This is something I came across today.
It's the announcement of a new product : a 3.5" solid state harddisk with capacity up to 155Gb ! (they also have a 2.5" version - up to 35Gb)

This could be the answer to everything I've ever wanted from my harddrives : no noise and fast access times !

Of course, I hope I won't have to sell an organ to be able to afford one.

I'm not sure one organ will be enough. That's going to cost an absolute fortune!

Hope away, sounds like you might need to lose a kidney to me though!

512mb CompactFlash is selling for $165.99 from Crucial. I don't know the margins, but let's say that the solid state RAM inside that costs $80.

That would make a 155GB disk cost $24,000! Or a 10GB disk cost $1,600.

Now my estimate as to the cost of solid state memory is a total guess, but can they really bring that down by a factor of 10, which is what it needs to make it even close to being realistically affordable for the average consumer?

That was also about the amount I came up with. There's no way I'm able to afford that kind of money for a harddisk (nor would I want to), and I suspect 99.99% of all the people think about it that way.
Therefor I expect that company's got something up it's sleeve price-wise because if they haven't I predict this product will die a fast death, probably even before it actually ships.

It's not marketed for consumer use. It's for when you've got very high IO requirements and need the fastest possible storage like enterprise databases. And if you've got such a large database that really does needs this drive then you're going to have bags of money to spend. So I'm afraid this isn't what you're wanting for if you want low cost solid state storage.

- Trevor

Something like this would be useful for databases. Currently, as incredibly useful as they are, databases are still constrained in what they can do by the hard drive access times. The lower the access time, the more powerful a database you can design.

Most of us don't have the need for an application like this, nor do we have the money to throw at it. I'm going to do the next best thing, personally, and design a system around a ramdisk at some time in the near future. However, copying 2 gB of data from the hard drive every time you boot up is going to be a pain, and, if the system crashes, all that data is gone unless you copied the changes back to the hard drive. For my purposes, it should work out, but for people that cannot deal with these limitations, one of these solid state hdd's might be perfect.

-Biscuits

That confuses me. The only vaguely affordable way to get such a large capacity would be to use NAND Flash, but that is way slower than modern hard drives. With a very efficient control algorithm you can get somewhere between 1MB - 2MB/sec out of most devices.

It'll would be interesting to see a standard benchmark of these solid state drives. The company quote data rates from 4MB/sec to hundreds of mb/sec and I can't help but think that the second figure is based on 100% cache hit.

Rob

I suppose if they stripe the data across several segments of the Flash memory then I suppose you could increase the throughput?
If it's NAND Flash then don't you have more issues with accessing it? A NAND ROM in a PDA needs the data copied to RAM before it will run as you can't do XIP.

- Trevor

Naw, too slow compared with SDRAM and spinning media.

These are for ruggedized embedded applications -- data recorders, mobile computing in extreme temperatures/conditions, etc..

At least until they get (MUCH) cheaper.

Cheers

Hmm...This is just confusing;

The speeds quoted suggest to me that the limiting factor is the IDE interface, and not the underlying media.

That raises the question "Why IDE?". Although IDE speeds are far superior to those of 5 years ago, and are now comparable to SCSI, IDE is still inferior to SCSI in one important aspect - Bus topology and contention.

You can still only have 2 devices on an IDE channel, and to get best performance, ideally you still only want 1 per channel. SCSI on the other hand deals with many drives on a controller without as much degradation.

If this was targeted to enterprise database applications, then surely it would have been designed with a SCSI interface. Sure, you could encourage customers to buy expensive multi-IDE controller cards to drive these, but as they are generally already set up for SCSI, that doesn't make sense.

It's my guess that we'll all own these soon - sort of. I'd hazard a guess that these have been designed with military and aerospace datalogging applications in mind.

If this was targeted to enterprise database applications, then surely it would have been designed with a SCSI interface

Their site's product section shows they also have SCSI solutions.

They seem to support multiple platforms. The only solution I can't seem to find amonst them is serial-ATA (although it's mentioned on their site, just not in their product section)

Hmm...I should have surfed the site a bit rather than just looking at the link posted before opening my big gob

I still stand by the statement that the IDE drives are not targetted towards the enterprise database market though.

You would be very suprised then..

The only thing really holding ATA drives back is that the manufacturers save their fastest mechanism (10K-15K RPM) for their high priced SCSI lines. RPM for RPM, ATA drives outperform SCSI in any decent benchmark, and several indecent ones too.

