Slashdot Mirror
Ultra Fast Disk Drives With No Moving Parts
saccade.com writes "Let's face it, the slowest part
of PC's today is the disk drive. Bit
Micro has come up with a nifty solution - flash memory based
disk drives available in typical
disk
form-factors. These e-disks are electrically compatible
with ATA, SCSI, etc. but run orders of magnitude faster - access
times down to 40 usec and transfer rates over 100 MB/sec. What's
the catch? Cost. Currently going for just under $1K/G, a 30G model
I recently held in my hand was worth much more than my car. However,
as flash memory prices drop, so do the price of these drives.
Within the decade the spinning hard disk may go the way
of the floppy and CRT."
Isn't an ultra-fast, no-moving-parts hard disk called a soft disk? You know, ROMs and memory and all that stuff.
I wonder how long you can beat at a device like this in a server environment before it croaks. I'd give it no more than a year life expectancy, but hey, I'm feeling pessimistic.
I am feeling fat and sassy
I need an EE to build an ata interface to a raid series of about 100 flash either (SD or compact). Now allow the end user to plug in how many cards he wishes and just use them. Imaging that if you have a raid 5 setup of say 128 256mb cards costing about $40 each would cost about $5000 1/6th of the $30k and it is end user upgraded and so cool to be able to ad more storage instead of rebuilding a whole computer and drive.
Flash disks. They've been around for quite a while, why do a slashdot story now?
Maybe not, but if they start going a little bit mainstream, we'll start to see the cost go down. I know I've thought about using some sort of flash device for my boot drive just to have extremely fast boots.
100,000 writes isn't gonna last long in todays bandwidth intensive video/mp3 world
no moving parts and non-magnetic media is a worthy goal but until we can cure terrible storage lifetimes they wont be much use if i have to worry about the mess backups of backups, as we know from sci-fi all it takes is a big EM burst from the sun and everything you and i have done is gone !
future generations will look back at us and say "they used to store it on WHAT !?"
I've always found the best way to deal with the problem of slow disks is to max out the memory in the PC and use a hefty chunk of it as a RAM disk. When done or needing to backup, tarball the whole disk, write it once to the hard drive.
:)
Of course, this assumes you're working on a stable OS with decent tools and good memory management. If you're not, you can be.
Everything in the Universe sucks: It's the law!
I know that 10000 writes seems like a lot, and perhaps it is. Anyone knows how this figure looks for normal harddrives?
Still it seems to me that the limited number of writes sets the biggest limitation.
I did an embedded application with a flash disk which emulated a floppy. In the autoexec: create RAM disk, copy whole sheboodle, run from ramdisk. Without this the device only lasted 2 years. Can't see you do that with XP on a 10 gig drive though... I guess it would be good for a non-dynamic server. Host all the Slashdot logo's on one?
10 ?"Hello World" life was simple then
...printer.
Technically, a printer is a peripheral, not a part. Whatever. All printers are evil: Too slow, too big, too expensive, too quirky. Ackk.
-- Slashdot: When Public Access TV Says "No"
Here's an idea: Performance will be nearly as good, reliability will be substantially up, cost will be a lot lower:
Use a traditional hard drive, but with a RAM cache that's as large as the drive. The drive controller uses idle time to preemptively load data into the cache. There's a battery backup so that the drive can continue operating after powerdown, and the system uses a long time period write behind cache with write combining to reduce drive usage in operation.
Until last year, I would have an employee come to me every 6-8 weeks with a beatup floppy containing their sole copy of some critical spreadsheet or database file... the floppies were clipped to a clipboard or had been flopping around in the bottom of someone's purse - the data was almost always unrecoverable. And despite my warnings, never a backup.
Our solution - new 'legacy free' PCs with no floppy drives. There was initial complaint, but now the users have discovered other ways to tote data around - and we don't lose that critical data like before.
Everyone will start to cheer when you put on your sailin' shoes.
Within the decade the spinning hard disk may go the way of the floppy and CRT.
I'm writing this from a workstation around a year old that has both a CRT and a floppy. They both get used (albeit one more than the other). Just because you don't use them doesn't mean other people do the same. I'm no futurist but I predict with my magic powers that based on cost/performance CRTs will still be around at the end of this decade. Floppies, maybe not so much.
Speak truth to power.
Hell, why don't we have that now? Why don't we have an affordable caching controller that will take a dozen commodity 512MB memory modules? Or a self contained 3.5" disk based on a 1.8" 20 or 40gb drive and a few gigs of battery backed cache?
Old enough, so the first 'generation' of SSD companies is already out of business. E.g. Platypus (I think that was the name) build RAM based solid state drives, some of them in the right shape and with appropriate disk interfaces to match existing disk drives.
I looked into SSD for a database at one point. But I found that you can get almost the same performance by using lots of drives in a fast RAID setup. Striping the content over multiple disks does wonders! And its much cheaper.
E.g. look at something like a 12 disk setup with RAID 5+1. You got a full mirror, and essentialy 4-8 times the speed of a single drive. So you are already close to the 'order of magnitude' they SSD drives claim.
---- join dshield.org Distributed Intrusion Detec
the 17-inch TFT on my desk cost me something like 450 euros, and it does gaming just fine. Max Payne, Soldier of Fortune, UT2004 etc, etc.... Zero problems.
Gaming wasn't that nice with those old 25+ms panels, but newer 12, 16 and 20ms panels are ALOT better! And we will be seeing 10ms panels in the near future!
Lesbian Nazi Hookers Abducted by UFOs and Forced Into Weight Loss Programs - -all next week on Town Talk.
