news

[lagfest]

Oh, and the real math:

Lets say the diameter is 3.5", that's 8.89 cm

circ = 8.89 x pi = 27.93 cm
speed = circ * 15k/min /60s = 6982 cm/s = 69.8 m/s = 251.28 km/h

Re:15k rpm -- old, OLD news

[Anonymous Coward]

you are wrong... your figure is hundreds of inches/hour, not km/hour. 3.5" * pi * 2.54 = ~28cm circumference. *15000 = 4.2e5 cm/min. /100 = 4.2e3 m/min. /1000 = 4.2 km/min. *60 = 251 km/hour. the edge velocity for a 3.5" as compared to a 2.5" drive is simply the ratio of their diameters.

Re:15k rpm -- old, OLD news

[smellsofbikes]

I suck at math. That's why I got a microbiology degree. Sheesh: good thing I'm not designing brake systems on cars.

Re:15k rpm -- old, OLD news

[rew]

So, the figures come to: 50m/s for 2.5" and 70m/s for a 3.5" 15kRPM drive. For the units impaired, that's 180 km/h and 250 km/h. Or 112 Mph and 157 mph.

However the numbers for 3.5" drives are unrealistic, because nobody has ever made a 15kRPM drive with 3.5" platters. Already the 10k RPM drives have smaller platters than 3.5".

Re:Moving disks are old SSD is in

[jdgeorge]

You do realize that the SSD you reference is based on flash, right? If you look carefully, you will find that no vendors list write seek times or write IOPS for such devices. The reason is that the performance is just plain awful.

RAM based SSD is nice, but flash based SSD won't touch a decent 15k drive for any write heavy application.

The reason "seek time" isn't listed for SSD devices is the same reason dynamic RAM manufacturers don't list "seek time" in their device specifications, namely, it doesn't apply. In storage device parlance "seek time" refers to the time it takes for the drive head to reach the target data on a rotating disk. Read the (ahem) authoritative Wikipedia article here.

Furthermore, the recently announce flash-based SSD's from Samsung and SanDisk have file access times far superior to any rotating disk-based storage device. However, it is true that the dynamic RAM-based devices have access times that are approximately 10 times faster than the flash-based devices, but the flash based devices have file acces times typically much more than 10 times faster than a disk drive's seek time. For reference, see the SanDisk press release for their SSD device.

SAS is about more than speed

[tppublic]

You don't see a reason to switch, because the benefits of SAS are in reliability, not in speed. The mechanism inside an enterprise drive is different than that in a consumer drive, and you can see that in the reliability specs and the warranty periods. Given that most consumer data really isn't mission critical (as much as people claim it is), RAID 1 SATA drives are sufficient.

Seagate Research presented a good technical article on SCSI vs. SATA back in 2003. Much of this is still relevant today (though it's SAS vs. SATA)

Speed

[certel]

Zoom zoom zoom! I've never been a fan of SCSI, but then again, I've never used it on any personal PC. Some opinion.

Re:15k rpm -- 2000 actually

[tppublic]

Forgot to say the first 15K drive was 2000... an article with the dates and speeds.

Re:Moving disks are old SSD is in

[KonoWatakushi]

Yes, seek time is no longer the proper term for it, but people use it. IOPS is the relevant measure, and as I have said, they do not list write IOPS. Read IOPS for the recent SanDisk SSD announced was 7000; this is much better than spinning media, but still pretty bad for SSD. Write IOPS are much worse, and they won't even list that.

In any case, the difference between flash and RAM SSD is far more than 10x.

kinda slow actually.......

[hurfy]

This is what i was trying to calculate myself.

Except this is not insane, it is actually slower than my vintage wang minicomputer with the 80MB drive that weighs more than me.

Best guess was 111 mph for this vs almost 125 mph for the Wang from 1980. (about 14" platter at 3000 rpm)

Re:What about for professional use?

[Anonymous Coward]

running a SAS drive in a laptop would murder the battery... unlike the SATA spec, the SAS spec does not provide for a method to turn the interface off when it is idle.

Re:What about for professional use?