Even with SCSI, it's desirable not to have more than a couple of heavily accessed drives per SCSI bus, to reduce contention issues.

Sure, SCSI can do a decent job if one opts for very very expensive fibre channel busses and the like, but in real life, ATA wins hands down for interface cost and performance.

Heck.. the current contract work I am doing here involves a very nifty ATA interface, that can handle up to 8 drives per channel, with full overlap between them.. requires modern drives, of course -- with tagged queuing. These days, that means IBM DeskStar drives.

Quite a fast system. I just wish I could obtain 10K or 15K RPM ATA drives to really make it smoke!

The reason for the enduring popularity of the ATA interface is it's electrical and software simplicity, coupled with quite high performance. This makes it ideal for embedded applications, and anywhere that cost is important.

Cheers

You would be very suprised then..

Well, not 'very' surprised...it's been obvious for a while that ATA has been biting more and more out of SCSI's traditional markets with every performance iteration.
But I wasn't aware that it was quite as close to being in a position to overthrow SCSI as 'king of the hill' as you have alluded. As I see it (and I am obviously not the expert that you are), SCSI's only real benefits over ATA have always been performance and scalability. Once those are removed ATA wins.

The only thing really holding ATA drives back is that the manufacturers save their fastest mechanism (10K-15K RPM) for their high priced SCSI lines.

Do you think that is to deliberately and artificially prolong SCSI's market share, or is it that the costs associated with developing the fastest mechanisms are what drives SCSI pricing? If the latter, would there be a strong market for 15K RPM ATA drives if that resulted in those drives being identically priced to 15K SCSI drives?

How do you see serial ATA affecting the SCSI market?

The reason for the enduring popularity of the ATA interface ... This makes it ideal for embedded applications

That's why I suspected that the intended market for these solid state disks would be military and aerospace. (And eventually in-car/personal mp3 players )

Sorry for all the questions, I hope that you understand that the opportunity to gain insight from someone who is a guru in this field is too good to pass up.

but can they really bring that down by a factor of 10,

Of course they can.

Don't you remember when regular old desktop RAM was approaching $50 per megabyte? That was only about six years ago if if I recall correctly. How exciting it was when the price came down to a dollar a megabyte just a few years later, a 50x price reduction! And today, what does it cost? Less than a dime a megabyte. That isn't a reduction by "a factor of 10" -- that's a reduction by a factor of 500, over a period of seven or eight years.

The cost of mechanical hard drive storage has dropped by an even larger factor over the same period of time. (From over a dollar per megabyte in 1995 to under $2.00 per gigabyte today)

So, yes, I believe they can "really bring that down by a factor of 10." It may take a year or two, but it will happen.

tanstaafl.

Well, yes, eventually. But by that time this company will be ground into the dust of bankrupcy...

Well, it's starting, first 10k ATA drive: http://www.storagereview.com/welcome.pl/http://www.storagereview.com/articles/200303/20030305WD360GD_1.html

About bloody time.. and as the article points out: WD was first with it, simply because they don't have an overpriced SCSI line to protect.

But pitty the performance isn't really there, but others will follow..

For now, the fastest ATA setup is IBM Deskstar drives on a Pacific Digital DiscStaq ADMA controller card.. very very impressive performance.

Cheers

For now, the fastest ATA setup is IBM Deskstar drives on a Pacific Digital DiscStaq ADMA controller card.. very very impressive performance.

Oooh, I'd not heard of those. Are they better than the 3ware kit then? (It seems to be 32/66 PCI for a start, and only 2 channels, whereas someone on LKML said his 8-channel 3ware could saturate 64/66 PCI.) Or is 3ware not an "ATA setup" within the meaning of the act? (ISTR the host perceives it to be a SCSI controller.)

Mind you, it is half the price of even the 2-channel 3ware.

Peter

An 8 Channel 3Ware Escalade 7500-8 or 8500-8 can easily saturate PCI64/66, and they also have 12 Channel versions.

They use ATA for the drive interface, but have a proprietary "switch" that allows addressing and full UDMA133 / ATA150 access to the drives simultaneously. In a RAID0 situation it can easily saturate the entire bus. So that's when you use PCI-X and get a Bus per slot hehe.

RAID5 is a little slower, but you can average about 200Mbyte/s read and about 120Mbyte/s write

They are pretty slick, and are awesome in action. One of the Escalades with the new Maxtor 1 Million MTBF near-line commercial IDE drives, and the sucker will last a life time.

Oh and they support hot swapping of the drives.

