Enter "Serial ATA"

[DrMordrid]

Jezzz...just when you have a handle on the drive and DV hardware something new shakes things up.

I was first alerted to this by John (shake-em-up) Dvoark's column in PC Mag, then some searching turned up more info in The Register and elsewhere.

It seems the ATA interface will be taking a quantum leap in the next year or two. That is when the next version, Serial ATA, is due to hit. The plan is for it to be used with all internal devices, and perhaps for some external ones too. More on that later.

The major players: Intel, Maxtor, IBM, Seagate, Quantum, APT and Dell. They plan to license it freely to other manufacturers.

The setup: a high speed serial bus; no ribbon cables.

The speed: 150 Mb/sec to start with going to 600 Mb/sec later, all the while maintaining OS, software and hardware compatability with current technologies. They state hardware adapters for legacy drives will be available.

The target: Apple's IEEE-1394, although not overtly. Apple has intended their IEEE-1394 will be the next-big-thing in drive interfaces. Since Apple licenses each IEEE-1394 hardware installation its use requires a royalty payment to Apple. The storage manufacturers don't like the idea of Apple getting a piece of their pies.

Speculation: once SATA hits full force what's to stop it from taking over IEE-1394's place as the interface-of-choice for digital video cams? Nothing.

Serial ATA would be both faster and cheaper. Also no royalty would have to be paid to Apple for each cam or card it's used in.

Such a cam would appear to the system as a standard system storage device, which would certainly simplifiy importing and exporting videos.

The proposed map has Serial ATA hitting in 2002.

The Serial ATA groups site is http://www.serialata.com

Dr. Mordrid



[This message has been edited by DrMordrid (edited 25 March 2000).]

[wkulecz]

Doc,

If it comes in initially at 150Mb/sec = 150/8 = 25 MByte/sec its DOA unless its a whole lot cheaper for the both the drives and motherboards. Looks lamer than USB to me.

Unless each device has its own dedicated channel or you've made a typo, this emperor has no clothes!

The scheme I've head kicked around is a version of gigabit ethernet (with 10 GB on the drawing board) where each device is simply a TCPIP connection! The lack of license fees for embedded Linux is one of the things that got this moving. I forget the name, but one the bigger (techie) names at Silicon Graphics is the author of this scheme.

--wally.

[wkulecz]

DOH!
Stupid me! 150/8 is of course ~18.8 MB/sec!

This is neglecting any framing or protocol overhead so actual thruput has to be a bit less that this. Once you start sharing this bandwidth -- what are these guys thinking?

Cheaper for us to make, and the public is so innumerate they'll never notice! After all 150 is bigger than 66, must be better huh bubba?

--wally.

[jeff b]

Hey, how have you been, Doc?

Don't forget, that's 600 megaBITS per second potential throughput. Current IDE controllers have a potential 528 megabits per second (66 x 8 bits) throughput that nobody's using, anyway. I mean, you only need 270 megabits per second throughput for full, uncompressed NTSC video. At least, that's the broadcast industry's standard for SDI's (Serial Data Interfaces) that they're using.

Personally, I have my doubts about any serial data interface being able to successfully work with a video signal coming in and being recorded on a hard disk on the same single-bit-at-a-time interface until it's up to something a lot higher than 10 times (600mbs) the total datarate of the two contending data streams. Sure, they can foresee resolving the burst rates with clever buffering, but that isn't going to cut it for 20 minute sequential transfers of DV video to a hard disk while both are contending for the same single bit line. Besides the technical problem, there's also the problem of getting all the camcorder manufacturers to do a complete abandonment of IEEE-1394.

Of course, my opinion of Dvorak is that he's just a gossip columnist these days....

[DrMordrid]

My typo was in the use of a small b in the datarate. It should have read mB/sec for megaBYTES/sec.

The actual BITrates they give for Serial ATA are 1.5 Gbits/sec early on and 6.0 Gbits/sec later.

This gives theoretical transfer rates of;

183.1 mBYTES/sec for 1.5 gbits/sec: (1,500,000,000/8)/1024000

732.4 mBYTES/sec for 6.0 gbits/sec: (6,000,000,000/8)/1024000

The 150 & 600 figures are what were listed in Dvoraks column and are likely his estimate of what the net rate would be after allowing for overhead.

Fast enough?

As far as dualstreaming goes, the best way to do that now is for each video stream to be on a separate drive and, if ATA drives are used, for them not to be on the same cable in a master/slave configuration.

