1+ TB Hard Drive & Application Proposal

[Dïscfärm]

Using existing platter, head, and spindle technologies, an "ultra-capacity" hard drive could be created for special applications.

In this case, with a generational throwback to 4500 RPM (or even 3600 RPM), along with the use of modern high-density platters and heads, a hard drive mechanism could be created today with a total usable capacity of over a gigabyte. The low rotational speed of this proposed hard drive mechanism would allow for a high level of read/write accuracy for a given data density -- hopefully significantly more than 100 GB per platter. The low rotational speed would also result in a relatively low amount of operational heat. The use of fluid dynamic spindle bearings would reduce operating vibrations, contributing to a higher level of read/write accuracy. The target MTBF rating of this drive would be 1.5 million powered-on hours (or more).

The primary application of this proposed drive mechanism would be in special fault-tolerant online disc libraries, though the use in medium-performance disc-based file servers would be acceptable as well. This fault-tolerant online disc library would be a competitor to lower-end and midrange tape libraries by providing a cost effect way of storing and receiving files without dealing with the long access times common with tape-based storage. The library would provide each hard drive mechanism the necessary power and cooling required to operate. Each drive mechanism would plug into a backplane providing hot-swap services, RAID processing, and a switched architecture for maximum channel throughput.

Even though fault-tolerance could be defeated in this online disc library, the normal mode of operation might be simple RAID Level-1 (mirroring) with 2 (or more) hard drive mechanisms to guarantee fault tolerance as far as the storage medium goes. In order to simplify compatibility with existing hierarchical storage management software, this disc library firmware will have to perform tape library emulation (or as far as that goes, M-O or DVD library emulation) where a virtual picker is picking virtual tapes from virtual tape slots and plugging them into virtual tape drives, even though all the disc drive mechanisms stay plugged into their respective drive bays in the backplane during normal operation. The only difference in operational characteristics with this online disc library is that the file access times are vastly better ( only a few milliseconds -- worse case) and the read/write throughput is generally much higher than most existing tape drives.


...CRITIQUE

[edit, by special request: It's clear you meant the title to say TB instead of GB and you probably meant 100GB per head but I'll let you confirm before I change it.]

 

[Dïscfärm]

By the way, the proposed hard drive mechanism is a 3.5-inch form factor hard disc drive (i.e. -- 3.5-inch form factor slow spinning, very high capacity hard drive).

 

[CougTek]

A good idea IMO. However, the HD's enclosure would have to be tougher than the one of ordinary hard drive disk in order to insure a very high reliability and a good protection of the valuable data. Maybe not something as tough as a Zip750MB media (which can resist to be rolled over by a car), but something significantly less fragile than a common IDE HDD. And with higher temperature tolerance too.

Just my 2¢.

 

[Cliptin]

A good idea IMO. However, the HD's enclosure would have to be tougher than the one of ordinary hard drive disk in order to insure a very high reliability and a good protection of the valuable data. Maybe not something as tough as a Zip750MB media (which can resist to be rolled over by a car), but something significantly less fragile than a common IDE HDD. And with higher temperature tolerance too.

Just my 2¢.

Media in HSM systems are not removed. Think of HSM as data that you do not need everyday but when do you need it faster than you can restore from tape.

 

[jtr1962]

Intriguing idea, although I'm not sure offhand if there would be enough of a market to justify the hard drive makers doing the R&D work. Once enough home users are into video something like this will definitely make sense. Right now we can get 320 GB in a 3.5" standard form factor IDE drive. I would imagine by slowing the speed down to perhaps 3600 RPM and not caring too much if seek times fall into the 20 ms range you could pack something like 125 GB on a platter, and 8 such platters will fit into a standard drive since they can be made thinner(less flutter at low rotational speeds). Therefore, I believe 1 TB is perfectly feasible without any engineering breakthroughs.