[morgan_greywolf]

Even if you had a laptop that had a SAS adapter like the i965, you still have the problems of power consumption and heat, and yes, these do apply to 'professional' or 'desktop replacement' laptops. Even if you completely disregard battery life (and who's gonna do that even in a desktop replacement laptop?), laptops generally don't have the larger capacity power supplies found in desktop and server machines. One of the easiest and simplest ways in terms of bang for the buck to reduce power consumption in a laptop is to put in a smaller, slower HDD. Heat is also a problem, since the tighter cases of laptops (which have less room and ventilation even as compared to blade servers) cause significant buildup of heat. I'm sorry, I just don't think we'll see 15K RPM HDDs like the Savvio anytime soon.

Cheeta X15 is a 2.5" platter in a 3.5" chassis

[cdn-programmer]

WOW.

So a 2.5" drive should be a 1.5" platter in a 2.5" drive?

This would mean if we can get the drive down to 1" then we can eliminate the platters eh?

That would be good. Such a drive would have zero latency. It could have zero seek times too. It could be a virtual drive.

It would probably be more than a billion hours between failures!

WOW.

170 sounds kinda high

[hurfy]

I came up with 125-137 mph for my 14" Wang ;)

I don't remember if it was 3000 or 3300 rpm and manual is at home.
Those made one helluva metal lathe if they crashed the heads tho :O

At least i wouldn't be scared to turn on this new little drive.....

Re:15k rpm -- 2000 actually

[smellsofbikes]

Seriously? That's crazy. Huh. I could've sworn the HP server systems I was working on in the late '90's advertised 15krpm. I still have a couple of them from 1999, but maybe they were just 10k. They're unbelievably loud: they sound like someone winding up a Ducati, especially under high disc activity.

Re:Moving disks are old SSD is in

[jdgeorge]

Yes, seek time is no longer the proper term for it, but people use it. IOPS is the relevant measure, and as I have said, they do not list write IOPS. Read IOPS for the recent SanDisk SSD announced was 7000; this is much better than spinning media, but still pretty bad for SSD. Write IOPS are much worse, and they won't even list that.

In any case, the difference between flash and RAM SSD is far more than 10x.

You are correct; I wasn't thinking about write speed, which is typically much slower than read speed for flash devices.

Regarding the difference between flash and dynamic RAM based devices, I haven't found a reference that shows a significantly greater than 10x difference between DRAM SSDs and Flash. The source I looked at for an (admittedtly random) DRAM-based SSD performance statement is this press release, which shows a DRAM-based SSD with Read IOPS of 50000 (less than 10 times that of the SanDisk flash device, which claims random Read IOPS of 7000). I have no sense of the difference in Write IOPS, though.

Heat and power?

[antdude]

Doesn't this faster speed make the HDD make more heat and use more power?

Re:Heat and power?

[Emetophobe]

Yes, SCSI drives run hotter and use more power than a consumer drive. A typical consumer drive runs at 7200rpm, while a 15k SCSI drive is over double that, at 15000rpm (they are quite a bit louder too).

According to this review, the Savvio 2.5" 15K drive uses less power than older 10K drives. Also, it is twice as quiet as previous 10K models, that's pretty impressive IMO.

These drives aren't cheap either, the 36GB version goes for around $450 while the 73GB version goes for around $840.

Re:Heat and power?

[antdude]

Wow, that's expensive. Thanks for thbe reply. I always didn't like the HDD speed in laptops/notebooks.

Flash Drives

[ProfessionalCookie]

And Flash drives have almost no chance of penetration in the server market

What do you mean? I fully expect that rotating drives are on their way out. There's too many advantages to flash and the disadvantages with using SSDs in a server environment are being worked out as_we_speak. I'm willing to wager that within 3 years SSDs will beat high end HDDs in every desirable metric sans price- and price is just a matter of time.

Re:Flash Drives

[drsmithy]

What do you mean? I fully expect that rotating drives are on their way out. There's too many advantages to flash and the disadvantages with using SSDs in a server environment are being worked out as_we_speak. I'm willing to wager that within 3 years SSDs will beat high end HDDs in every desirable metric sans price- and price is just a matter of time.

I doubt SSDs are going to come within a bull's roar of magnetic media in terms of cost-effectiveness any time soon (if they ever do).