That situation won't exist with Serial ATA as it will use point-to-point connectivity. Each drive will have it's own controller port with NO master/slave status. With this setup each drive can feed its stream at full tilt and in parallel.

An abandonment of IEEE-1394 would only take one thing: a fast enough interface that would not require paying a royalty to use. Serial ATA would certainly provide that. We'll see.

Dr. Mordrid


[This message has been edited by DrMordrid (edited 26 March 2000).]

[jeff b]

Regarding the 1.5 gigabit per second stuff, I work at a company that makes 1.5gbs video interfaces (among other things) for running video over fiberoptic cables. When I asked our engineer (who designs this stuff) what he thought about having two 30mbs compressed video data streams contending for throughtput on one line, his respone was, "I don't want to think about it... Why don't ask me something that won't keep me awake at night?" The problem isn't throughput of a single data stream in one direction, because that's being done all over the place. It's time-slicing the two data streams in a way that they will stay contiguous for the ultimate destination. No matter how you cut it and buffer it, there's still that 'little pause' in there, sooner or later...

[jeff b]

Well, that makes a whole lot more sense, since if you set things up with multiple ports that don't share the same channel(s) for IRQ, DMA, or whatever, then you've got dedicated throughput for as long as you want to hold the line. The license free side of it is also enough to move industry sectors in that direction, indeed!

A little more reading on this seems to reveal the prospect of some level of compatability with IEEE-1394 as used in current camcorders, and maybe even the use of the same plugs (which would certainly be a BIG plus in the prospect of more people going this route when it's available). At the very least, I should expect something that would do a conversion...

The one thing that gives me pause, though, is how LONG it took for NLE to get to the current state that we have with IEEE-1394. If the drive manufacturers were just going to be the only ones that are really hot for this new interface, then it seems to me that there'll be a point where we'd still be using IEEE-1394 for the video in and out, on systems that ran screamer drives using this new interface. Maybe that's the only side that will be in evidence, two years from now...

[DrMordrid]

Another aspect that got my attention is that this port can be added to existing drive designs with just a connector and what they discribe as a "small bit of silicon".

They also predicted adapter "dongles" for legacy devices. I imagine someone will eventually market PCI SATA cards as well.

They went out of their way to emphasize it will be transparent to the OS and software. If it IS that easy to implement then there's no telling where it could end up in rather short order.

Dr. Mordrid

[jeff b]

The thing that I had such a hard time trying to shake out of my old head is the difference between bits marching along serially (one at a time), and bits being lined up in parallel 8 at a time, 16 at a time, 32 at a time, etc. All these years, the drive manufacturers have worked diligently to get those parallel chunks of data marching into their onboard drive buffers as fast as possible. But at the end of the process, the bits are all written onto the disk one at a time...

Now they see the prospect of getting rid of that whole elaborate parallel to serial hardware structure inside the drive, and it must be giving them all some VERY serious wood!

[DrMordrid]

I have to admit I've spend some serious time over the years wondering why they used parallel connections for drives. I guess the main factor was a limited (at the time) ability to push bits over each wire. That has changed drastically with the new error correcting technologies developed in the last few years. Look at what's happened to modems and phone lines.

Then there was Commodore. Remember their systems (PET, SuperPET, C=64 etc.) where they had a single serial bus that handled drives, printers & floppies all on one daisy-chain? Each had it's own device number to keep things straight, but it only operated at 300 kbits/s. I had all of 'em. In fact I STILL have all of 'em.

Talk about dating ones self ;-)

Dr. Mordrid


[This message has been edited by DrMordrid (edited 27 March 2000).]

[jeff b]

My C64 is carefully packed away, underneath my workbench behind me, here in the cellar... I could never possibly part with the first computer I ever owned! I mean, I learned how to program on that thing!

Did you have a CompuServe account back then, too? It was the one place you could get all sorts of programs for the C64, store them on the tape drive, and make your own source code modifications as needed.

Ah... the wonder of it all...

[salacious]

I was reading about the IBM 70GB IDE HDD and it has a ATA-100 interface. Do any motherboards have this interface yet?

Salacious

[DrMordrid]

If any not many.

ATA-100 is supposed to be the last of the parallel ATA controllers and will likely be short lived given SATA appearing in the next year or so.

Yes, it'll be a tad faster than ATA66. Yes the drives will be huge. No, I don't think it's worth it unless you're building a new system from the ground up.

I sure wouldn't go out of my way to upgrade to it.

Dr. Mordrid


[This message has been edited by DrMordrid (edited 28 March 2000).]