Slashdot Mirror
Serial ATA Technology Explained
Mike Parsons writes "Explosive Labs has an interesting article on Serial ATA . Here is a quote: 'In the rapidly moving computer industry, there are rarely the kinds of revolutionary changes like what is about to take place in secondary storage segment. Soon the hard drives and configuration methods that have existed since the origins of the personal computer will change forever. The basic IDE technology has been around for nearly twenty years. When the lifetimes of other computer components like CPUs and video are measured in months, twenty years ago seems like prehistory.'"
Um, no. Gen 1 SATA is 1.5 Gb/s. Firewire is not faster.
"The defense of freedom requires the advance of freedom" - George W Bush
Actually they said SATA 3 (600 MB/sec) will debut in 2007.
I was a bit confused by this article. They talk as if this thing is the Second Coming of Christ, but then they talk about how desktop pcs are just going to keep taking baby steps. Also at the beginning of the article they say that serial seems to be a step back from parallel (ya think?) but it is faster and better and Oh! Look! An elephant!
>You mean the oversized 40-conductor ribbon cables are solved by 68 conductor ones?
:-)
Nope, they're solved by using high-density cable connectors, which IDE still hasn't figured out with 80-pin cables (instead IDE just cheats).
>And 10x more expensive.
Because... why?
Nobody is buying, that's why. Lower speed SCSI drives are still available, but are still expensive because IDE is stuck in everyone's heads as the only storage method for a PC.
>You are the first SCSI fanboy I've ever seen.
You haven't been here very long. Let me be the first to say welcome!
If you could be told what you can see or read, then it follows that you could be told what to say or think - BoC
Since the site is slashdotted, here are further links about Serial ATA:
Cnet
SATA and ISCSI
Intel Dev Paper
Maxtor White Paper
So what does SerialATA offer that a Firewire connector on my hard-drive won't?
150 MB/sec?
"The defense of freedom requires the advance of freedom" - George W Bush
Read the article.
The problem with SCSI is heat and noise, things that work fine in a server room, but are bad for the home user.
I had SCSI for a while in college. It was cool to show off, but having to turn up my TV to hear over the jet engine in my PC was annoying.
A speech...
The battle between serial and parallel communications is neverending. Show me a Serial WAN connection like a DS3, and I can say "Well, since you never send partial bytes, we could strap 8 of these side by side, send one byte at a time with the bits split up over the 8 DS3s in parallel frames, and we get an 8x speed improvement that's usable by a single connection and no additional latency".
Or show me a parallel bus like IDE, and I can say "Look, having all those data lines next to each other causes additional interference we have to account for, and they're bulky, cost more, overly complex, blah blah. If we just put a serial bitstream on a pair of wires, it would be so much simpler that for the same cost we can turn up the bitrate more than enough to make up for the lost parallelism."
It's all the same. Various communication technologies tend to rise and fall, serial replacing parallel replacing serial replacing parallel ad infinitum. In some cases (like PCI busses) parallel just makes a lot more sense, but in a lot of cases (network stuff, storage stuff especially) there's a tradeoff where both are better and worse than the other in different ways. You could just pick one and stick with it and do you incremental improvements, but the occasional switcheroo provides upgrade revenues and more user "wow" factor and buzzwords.
11*43+456^2
>Nobody is buying, that's why. Lower speed SCSI drives are still available, but are still expensive because IDE is stuck in everyone's heads as the only storage method for a PC.
I have to say that all of my computers up until my most recent have been SCSI, and I am glad that I stuck to IDE this time.
Firstly the price to preformance ratio is dismal. Twice the preformance for 10x the price is not worth it.
I would like to say that the price of SCSI drives is not 10x more because nobody wants a SCSI drive, but because they are simply more complicated to manufacture and interface. Attributing cost comparison between two different technologies solely to suply and demand is a very very basic understanding of the way the economy works.
Also, the preformance hit you take going from 7200rpm SCSI to 7200rpm IDE is not noticable at most times, but I suppose i am tolerant because i can wait more than 3 ms seek.
SCSI is loud and hot and expencive, just like all preformance computing componants, and thus will never be a consumer standard.
Yawn.