What I *can* see, is the growing use of flash [drives] as an intermediate caching device - in SANs/NASes (eg: each physical array comes with an SSD for caching purposes), individual drives (the drives with flash RAM that have been talked about recently), some magic device that plugs in between the regular drives and the disk controller and the poor-man's DIY version at the OS level (eg: Vista's "ReadyBoost").

I can also see them being used in small scale, very specific tasks (eg: DB transaction logs).

But, flash completely - or even meaningfully - replacing magnetic media in the forseeable future ? No way. It just can't provide sufficient density at a reasonable cost. Price out a 500G (usable) array of flash disk. Even being generous and using a parity-based RAID scheme where you only need n+1 or n+2 disks is still going to have a cost vastly in excess of an array of regular disks (and potentially requiring more physical space as well).

Re:Flash Drives

[ProfessionalCookie]

You're thinking in terms of today. SanDisk has already released 16GB CF cards. Density is increase at a fairly quick rate. While I agree that it'll be more expensive for quite some time I find it hard to imagine a future world that still relies on spinning disks and moving parts. What I am saying is that the technology for flash memory & interfaces is quickly approaching the point at which it will outperform any disk solution, even if the only fightable factor left is rotational latency.

All that to say perhaps my foreseeable future is a little farther away than yours.

solid idea?

[deviceb]

Yeah ok... try to fix a common bottleneck. I'm with ya on that.. But i would prefer more effort into solid state disks not something that spins fast enough to eject from my box & into orbit.

No moving parts please~~

(that being said, when can i buy one?)

You call that a head crash? THIS is a head crash.

[mkcmkc]

I was wondering what sort of angular momentum we're building torwards. Here are some interesting photos of a 55,000 RPM "disk" crashing.

(okay, so the platters are a little on the heavy side)

Faster Porn?

[chromozone]

http://money.cnn.com/2006/11/30/magazines/fortune/ obrienseagate.fortune/index.htm "Not so with Bill Watkins, the mercurial, salty-mouthed Texan who runs the $15 billion hard-drive king Seagate Technology. At a San Francisco dinner on Tuesday evening, he was candid about his company's ultimate mission: "Let's face it, we're not changing the world. We're building a product that helps people buy more crap - and watch porn."

Re:Faster Porn?

[Hoi Polloi]

God bless that man. Finally, a big businessman who sees the truth and speaks it too.

Blade Servers

[muckdog]

I seen 2.5" SCSI drive in some Dell blades 6 months ago so if the technology is go enough for servers I would assume it should do well in laptops, except battery sucking issue of course.

A dying technology

[imsabbel]

Dont missunderstand me, i thing magnetic storage has quite some time left.
But those kind of disks (very low capacity, high price, low access time) are going the way of the dodo very soon. This may well be the only generation ever to be created.

Because solid state/flash disks will rape its ass very soon, and are rapitely dropping into the same pricerange.
2.8 ms wont look nearly as nice if you can get the same capacity for a few bucks more with 0.1-0.3ms.

(and please dont start arguing about write cycle count. If you want to, then properly inform yourself about the current state of affair in the area. Hint: its really no issue, especially in the market those drives are targeting)

It's still not fast enough!

[AbRASiON]

2.5" or 3.5" or 5.25" it's JUST not fast enough!

(the snippet below is a re-post)
I love these things and I hate them, as an enthusiast I've always been a big fan of the high performance hard disk. I've done my best to learn about them, I've theorised about ways of speeding them up, I've discussed the technology with friends for hours at a time in a geek like fasion.

As much as I love a fast hard disk and I love a big hard disk I also hate these hard disks, because ultimately it's a very old fasioned method of storing our data, it's just some magnetic disc spinning same as it did 50 years ago.

When you really think about it, it's just a really extreme tape drive with better random access, there's moving parts, it's delicate, they can run hot, they can be noisy etc.

I recall my C64 as a boy, sure it had that weird "computer high pitch whine" to it but when the 1541-II wasn't reading data that baby was pretty damn quiet, I miss those days and hard disks don't help.

What we need is to finally see the end of the hard disk, some new method of storing data, something which holds more, reads and writes faster, less delicate and no moving parts - of course solid state sucks right now but damnit I recall discussing holographic drives storing data on a small cube the size of a peice of sugar at 2tb or something (so the rumours went, like 5 or 10 years ago)

The oven had the microwave replace it with a whole new tech, the television had the LCD / plasma, sending data has gone (at points) from copper to light - cmon where's the magnetic storage replacement, something to put us in the 21'st century?