In order to sell enough drives it will need to be no more expensive on a per GB basis than current bulk storage IDE drives, which can be had for $1/GB or so. My guess is that you can make an 8-platter 1 TB IDE drive reliable enough for home use for under $500. These will sell like hotcakes once enough people start using video on their machines. In the meantime, you can sell a better version of the drive, using more reliable parts to get your target MTBF to those who currently use tape libraries. Access times will be much faster than tape, and the drives will be more reliable as well. These drives will fill the void between tape and cheap solid state storage, which will come along within a decade, if not sooner.*

*To qualify this statement, right now many different types of solid state are being researched. My guess is that some type of polymer film will come out on top. Initially, the price per GB might be more than current magnetic hard disks, but the technology will quickly scale to ever higher bit densities, and it can be stacked much more easily then platters. Consider a polymer film drive in an IDE form factor. You can put perhaps 100 sheets stacked on top of each in the case. Each sheet can use almost the entire area, unlike the 5 or so square inches on each side of a platter. This is about 15 square inches per side. Let's say only one side can be used. Even at a bit density one-tenth of today's drives, you can put about 12 GB per sheet. 100 sheets=1.2 TB. Eventually, you can get densities far exceeding what is possible with magnetic spinning disks, so PB capacities should eventually be possible with this technology. Access times and reliability will be several orders of magnitude about magnetic disks as well. STRs will probably eventually be limited by the interface, not the technology, and this would be where SATA-300 and 600 come in.

 

[Explorer]

...the HD's enclosure would have to be tougher than the one of ordinary hard drive disk in order to insure a very high reliability and a good protection of the valuable data...

The casing (chassis) of these hard drive mechanisms wouldn't really need to be and stronger than any other existing hard drive since they supposed to sit dead still inside a disc library during use and only come out if they have failed. As far as security goes, tape and disc libraries normally have a decent level of physical security -- locked doors, intrusion alarms, etc. A bit of emphasis on passive cooling could be put into the design of the hard drive's metal case possibly, but the library is supposed to provide a bit of forced air cooling across all hard drive mechanisms which could easily be done with 2 to 8 fans with good air filtration (including a front panel that provides messages / alarms to change out the air filters on a regular basis). Each drive would be best suited to be mounted in SCA frames.

Maybe not something as tough as a Zip750MB media (which can resist to be rolled over by a car)...

Knowing Iomega, a Zip 750 drive cartridge could be run over by an automobile and survive, but be destroyed by their drive unit (i.e. -- "click of death").

 

[James]

But why? I don't understand the problem you're trying to solve.

 

[Explorer]

Intriguing idea, although I'm not sure offhand if there would be enough of a market to justify the hard drive makers doing the R&D work.

Unless you've never had any contact with it -- and many people haven't -- the enterprise storage market is absolutely a HUGE market that has been around for decades. You might be floored if you ever saw the yearly budgets of many companies for computer-based storage.

...I would imagine by slowing the speed down to perhaps 3600 RPM and not caring too much if seek times fall into the 20 ms range you could pack something like 125 GB on a platter, and 8 such platters will fit into a standard drive since they can be made thinner(less flutter at low rotational speeds). Therefore, I believe 1 TB is perfectly feasible without any engineering breakthroughs...

You've got the gist of what I was saying earlier in that there wouldn't really need to be ANY bleeding edge technology required to develop one of these high capacity hard drives.

In order to sell enough drives it will need to be no more expensive on a per GB basis than current bulk storage IDE drives, which can be had for $1/GB or so. My guess is that you can make an 8-platter 1 TB IDE drive reliable enough for home use for under $500.

To make this hard drive price-competitive with existing storage technologies (tape, magneto-optical, DVD-RAM, DVD-R, CD-R, etc.) its market price would have to be set so that it would provide a reasonably attractive price/performance ratio for new installations compared with other existing technologies. Once enough new installations came along providing user feedback, the customers looking to upgrade and/or convert from "conventional" storage technologies would come along.

These will sell like hotcakes once enough people start using video on their machines.

There could be a market for this hard drive mechanism with top-end consumer TiVO-like video set-top recorders and possibly with desktop video professionals as long as its relatively-slow seek and access times weren't a problem.