- The PC parallel port was defined, in a very loose manner, by IBM when the IBM-PC was introduced in the early 1980s. The original PC parallel port was unidirectional and good for little else than driving a printer. While a number of manufacturers offered bi-directional versions of the PC parallel port (including IBM, in the late 1980s, with the IBM-PS/2) no actual standard existed until the release of IEEE-1284, in the mid-1990s, which gives us the modern ECP/EPP option.
- The VGA pin-out dates from the introduction of the IBM-PS/2 in 1988/1989.
- The PC keyboard pin-out and protocol have only been around since the introduction of the IBM-PC, and have undergone at least two revisions, first for the IBM-AT keyboard, and second for the IBM-PS/2.
The reference to IDE having been around for 20 years is a pretty dubious one, unless you count IDE as just a variation on the vernerable IBM-PC expansion bus. (commonly known as the ISA bus) Still, 20 years ago, there was no such thing as IDE, even as a glimmer of a hope in anyone's eye.All data goes through the PCI bus
No it doesnt. Data goes through the PCI bus if the address is not claimed by something else along the way. That means that everything from the southbridge up is not limited by the PCI bus bandwidth. That means that integrated SATA controllers (not available until next year) are only limited by the bandwidth between the northbridge and southbridge.
Ever read the actual throughput specs on a drive?
Drive throughput has been steadily increasing, and it is predicted to pass up PATA within a few years, and that is not counting RAID striping or the 8 MB drive caches. Its always desirable for the bottleneck to be the drive rather than the controller.
"The defense of freedom requires the advance of freedom" - George W Bush
Personally, I could give a rats butt about the speed. I don't want SATA so my drives go faster. I want it so I'm not having to spend twenty minutes doing finger gymnastics everytime I need to do _anything_ in my case.
Is it worth upgrading for? No, probably not. But id damn sure is worth waiting an extra few months for that next machine to save the hassel of those f'ing ribbon cables.
jello.
aka aron.
Bingo, it is exactly what Serial ATA does not have over firewire that makes it so desirable.
Namely the $50 or so price premium. . . .
Need help treating your acne? Come here!
> Your top of the line P4 xeon is, when powered on, a fast 8086
No, but it's backwards compatible. I don't remember 8086 being highly pipelined, having an integrated FPU, and MMX and SSE/2 instructions.
> Your Keyboard will plug into an IBM PC from the early 80s
They had USB in the early 80s?
> VGA is 15+ years old, and SVGA 10+.
Thats close to true, but irrelevant
>Almost all consumer sound cards include sound-blaster 16 emulation
That would be a software emulation - they aren't register compatible.
The common thread? That they're all backwards compatible. As will be PATA.
If I wanted my next computer to be incompatible with this wall of media behind me, I'd buy a mac. (Not an anti-mac troll, just saying)
Yes- but AFAIK command queuing is not implemented in a lot of the 1st generation controllers because it can break backwards compatibility with PATA software. Most vendors went for an easy upgrade path instead. Look for command queuing in the next generation of controllers.
"The defense of freedom requires the advance of freedom" - George W Bush
And this made me wonder... how long will it take until Linux (and the *BSDs) support this new standard? Will it happen after Longhorn's release? Or has it already been done?
Gee, where have you been? The latest chipsets on the PC market skip the PCI bus for a lot of things. The latest offering from SIS has a south bridge with a 1 GB/s max bandwidth, and if it had a Serial ATA controller on it.. Well, I don't think I need to go on.
The red stripe points towards the power supply connector, or marked pin 1 on the mobo. That's SOOO difficult. 8-)
Actually SATA is 1.2Gb/s, but that doesn't change your point.
Slashdot - News for Herds. Stuff that Splatters.
> (less wires = less logic).
If you've ever looked at the circuit diagram for a simple multiplexer, you'd take that back. Really, how complex the is logic depends on what you want to do with the data off of those wires. To really simplify the idea: if you use serial, you've got to have a muxer/demuxer on the end of that line, and if you use parallel you need to have a clock to syncronise communications on those lines.
In any case, the difference in circuit complexity due to parallel vs. serial, even if one required less than the other, would be a few orders of magnitude less than the complexity of the circuits required to manage the disk itself.