So in conclusion, I love them but I also hate them - it's really time for something new,...

Re:It's still not fast enough!

[Gat0r30y]

Expect something to replace hard drives around the time quantum processors show up on the market. I would venture a guess that when something replaces light lithography in chip design, you will see your holographic drive.

Re:It's still not fast enough!

[macraig]

So what constructive storage inventions came out of those endless hours of brainstorming? Were you actually intending to share them and just forgot, or were you mum because the patents haven't come thru yet?

Re:It's still not fast enough!

[AbRASiON]

A conversation with a buddy who has a basic understanding about how a disk works and theories on how to speed them up does not make one an engineer or scientist who works at a laboratory at Seagate or something!
Sorry dude, it's more the musings of a couple of tech geeks who just hate shitty load times and archaic storage mediums.

Re:It's still not fast enough!

[AbRASiON]

I just re-read your post and I think I mis-read it as aggressive.

Ok well here's a couple of ideas I've speculated on, one of which I shot down myself as it's simply too expensive and complex to do and the other I beleive someone who actually had a real clue about disks straightened us out on.

1) Why not make a disk which is simply 25% longer and add a whole second disk head on the opposite side of the disk?
Example, normal HDD
=0

The 0 being the platter the = being the read heads.
Why not do this
=0=
2 sets of heads one at each end, it should reduce the seek time incredibly - unfortunately likely far more chance of failure, incredibly expensive etc - but hey! I really hate how slow these flipping things are!

The second theory was this.
When a disk is being written to, there's of course generally 1-10 heads, depending on how many platters you have.
Why not do a kind of "internal" RAID0 where head 1 writes the first byte, head 2 writes the second byte, head 3 writes the third byte and so on! since the head is physically in the same location as the other heads on the other platters, it would be awesome - theoretically 2->10x faster sequential read and write times.
Apparently though there's a very mild alignment issue with heads on a hard disk and it's just not that simple, the precise physical location for say sector 1121 on platter #1, side #1 is not exactly identical to sector 1121 on platter 1 side#2 or platter 4 side #2 for example :(

The third theory was the Y shaped head / actuator thing, it literally has 2 heads per platter!
So it only has to travel 50% the distance of the disk for head #2 to be at the "end" mean while only 50% back again for the head #1 to be back at the start - this again goes under theory number 1 - it's too expensive / delicate and the potential for breakdown is higher :(
(You could also mix it up with a similar kind of RAID0 theory as theory #2, where head 1 and 2 write at the same time, theoretically 4 heads per PLATTER - hot in theory,.. but not gonna happen :( )

and the final theory I / we / pals have theorised is that the disk heads can move independently, example platter 1 can be at the start of the disk, platter 2 the head can be at the end - however you'd need multiple actuators /motor thingos, again the cost / risk of failure increases and unlike the other suggestions, I don't know how much it could speed stuff up.

Sorry about the other post, I thought you were goading me or some such.

I _really_ REALLY! hate hard disks, I love them but I hate them - such a slow shitty technology - CPU's GPU's, internet / usb / firewire / serial / cdroms - all ALL have increased to relatively reasonable levels of performance but still to this day hard disks suck.

When I was 13 (goodness me, 16 years now) I recall my 286 taking about 30-60 seconds to fully boot with bios scans, floppy seeks and running the autoexec and config sys.
Now, 16 years late the PC STILL takes 30-60 seconds to boot, we need a faster way of getting things happening.

BTW I fully endorse Vista copying Mac OS with the suspend to ram AND write to disk incase kind of standby - where the battery still lasts but instant startup is always an option - brilliant.

Ok rant over - laugh or flame away

- Scott

Re:It's still not fast enough!

[macraig]

The post just struck me as annoying because, to me, it tempted with hints of ideas and then delivered anecdotes. Gimme the meat, I can take or leave the potatoes. Your followup was more "constructive". :-)

I can see the multi-head design being left on the cutting-room floor for perhaps three reasons: expense, volume, and complexity. Expense because of the need for not only extra heads but a second actuator for them; volume because that second actuator would force the drive to become relatively huge; and complexity because the signals from the opposing heads would have to be coordinated and interleaved.