>ooh, I'd not heard of those. Are they better than the 3ware kit then?

Dunno. But for just two channels (4 drives, all of which can be in use simultaneously), the PacDigi board is awesome, and can saturate the PCI bus with just the two channels moving data around rapidly.

The nice thing about it is that it is NOT raid-specific --> very useable as just a desktop interface for everyday stuff. It really offloads all of the ATA register twiddling from the host (a typical ATA transfer requires 8->12 writes to the ATA registers, at 120-667ns per access, depending on the PCI bridge. That's a lot of CPU cycles to spend just sitting around writing registers.

The PacDigi offloads all that from the CPU, and does neato things like tagged queuing (to IBM/Seagate drives, nobody else seems to have it yet) in hardware, and handles auto-polling to keep both drives per-channel active at the same time.

Cheers

Mark,

considering that you seem to know about this card, and given that you are the Linux IDE guy, is there any support for this card in Linux, as their website only seems to mention MS OSes.

Thanks,

There's basic PIO support available for it in Linux -- but Alan blew the 2.4 merge, so it doesn't actually get enabled in the current 2.4.xx kernels -- patch pending for that.

For the much faster "Automatic DMA (ADMA)" operational mode, a special driver is required. They have a binary-only RedHat-7.3 driver on their website, and I'm trying to convince them to release source code for it to make kenel compatibility a non-issue.. soon, I think.

Meanwhile, Andre H. has threatened to write an ADMA driver for it from scratch.. not something I'd really trust, knowing what I know.

Cheers

Cool, thanks for the info, I will let that sway my current purchasing decisions, or I may just pick one up anyway as I have been spending so much on stuff for the house lately I feel the need to buy some 'me' things

Thanks,

Mmm... I lead a somewhat sheltered life here at times, so perhaps somebody out there can tell me if this (below) is normal or not.

I have an ATA ("IDE") controller in my desktop box here that I'm writing drivers for, and have a small arsenal of high speed hardware to plug onto it.

Here's some of the stranger (to me, normal?) things I've been torture testing it with:

1. Two high-speed CD-RW drives (24X and 48X) on a single cable, both writing CD-Rs simultaneously at 20-30X.

2. One high speed CD-RW writing at 48X while watching a movie on a DVD-RAM drive, on the same cable at the same time.

3. A high speed tagged queuing IBM Deskstar harddrive, doing a full image backup at the same time as watching a movie on the DVD-RAM, on the same cable as the Deskstar drive.

4. One high speed CD-RW writing at 48X while dumping my entire root disk to a Deskstar drive on the same cable as the CD-RW.

This all works fairly smoothly, with only the VERY rare glitch in playback/recording.

Needless to say, this is under Linux.

???

2. One high speed CD-RW writing at 48X while watching a movie on a DVD-RAM drive, on the same cable at the same time
3. A high speed tagged queuing IBM Deskstar harddrive, doing a full image backup at the same time as watching a movie on the DVD-RAM, on the same cable as the Deskstar drive

Watching a DVD while waiting for a CDR to burn or a backup to complete is probably a valid use case.

4. One high speed CD-RW writing at 48X while dumping my entire root disk to a Deskstar drive on the same cable as the CD-RW

And burning the important data to CDR while only copying less than important stuff to disk is something I normally do quite often.

I guess the question of what is connected where means some of these things could happen together, as generally I do not reorganise my IDE devices depending on what I am doing.

This all works fairly smoothly, with only the VERY rare glitch in playback/recording

As an end user I could handle my movie briefly glitching in those sort of situations. Are the glitches with recording present of the finished copy, or is it just the process that glitches? If it is the process then I wouldn't mind, If I have data corruption then I am not happy!

On the vast majority of computers, the CD burner and/or DVD player are NOT on the same cable as another high usage device like a hard drive.. so of course they're mostly glitch free.

I'm just wondering about the more extreme situation of when two very high usage devices are sharing the same cable.. or even just two ATAPI devices.

Cheers

I think his point is that there usually isn't such a thing as a small glitch in a recording process. That's what the buffer is for. I was going to ask the same question. I was going to ask what you meant by a glitch. A hit on the buffer I would consider a small glitch or hiccup. A "glitch" in the recording process is what I would call a recording failure!

Well, with modern drives, the recording can completely stop and restart without making a coaster..

But the CD-RW "glitches" in this case were confined to "dropping to 8X speed" for very brief intervals..

Mostly I mean DVD playback glitches, mostly due to the soundcard rather than the DVD drive.

Cheers