-- "Perhaps the truth is less interesting than the facts?" -Amy Weiss, RIAA
Note that Sun, IBM, AMD, nor HP are disk manufacturers. (Well, IBM might still be, my memory is being bad tonight, but I digress...) Some AMD and Intel motherboards are already coming with SATA RAID interfaces. Intel is right behind the technology, as they are a chipset manufacturer. AMD isn't. VIA, a chipset manufacturer is, along with a ton of other manufacturers who are core to desktop/workstation storage. Just because the big power houses that you name aren't on board doesn't mean anything. Most of these places leave their disk interfaces up to someone else. And those companies *are* adopters.
>FYI, most modern bioses recognise USB keyboards and can be accessed from them.
Oh, I know that.
I also know its an option I disable on all my boards, and an option that isn't always the default.
This means that if you only have a USB keyboard and buy/receive a motherboard with USB disabled by default, you just bought a heap of trash.
If you could be told what you can see or read, then it follows that you could be told what to say or think - BoC
Forgive me if I sound a bit naive but wouldn't parallel be faster than serial?
On paper, parellel can be made to be faster than serial. However, in the practical world it is very difficult to make a high bandwidth parallel bus. It is even more difficult to run that bus any considerable amount of length. By using a serial bus with embedded clock you only need 2 signal wires. If those signal wires are a differential pair (perferrable Low-Voltage) then you can run them a considerable length at an extremely fast rate. If you have a parallel bus, you don't have the option of embedding the clock. If you are running with out an embedded clock you must send a clock syhncronous with the data. Now you have to deal with skew issues between each individual channel as well as all channels relative to the clock. Not to mention other aspects such as crosstalk between the channels. If all you have to worry about is two signals (which if are differential can be considered 1) then many of those issues ago away.
There's a lot of physical behind why its extremely difficult to run long lengths of parallel lines. Yes a parallel bus is faster, but it is almost impossible to implement a reliable parellel bus running at 1.5Gb/s, through cables, and with connectors. Take a look at a bus like hypertransport. It can be up to 16bit wide and run 1.6Gb/s. However, it is a point to point protocol, is run over a control impendance, and is run over very short distances.
I hope that helps.
Well... Seagate has announced their first SATA drives, the Cuda ATA V, but they haven't hit the shelves yet as far as I can tell. It is supposed to be available this fall... so they should be out soon. Also, the Cuda V is a true SATA drive, not just an IDE drive with a bridge slapped on it.
"Ironically, SCSI stands for Small Computer Standard Interface, but SCSI is most frequently found in Servers (large, not small computers), in large RAID arrays. And more ironically, the SCSI drives usually used in RAID (Redundant Array of Inexpensive Disks) usually are not that inexpensive at 3x the cost of IDE. "
You have to look at when these terms where invented.
The following definitions applied back then:
Small Computer = a computer that did not require an entire building to house it.
Inexpensive = cheaper than solid state.
IDE also becomes outdated very quickly.
Try plugging your 7200RPM 120Gb IDE drive into a 386 era IDE controller and see what sort of performance you get. You'll probably only be able to access 8Gb of its capacity also.
IDE hasn't "just been there" it has been constantly evolving.
400Mb/sec vs 150MB/sec
Pay attention to case, it does matter. As for what's in development, call me when its actually available.
Matt
Heh. Here's a list of IDE's shortcomings SCSI makes worse:
The bandwidth per drive thing is one of the great things that SATA brings to the table. With a modern large SCSI setup it seems like you have a lot of bandwidth but on a per drive basis you really don't. 160MB/s divided by 12 drives = 13MB/s (1980's speed). To contrast that look at a 12 drive 3Ware SATA controller. That has a full 150 MBytes/Sec to each of the 12 drives.
To see the usefulness of this take the example of a 12 drive RAID 5 array doing a rebuild while the server is trying to read from the drives. The controller has at it's disposal 1800 MB/s worth of bandwidth that it can use. It can run those drives as fast as they can go keeping the write buffer full on the drive it's rebuilding and using the leftover bandwidth to service the server's requests. Modern ATA drives can read at up to 56 MB/s. With 12 drives you get a total of 672 MB/s throughput which is far more than even the new Ultra320 SCSI is capable of. With newer faster drives and 16 drive RAID controllers this problem gets even worse.