In the meantime, you can sell a better version of the drive, using more reliable parts to get your target MTBF to those who currently use tape libraries. Access times will be much faster than tape, and the drives will be more reliable as well.

There would have to be only ONE grade of this hard drive mechanism coming off the assembly line. As far as interfaces go, SATA, F-C, SA-SCSI, Ultra320 SCSI, and even ATA6 (parallel ATA) would all factor in for sure. For a disc library, F-C, SA-SCSI, and Ultra320 SCSI would be ideal.

...My guess is that some type of polymer film will come out on top.

Yes, I was recently talking about "Plastic Memory," as I have been for the past ...er ...couple of years or so on occasion. There seems to be a lot of potential with polymer thin-film memory and circuits in general. Initially it will show up as a cheap alternative to Flash Memory, then maybe a bit later something more.

 

[Explorer]

But why? I don't understand the problem you're trying to solve.

One of the problems that would be solved with a hard drive mechanism like this would be the poor access time of tape in tape libraries. As a "tape alternative," these high capacity hard drives would provide far better access time than any tape format, but still provide a level of storage density that can only currently be attained with AIT3, LTO2, SuperDLT, SuperAIT, and a few esoteric tape formats like DTF -- all a competitive price that magnetic hard drive storage can provide today.

The cost of a switched backplane for these hard drives to interface would be significant, but that cost would be offset by the very high price of multiple enterprise-level tape drives.

 

[Jan Kivar]

Hmm... How much more capacity would 5,25" platters give?

Jan

 

[CougTek]

Hmm... How much more capacity would 5,25" platters give?

The ghost of the dreaded BigFoot is still wandering...

 

[Cliptin]

Descriptions and an article about HSM including "esoteric DTF":

http://itmanagement.webopedia.com/TERM/H/HSM.html
http://searchstorage.techtarget.com/sDefinition/0,,sid5_gci214001,00.html

http://www.enterprisestorageforum.com/management/features/article/0,,10558_1011401,00.html

 

[P5-133XL]

Hmm... How much more capacity would 5,25" platters give?

Jan

Assuming a 1" inner dead-zone and a 1/2" outer dead-zone (The dead zones may vary depending upon armiture design) then
(PI*(5.25-.5)^2-(1)^2)/(PI*(3.5-.5)^2-(1)^2) = 21.6/8 = aprox 2.7X the capacity.

Note, that this is a rough model, just using the area of the two circles for compareson. Inactuallity, you can get a more accurate measure by using TPI (Tracks per inch) for a specifc model and BPI (Bits per inch) * the circumfrence of each track. However, the area compason works fine as a "rough estimate".

 

[Platform]

Hmm... How much more capacity would 5,25" platters give?

A drive using 5-inch platters would give you more storage density, but at a high cost today because the assembly lines for manufacturing 5-inch hard drive platters have long since been shutdown.

CougTech mentions the Bigfoot hard drive. The Bigfoot drive was a joke in its own time, a time when 5.25-inch hard drives were well on the way out. Bigfoot drives were essentially a product devised by using some spare end-of-life capacity that Quantum had somewhere. The economics were marginal at best and they knew they had a product that wouldn't last more than a couple of years, but they pursued it anyway. Needless to say, the failure rate with Bigfoot drives was pretty high and they were performance slugs.

The general idea with this proposed hard drive is to use as much off-the-shelf parts and current engineering knowledge as possible. With a reduction in spindle RPM to 4500 RPM, or even 3600 RPM, various operating parameters that govern platter density can be tweaked to elevate information density significantly, while maintaining the same level of detectable and recoverable read/write error rates that are acceptable with current 5400, 7200, 10000, 12000, and 15000 RPM hard drive mechanisms.

 

[Jan Kivar]

How much data can tape libraries contain? 10 TB?

Jan

 

[Mercutio]

As much as you have room for. I've seen PB (petabyte) libraries... sealed rooms with robots running around on tracks at 100km/hr. Hard to compete with tape on that level, I think.