I recently bought a 200GiB hard drive and if it was made of flash memory and cost the same, I should have payed 1600$ worth of taxes. Or roughly 10 times as much as the hard drive itself.
Until this tax insanity blows over, I don't see the technology going anywhere regardsless of how cheap they can build it.
(*): probably a little less, but I didn't bother to look it up. 3.20 DKR per 64MiB - do the currency conversion yourselves.
________
Entranced by anime since late summer 2001 and loving it ^_^
Our current mass storage interface standards encompass concepts firmly attached to the physical model of rotating disk(s) with read/write head(s) that can operate on cylindrical tracks.
If flash memory drives become the norm, are these interfaces (ATA, SCSI, etc.) obsolete? Is there a set of primitive operations that map to a flash drive better than retaining those created for spinning media? Could flash drives like these simply be memory mapped and treated more like a cache?
Babies are cute because they have to be.
Not always. Sometimes things are expensive because they are technically difficult to manufacture, or because the raw materials are expensive, or because the environmental regulations are expensive.
memory chips require many expensive and hazardous chemicals to manufacture like fuming sulfuric acid for dissolving the photoresist inks and hydroflouric acid for etching the circuits. These chemicals have a large environmental regulation cost associated with them that's not going to go down any time in the forseeable future and is entirely outside the control of any manufacturing process.
You can tell a great deal about the character of a man by observing those who hate him.
You young whippersnapper! Why, I remember, back in the day, when the sysop of a BBS I was on was collecting donations to get a 1GB drive... it cost $1000.
And we liked it! Uphill, in the snow, both ways! And at 2400 baud!
Live simply, that others may simply live. -Gandhi
What if you merged a flash device with a battery-backed RAM drive? Keep all your ordinary I/O interface with the RAM drive and then periodically mirror RAM to flash with a single write cycle?
It still wouldn't last forever, but it might be a lot more practical for ordinary use; although you might consider just mirroring it to a HDD as well.
Anyone remember magnetic bubble memory?
I know Nat Semiconductor does. They sank ALOT of cash into the concept in the "early" PC era.
It worked. It worked well. Capable of storing data w/no power. It was going to replace disk drives an system memory.
But while it worked, it worked not as well as the SDRAM of the day or the less that 1 gig drives that were common then.
They never got close enough to breaking the price/performance/capacity "wall" that the others did. The ecomony of scale they hoped for never came through.
I'm not sure, but it might have some uses still as NVRAM (or might be renamed flash memory for all I know)
"...and is entirely outside the control of any manufacturing process."
ANY process? I think that was the point - if someone can come up with a new process, we could reduce costs. The more these are used, the more incentive there is to research new processes.
As far as I can recall, there ARE people working on alternatives to memory as we know it.
The same thing happened with LCDs, as pointed out - CRTs have a bottom line cost - the cost of the components have a bottom line that means that LCDs should, at some point, be cheaper - the processes are still be refined and improved, and there's not a whole lot of leeway anymore with CRTs.
Stupid sexy Flanders.
Without giving away too much (and getting fired in the process) there is a whole new tech on the horizon. It still uses all the nasty chemicles, but in traditional flash memory, the chip is broken into three major components:
charge punps (to provide the 9.5-12 volts required to program the chip from the punny 1.8 - 3.3 volt supply
the control circuitry (basically a mini CPU)
the flash array
all these elements are "flat", that is they are one structure deep. This new tech coming up, if someone can perfect it, uses multiple layers to make the flash array several layers deep. Thus you could (in theory) shrink your die size while increasing the memory density.
-nB
whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
Doesn't the technique of 'sleep to ram' solve your problem entirely?
The system is 'perpetually' on and a booted system is stored in (low power) ram, mirrored to the hard drive of course in case power goes out, so boot only takes seconds?
I mean, that's what *I* do. Start up the computer on a daily basis in less than three seconds, most of the time just waiting for the monitor to rez.
GPL Deconstructed
A CF/IDE adapter is a cheap, commodity item.
With COTS parts, you can run 4GB of flash for
about $500. Problem is, you need a filesystem designed for memory with limited write cycles. Just turning off metadata updates would help a lot.
-I like my women like I like my tea: green-
all these elements are "flat", that is they are one structure deep. This new tech coming up, if someone can perfect it, uses multiple layers to make the flash array several layers deep. Thus you could (in theory) shrink your die size while increasing the memory density.
This turns out not to help much. Multi-layer chips add mask steps roughly in proportion to the number of layers. While you save on the cost of wafer area, your processing steps cost a lot of money too, so you rapidly reach a point of diminishing returns. Building multi-layer devices also requires making transistors on epitaxial silicon layers, which generally have far worse performance properties than the monocrystalline wafer (even SOI processes generally work by building devices on a silicon wafer, and either flipping the chip and back-etching or using a buried oxide layer, as opposed to depositing a silicon film).
3D chips have been a holy grail for density reasons for decades, but they turn out to be expensive to manufacture and poorly-performing for the reasons noted above, and for microprocessors, at least, they're now a pretty much obsolete solution, as heat generation is what limits chip performance (and a multi-layer chip gives you that much more heat generation per unit area).
If your company can pull it off in a useful way for storage, they'll deserve kudos, of course.
When I was in college in 1998, my college had a SSDD (Solid State Disk Drive) setup as a swap-file disk on our class registration database server. It was only Ultra Wide SCSI-3 (40 Mbit/s), but adding that drive as a swap partition and storage of temporary tables cut the time required to register for any give class from 3 minutes to 15 seconds.
Carpe Pisces