> If more people used it, it would be a cheaper solution
SCSI is quite widely used. There is a lot more SCSI out there than SATA and yet a motherboard with a SATA raid controller costs about the same as one without it whereas a motherboard with a SCSI raid controller on it costs about 3 times as much. SCSI is simply an expensive, complicated technology to implement.
>15k rpm scsi drives get seek times in the low three range--that's three times faster than your average 5400 rpm ide hdd.
The low seek times are a result of expensive server class drive technology, not of the interface. Seagate could just as easily drop a SATA interface on those 15K Cheetah drives and I suspect in the near future they will because:
All of SCSI's shortcomings are fixed by Serial ATA
Yea, I know, it's a cheap shot but really SATA is poised to replace SCSI in most of the markets SCSI still occupies. SCSI was mostly popular in server systems because of it's hot swapablility and plug and play operation (no jumpers to set on 80 pin sca drives.) These are advantages that Serial ATA shares. Motherboard integrated SATA RAID will take over for SCSI RAID in server class systems because of cost, size, power and bandwidth issues. 8 - 16 drive SATA RAID arrays will take over the low to mid-size storage array market. (If you can count 4.8 Terabytes as mid size.) Fiber channel will be left for SANs and large storage arrays. SCSI will be relegated to connecting external drive systems but I imagine fiber channel will eventually take most of that market.
People who like SCSI will probably like SATA even more. It will be faster, much cheaper, more reliable, more compatible, and easier to maintain and troubleshoot. True, you won't be able to run a printer or scanner off it but I doubt there will be a lot of people missing that particular piece of SCSI functionality.
set softtabstop=4 shiftwidth=4 expandtab nocp worlddomination
The IDE I/O register mapping/interface dates to the ST-506 interface (early '80s), defined by Shugart Technologies.
The (16-bit) IDE physical interface is an extension of the AT bus (1984).
General Relativity: Space-time tells matter where to go; Matter tells space-time what shape to be.
It, um, reads less like a press release than does the Explosive Labs piece :-).
ACARD makes a series of SCSI-IDE bridge cards which connect to the SCSI chain on one side and an ATA hard disk on the other. They have several models, mostly depending on what type of SCSI cabling you have, costing from $50 to $80. They support large ATA disks, the cost of which plus the $70 for the bridge is still cheaper than most SCSI drives. If you don't need the warranty and physical traits of SCSI hard disks, but you want to be able to hook up 6 drives to your PC with only 1 IRQ and IO address or add 60-80 gigs of space for under $200, this might do the trick. They also come in handy for old workstation-era machines, like PowerMacs, SparcStations, or VAXes. The bridge doesn't require any drivers or software to work, since it just tunnels ATAPI and makes the IDE drive look like just another SCSI disk in the chain.
http://www.acard.com/eng/product/scside.html
Microland sells them in the US:
http://www.microlandusa.com/microland/
Some downsides:
- The hard disk has to be formatted while cabled to the SCSI-IDE brige. You can't move a drive from a regular IDE controller to the SCSI-IDE bridge without getting geometry errors.
- The interface is ATAPI only, so not all commands for the device may work. FE, firmware updaters and vendor utilities designed for the hard disk probably won't work the bridge.
- The utility to update the bridge's firmware is only for DOS/Windows.
There will probably be LVD-SATA bridges too in the future, if SATA truly catches on.
Democracy. Whiskey. Sexy. Pick any two.
LRC, the best-read libertarian site on the web
- Granted, those cables are annoying. But really, how many times have you felt the need for a cable much longer than 40cm? People with full-sized cases may benefit, but then the author says that the current trend is "small footprint machines". So, why do I need a cable that is bigger than my server?
Quite a lot of people have been bitten by the IDE cable length restriction. The only way to use the top drive bay in a full tower case is to use an out of spec case, or use an PCI IDE controller in the closest PCI slot. Besides the SATA cables are much smaller, they are also much more flexible so they are useful for lots of reasons: eg:
a) Route the cables out of the way, no longer do you have to remove drive cables to access poorly positioned RAM, etc
b) Smaller thiner cable, make for better ventalation which is desirable in both cutting edge gamer boxes and in mini PCs.
c) You no longer have to worry about multiple drives on the one cable, each cable is independent, useful for either neatness, or for cable patterns in show boxes (gamer, with neons, etc).