 

[Cliptin]

As much as you have room for. I've seen PB (petabyte) libraries... sealed rooms with robots running around on tracks at 100km/hr. Hard to compete with tape on that level, I think.

One advantage to tape on that scale is that the tapes don't have motors. Only the robot and the tape drives are producing heat. It's a moot point if the system can't get data moved around fast enough though.

 

[jtr1962]

The main advantage of this whole idea over tape is the huge decrease in access times. Even if the 1 TB+ drive is a real slug with access times over 30 ms, it will still be orders of magnitude faster than a tape library. Heck, I could probably look in a card catalog, take a book off a shelf, and go to the page I want faster than it takes to get a tape and wind to the correct spot. I remember well backing up on QIC-120 tapes and then needing a particular file. No exaggeration that I could sometimes take a shower and the file still wouldn't be retrieved. That's why, despite it's drawbacks, I find CD-RW such a wonderful backup medium. I can retrieve the file I want in a NY minute(sorry, I couldn't resist that :mrgrn: ). A tape library replaced by drives would be similarly faster.

Heat production of a slow drive should be minimal, even with 8 platters. Power needed increases with the cube of the speed, so cutting the rotational speed to 3600 RPM should cut the power needed by a factor of eight compared to a 7200 RPM drive. Such a drive at idle uses ~7 watts even with 3 platters, so the 1TB drive should come in at around 1W or less, even accounting for the extra platters, since most of the friction at 3600 RPM is bearing friction, not air resistance. You could put 20 of these in a metal case about the size of a normal PC, and not even need any cooling fans.

 

[Fushigi]

The main advantage of this whole idea over tape is the huge decrease in access times. Even if the 1 TB+ drive is a real slug with access times over 30 ms, it will still be orders of magnitude faster than a tape library. Heck, I could probably look in a card catalog, take a book off a shelf, and go to the page I want faster than it takes to get a tape and wind to the correct spot. I remember well backing up on QIC-120 tapes and then needing a particular file. No exaggeration that I could sometimes take a shower and the file still wouldn't be retrieved. That's why, despite it's drawbacks, I find CD-RW such a wonderful backup medium. I can retrieve the file I want in a NY minute(sorry, I couldn't resist that :mrgrn: ). A tape library replaced by drives would be similarly faster.

QIC-120 is a very very very dated format and was never Enterprise-class. Modern Enterprise-class tape drives can load a tape and seek the desired file in nicely under a minute. IBM Ultrium, AIT, and probably S-DLT are all capable of finding & restoring data pretty quickly and they are among the smaller tape systems available nowadays.

AIT3 tapes, according to pricegrabber, are $62 for a 100GB native; 200-300GB compressed capacity tape. AIT tapes have a microchip that stores tape positioning info to speed up the location of files on a tape and can generally find any file in well under a minute. The tape is smaller, lighter, and more durable for transport to offsite locations than any 3.5" and most current 2.5" HDs. Which makes tape a better medium for offsite storage.

Enterrpise-class backups will continue to be tape based for the foreseeable future. Tape technology is currently very good and scales as high as necessary. Check out http://www.storagetek.com/pdfs/9310PowderHorn_DS_MT1006F.pdf for a tape system that does up to 100TB/hour backup.

- Fushigi

 

[Corvair]

...It's a moot point if the system can't get data moved around fast enough though...

Which is why larger tape libraries have the ability to house several (expensive) tape drive modules so that a tape cartridge can (statistically speaking) find an idle tape drive to insert into.


There are a few of these beasts behind now here at work...

 

[Corvair]

> There are a few of these beasts behind ME now here at work...

I think the ECC circuits in me noggin haven't been working properly lately (wacky work hours), as I have regularly been omitting words from sentences for some reason.

 

[jtr1962]

QIC-120 is a very very very dated format and was never Enterprise-class. Modern Enterprise-class tape drives can load a tape and seek the desired file in nicely under a minute.