Your second idea doesn't really require perfect alignment, at least to any greater degree than the normal read-write process. It does add that complexity factor again, though, and if data is being read or written through multiple heads/platters simultaneously it might hit a practical wall, might slow things down. There's also the matter that not every drive has multiple platters, or the same number of multiple platters: it might mess with "scalability" of designs (being able to alter the number of platters to produce different models) that disk makers so cherish.

The last one is definitely the stuff of dreams... I think it might even cause problems with maintaining proper head separation from the platter (not to mention alignment).

Actually, if I was gonna dream of any drive with integrated RAID behavior, it would at least be RAID 5! Not sure how you'd string such drives together in a single spanned volume then, but as long as we're just daydreaming....

Where's the IMPORTANT spec, platter-to-buffer rate

[macraig]

Great, this article mentions pretty much every spec except the most important one when assessing claims of disk drive "speed": the disk- or platter-to-buffer transfer rate.

On all the mass-market drives that I investigated up until fall 2005, that rate - which is effectively the sustained transfer rate of a drive - was less than 100MB/s... in other words, ridiculously less than the maximum bandwidth of SATA II/300, substantially less than even SATA I/150, and barely faster than ATA100 if at all. In the fall of 2005 I bought four 250GB SATA II Hitachi drives which, even though they had the SATA II interface, had a disk-to-buffer rate no better than about 75MB/s.

Now, granted, this new drive has an awesome rotational speed and perpendicular recording, but I wonder just how close it really comes to testing the limits of a new SCSI or SATA II interface's bandwidth? Had the most important of all the "speed" specs been reported, we might have been able to answer that.

I hate technical artiicles that leave out the most relevant detail....

world's fastest disk drive dollar cost?

[bobp0303]

how much???

Don't be too fast - even telex is still around ..

[cheros]

I know even telex is still making a tidy mint for some telcos (I think C&W still run a setup, even post Y2K where parts of the infrastructure needed upgrading).

Following that scenario it would give hard disks another two decades or so. Most likely better, smaller, more efficient (a 10MB disk used to be the size of a washing machine), but not yet abandoned..

Re:Perpendicular!

[Yvan256]

Anyone know how to convert the Flash animation into a real-world video format via QuickTime Pro?

Same as LD50

[RingDev]

Maybe I'm off in left field or something, but I was under the belief that MTBF was similar to LD50. Meaning that if something has 182 years MTBF, and I have 100 of them, 50 of them will be dead at 182 years.

-Rick

Re:Same as LD50

[pe1chl]

No. It means that when you have 182 drives in use for 3 years, about 3 of them will be defective.
This cannot be extrapolated to the number of defects after 182 years, because the MTBF is only defined for a certain service life (like 3 or 5 years).
After that, the number of failures will increase so the MTBF will decrease.

No really, 1.6 million hours is 1.6 million hours.

[RingDev]

Cisco defines MTBF as "Mean time between failure. Time at which 50% of the units of interest will have failed; used as a measure of the time a user might reasonably expect a device or system to work before a fault occurs." http://www.cisco.com/univercd/cc/td/doc/product/me ls/dwdm/dwdm_gl.htm#xtocid1301111

wiki defines MTBF as the reciprocal of the Failure rate (1 failure/16 million hours -> 16 million hours until a unit fails) http://en.wikipedia.org/wiki/Mean_time_between_fai lure

Seagate defines MTBF as "The average time before a failure will occur. This is not a warranty measurement. It is a calculation taking into consideration the MTBF of each component in a system, as well as, the statistical average operation time between the starting lifetime of a unit and the time of a failure. After a product has been in the field for a few years, the MTBF can become a field proven statistic." http://support.seagate.com/support/glossary/terms/ mean_time_between_failure.html

Western Digital's definition is also very similar http://westerndigital.com/en/library/gloss0803.pdf

In none of the definitions I can find online have I seen the warranty time span come into play.

If this drive has a 1.6 million hour MTBF you should indeed get 1.6 million hours of use from it.

From a batch perspective, if you had 182 of these drives, after 1 year, 1 of them would have a 50% chance of failing.

-Rick