-Yes, they are! But you'll require at least twice as many S-ATA connectors, as only one device is supported... In the end, the real state on the mobo is going to be similar.
RTFA! I mean they had pretty pictures and everything. 2 SATA = 25% area of one PATA connector. Obviously a plus for either miniturising the motherboard, or for making room for more features (Dual ethernet).
-C'mon... This guy must be joking. I couldn't believe my eyes when I read it! One device per controller is an *advantage*???? Why??? I wish I could add more devices (like SCSI and Firewire) to my curreny P-ATA technology. And then he says ONE is good for me? Don't think so...
I have not yet seen a single controller where you end up with less useable drives than a traditional IDE setup (2 channels, 4 drives). And this is an advantage. Why? A Master/Slave arrangement is horribly inefficient, it causes not only bandwidth sharing but also latency issues. Not a problem on a 4 channel SATA solution, and with no useable loss.
-What kind of RAID? RAID 5 is slow in writes due to the computational power needed to calculate the XOR. Adding bandwidth won't help. And I can't see why or how only RAIDs will benefit from higher throughput.
Depends have the RAID controller is implemented, it can be either integrated so that it communicates directly with the south bridge (SATA no advantage), or it can appear as a logical drive over a normal IDE channel (advantage). However the former is the most common method.
As for your comments on RAID, don't confuse a RAID methodology (1,2,3,4,5,50,etc) with the number of drives. You can have 20 disk RAID 1 sets if you like , which would definately overload a single SATA channel iff using the latter method in the above paragraph. RAID 5 write speed is not _slow_, It would only be a problem, if the drives were being continuously written to, in which case RAID 3 is a better idea. Burst writes are handled by both the controller's and the drive's cache.
RAID 5 controllers have their own dedicated CPU's (commonly i960s) to do the relevant calculations. The speed penalty is very tiny, and the difference from a single non-RAID drives performance is slight. Remember the controllers are spec'd to be able to reconstruck data on the fly, and practically all drives read faster then they write.
-Granted. It may be faster than P-ATA. But what about established technologies like SCSI and Firewire? I *think* (not sure) Firewire can go much faster than S-ATA in its initial version.
SATA is really IDE over SATA, IDE is emulated on a different faster data link. There is in fact a SCSI over SATA standard proposed as a cheaper alternative to Fibre Channel (Another SCSI emulated over a serial data link standard). SCSI theoretically already has a faster standard (Ultra 160 and Ultra 320) but these are buses, that is all 7 (or 14 for wide) devices must share that bandwidth, wheras SATA is a peer to peer technology, one device only. The difference will only be usefull with integrated RAID controllers and/or when you have the money to buy enough drives to saturate a PCI-X bus.
Firewire (1st gen) has 400Mb/sec, SATA 1.5Gb/sec
Firewire (2nd gen) has 800Mb/sec, SATA (II) 3.0Gb/sec (Faster than the current AMD platforms FSB at 2.4Gb/sec)
Road map also includes SATA (II) at 6.0Gb/sec
-I'm disappointed...
You need to do research...
1. You can't HOTSWAP an IDE drive without risking blowing your drive, crontroller, or upsetting the powersupply.
With SATA you can.
2. You can't WARMSWAP an IDE drive, without risking blowing your drive, controller, or upsetting your powersupply.
With SATA you can.
3. IDE still only supports 2, yes 2 drivers per controller, which makes it impossible to do hardware RAID-5. That leaves us with software RAID-5 as our only option.
Who cares when you can get hardware RAID-controllers with 12 ports on one card? What is the great advantage of having the cable be the single point of failure for your whole RAID, like SCSI does?
4. IDE cables can only stretch so far, so even if you could somehow manage to get 8 IDE controllers into a box, for a total of 16 drives, there would still be cable length issues. I think 1 m is max. We need differential IDE :)
Ok, 1m can be a problem for some people. However most people do not have cases larger than 1m.