I had no idea they had gotten that fast. Still, if there's a situation with many random seeks, the drive solution might be better. 30 ms is still 3 orders of magnitude faster than a state-of-the-art tape, although a HDD is obviously not as portable or duarable. I suppose there's enough markets for the 1 TB HDD and tapes to coexist.

Check out http://www.storagetek.com/pdfs/9310PowderHorn_DS_MT1006F.pdf for a tape system that does up to 100TB/hour backup.

Very interesting, and frankly I'm floored that any enterprise would have need to store up to the 28.8 PB of data that this device scales to. Obviously such demand exists, or they wouldn't be making it. I heard that the collective works of mankind or the entire Internet would occupy at most a few tens of PB, so exactly what do the people who use these things store on them? I get a headache just thinking about it.

 

[James]

But why? I don't understand the problem you're trying to solve.

One of the problems that would be solved with a hard drive mechanism like this would be the poor access time of tape in tape libraries. As a "tape alternative," these high capacity hard drives would provide far better access time than any tape format, but still provide a level of storage density that can only currently be attained with AIT3, LTO2, SuperDLT, SuperAIT, and a few esoteric tape formats like DTF -- all a competitive price that magnetic hard drive storage can provide today.

The cost of a switched backplane for these hard drives to interface would be significant, but that cost would be offset by the very high price of multiple enterprise-level tape drives.

Yes, I get all that. But my question is, is there actually a requirement for what you're trying to build here? In my experience there are requirements for things that clearly require HD technology and ones that clearly require tape technology (of various types depending on specification).

Sure, large scale tape systems are very expensive but then the demand for them is not widespread (compared to the enterprise drive market, say). They have a specialised application, and in those cases a minute or so to seek a particular file is not critical - I have seen them deployed in environments where the data needed is very large but only retrieved very infrequently. (For example, land and title information, mechanical drawings, and suchlike.) In this sort of enviroment the additional moving parts of the drives you describe are a liability since they reduce the MTBF.

Now tape is not really a suitable long term storage medium, but since the media itself is relatively cheap and portable it is practical to make several copies (and store extras in a duplicate facility, fireproof safe, etc.). This is much more problematic with hard drives.

Most large enterprise storage systems have a quickly-accessible database at the front end with the data that is required 90% of the time available immediately, with the low priority bulk stuff available on tape. It's a system that works well for most applications.

The only application I have encoutered where I had several TB of storage requirement and tape was not suitable was where the volume of data was such that the transfer speed off tape simply wasn't fast enough to meet the distaster restore scenario. In that case I deployed NetApp filers and it was horrendously expensive, but then the customer could afford it. This seems like really the only space the device you describe could play in, and it's a very small niche indeed, surely?

 

[Fushigi]

Very interesting, and frankly I'm floored that any enterprise would have need to store up to the 28.8 PB of data that this device scales to. Obviously such demand exists, or they wouldn't be making it. I heard that the collective works of mankind or the entire Internet would occupy at most a few tens of PB, so exactly what do the people who use these things store on them? I get a headache just thinking about it.

Maybe they want to store every copy ever made of mankind's collective works.

Seriously, it is common to have multiple copies of most data floating around in these systems. Since the tape silos are stationary, and they hold a lot of tapes, it is possible at any given time that you will have new tapes ready to be used as well as used tapes ready to be reused as well as used tapes not ready to be used (their contents have not expired yet).

The silos are huge and typically require not only a computer room (floor tiles, etc.) but a raised cieling as well as they are rather tall (at least the old Timberline models did). Capacity is expanded by attaching multiple silos together. There are safety sensors to make sure the robot arms don't move if a human is detected as the arms move at 70MPH and can easily injure a person. $eriou$ $tuff.

Anyway, a growing method of storage is Hierarchical Storage Management. The article is for the AS/400 and iSeries world but can be applied to other architectures with some changes. Where I see a possible use for the TB disks is replacing the middle layer in an HSM structure with high-capacity-but-slow disks. A further advantage could be gained if the disks feature onboard compression units; most database data is very compressable and that TB could turn into 3 or 4 TB of compressed data.

- Fushigi