5. IDE drives are just now able to verify data integrity, but thats good since we can start using IDE drives in servers that don't need 100% uptime.
Err, why is it a problem when it is already fixed as you say?
6. ATA/100 Round IDE cables are already available. In fact I just ordered some that have a UV reflective coating for my next case mod which features a black light. Airflow isn't a big issue, in fact Compaq has been slicing up IDE cables for a long time now to increase airflow.
Round IDE-cables are expensive to produce and still large and inflexible. SATA solves it.
7. The SUSTAINED TRANSFER WRITE RATE of IDE drives is still not fast enough to store uncompressed NTSC video at 60 frames per second, or store high bandwidth Satellite streams.
So get the hardware RAID-controller and start streaming away. Oh wait, hardware RAID for SATA doesn't exist. 3ware is a figment of my imagination.
8a. Size increase (GB's) are not keeping pace with read/write access speeds and simply adding cache RAM and tweaking seek algorithms isn't going to remedy this problem.
You can't blame the interface for that. 150MB/s per drive for 12 drives on one card is way more than any SCSI solution supports -- and way more than current drives need.
8b. As, internal volatile write caches grow larger, the risk of uncommitted writes being lost in a power outage or crash increases.
So turn off the write cache. ATA supports Transaction Command Queueing although not all drives support it yet. By the time SATA drives become available, TCQ should be common.
Finally! A year of moderation! Ready for 2019?
There is a certain elegance to such things as SCA backplanes which ATA (in any incarnation) will never achieve.
Which is another point:
They botched SATA by not including power on the bus, having no standardized connector locations, and (at least so far) having no facility for connection of external devices.
The 1-device-per-wire rule of SATA is another detriment: Sure, the cables are fairly small, but can you imagine the rats nest that would be a 12-device SATA system?
SCSI daisychaining is an easy fix for that, and now-common LVD SCSI is quite able to support this number of devices with a single ribbon. And LVD, by its differential nature, is quite resistant to the electrical problems introduced by slice-and-jacket cable rounding techniques that are all the rage these days..
Oh. And you've got your math wrong, Son:
It doesn't matter if 3ware makes a controller with 12 150MBps ports. The 12-port Escalade 8500 you speak of has a 64-bit 33MHz PCI interface, topping out at no more than a paltry 264MBps to be shared by all connected devices.
If you're serious about throughput, try something like this: Two 320MBps channels on a 64-bit, 133MHz PCI bus. Good for real-world transfer rates in the realm of 640MBps, more than twice that of the 3ware product, while keeping a good portion of the PCI bus free the -other- 30 SCSI devices you've got plugged into it.
And none of this says anything about the benefit of SCSI for the home user:
Just bought a new DVD-R, but don't want to toss your old but dead-solid Plextor? SATA requires you to buy another port. SCSI just requires you to plug it in.
Kid-proof tablet..
It had a different physical connector too, most connectors on the RT was in the same style, kind of like a ribbon cable connector, but with a housing that prevented you from connecting stuff upside down.
The physical interface for RS-232 has changed quite a lot, too. Back in the olden days (terminal nets), 15-pin D-subs were very common, and you NEVER see those today. People would think it was a game port I guess. But you can sometimes see that there's one of those push-out D-sub blanks on the back side of a PC box that is too big for a 9-pin and to small for a 23-pin connector. That's what that's for.
Of course, the RT had the flat, small (8-pin IIRC) connectors for that too. It was pretty handy, you could get a 4-port serial card that didn't need octopus cables or external connector boxes or anything.
"An object declared as type _Bool is large enough to store the values 0 and 1." -- 6.1.2.5, C99 standard.
Throughout the history of ATA, when drive capacities climb towards the addressable limit of the spec, the protocol is kludged by a team of drunken baboons to extend it for another generation
Yes. We should run out of space in the latest incarnation in roughly 50 or 60 years.
Unless, of course, you're expecting to implement a single drive with more than 144,115,188,075,855,872 bytes (that's 128 petabytes or 131072 terabytes) anytime soon.
Yes, previous extentions have been poor. Maxtor got it right for ATAPI-6 which has been adopted by the industry. 48-bit addressing of 512 byte sectors.