Slashdot Mirror
Serial ATA, Here and Now
Xev writes "We have heard a lot about this new technology; over at HEXUS.net they have a review of a retail drive. The first on the internet, it is interesting to see the performance of the unit as well as the hotswap feature, and other new functions. Is this a solution to cheaper hot swap?"
is serial ATA drives that will swap into My portable music player, My PDA and My desktop and laptop computers. Or, to be more exact, I want all of thos different pieces of hardware to HAVE serial ATA functionality... THEN I will be content.
I'm currently building a computer for a good friend of mine, and we had planned on building in a Serial ATA RAID for fault tolerance.
Yeah, well... We had all the parts weeks ago... all except the !@X&@! serial ATA drives. Nobody had 'em, and nobody could get 'em. We also couldn't find Serial ATA mobile racks to mount the RAID drives... apparently nobody has those either.
We ended up having to use standard Parallel ATA drives (spare me the "SCSI R0XX0R5!!" flames... this is RAID on the semi-cheap, and it's not a server).
Ah well, nice to see that Somebody can finally lay their hands on these.
Even if a man chops off your hand with a sword, you still have two nice, sharp bones to stick in his eyes.
Why the new connector? After all, it does plug, ultimately, in a standard ATX power supply. And they even provide a conversion cable ( == less reliable).
Sigged!
Obviously you don't read many hardware reviews.
Welcome to the world of hardware reviews.
As you'll start to notice, 99% of reviews are spread over billions of pages and generally consist of cut-and-paste descriptions.
Any actual analysis is so horrible and incomplete as to make the review worthless. Very fiew hardware review sites do an actual good job of doing hardware reviews. Storagereview I view as the best. Too bad they only do hard drives!
For other stuff, skip right to the Arstechnica hardware forums. Best place there is for mildly biased hardware discussion.
As I see this technology, it's the poor man's Fibre-channel. Seriously, it improves on (parallel) ATA the same way as Fibre-channel improves on SCSI: it increases throughput, it increases scalability and it drasticly increases flexibility and manageability.
Well, to be completely honest, Fibre-channel is much more sophisticated than SATA, having actually a (double) ring topology, storage-area networking (you have Fibre-channel hubs and switches), support for up to 10.000 m distances between devices and up to 400 MB/s (that's megabytes per second), diverse physical-layer techs (copper included) etc. etc. But you pay for it dearly, of course.
Sigged!
Have you looked at why Fibre Channel is more expensive? Well, there are a few reasons. FC is designed to be used in HUGE, HUGE datacenter environments. An FC switched fabric can have 16 million devices on it, all speaking SCSI and/or IP. You can run your massive disk array in its own room up to 5km away, thanks to single-mode fiber optics, which are a part of the spec. FC itself isn't too impressive until you look at the bigger picture for it. Another thing it has going for it (like scsi except on a larger scale, unlike Serial ATA unless I'm mistaken) is that in a switched fabric, any device can talk to any other device. For example, my workstation could go over and read a disk that was currently being accessed by 6 other devices from around the datacenter, and it would just put my request in the command queue and respond to it in turn. (Writing is the same way, but you must use a file system designed for multiple writers unless you like FS corruption.)
Another thing, it's currently faster than SCSI. As far as I remember, 4Gbit FC is out, and 8 (or maybe 6) Gbit is on the way. You don't even need all of your devices to talk that fast (most drives are 1Gbit, maybe 2Gbit), as long as your switch can handle the speed differences, as it should be able to. Even the 4Gbit is faster than Ultra/320 SCSI, and the 6 or 8Gbit will kick the pants off of it.
Also, unlike SCSI, the cabling requirements are EASY, and the interface cards are inexpensive. I built a JBOD (just a bunch of disks) out of ST39102FC drives (9.1gb 1" 10000rpm Seagate), and you know what I used to cable it together? Cat5. Standard category-5 ethernet cable, at less than 10 cents a foot. Of course, that doesn't include the power cabling, but that's all standardized anyways. Interfacing to an FC drive? Nothing. I grabbed a copy of the drive's tech manual from Seagate's web site, which had all of the pinouts for the FC connector, the SCA40. I whipped up a board in ExpressPCB (because it's easy to order boards from them and the software's free), ordered the mating SCA40 connector from Mouser Electronics, and soldered it all together. It didn't even require any passive components besides a couple of status LEDs (which are optional of course) - just the connectors. Total cost per drive? $10 for the interface card, give or take a dollar, and the cabling is negligable. Buying the connectors and PCBs in bulk will cut down even farther on the cost (mainly because the SCA40 connectors are $6.50 or so in singles). You can have up to 120 drives on a non-switched loop. I'd love to see a SCSI card do that, not to mention the cabling associated! The FC HBA was a surplus HP part which cost me a whole $25 from Fleabay. I soldered my own cable to it where the GBIC module would normally go (thanks, IBM, for the manual on your GBICs!) and popped it in. Finding drivers aside (hpaq's website sucks), works perfectly.
Of course, one problem with a simple FC loop like what I built is that if you remove a drive, the loop is broken. That problem can easily be solved - Maxim Semiconductor makes a neat little chip that will do port-bypass for you, and the signal for it's already on the SCA connector. Add the chip, the resistor, and the extra PCB space - you've blown a whole $4 more per interface card. Optionally, you could just hook up everything to a FC hub, which handles the bypass automatically. Maxim even sells a chip that basically allows you to build a 4-port FC hub for a few bucks, and they're daisy-chainable to the whole loop capacity of 120 drives.
Every cloud has a silver lining (except for the mushroom shaped ones, which have a lining of Iridium & Strontium 90)
Right now, I use IDE on most of my machines. Why? Because most of the time I don't need great disk IO, I don't need more than 2-3 drives, and I can usually live with the lack of reliability that is IDE.
That said, SCSI is far better, and is doing now, for reasonable prices, what Serial ATA is only claiming it will be able to do eventually, and a lot more in addition. SCSI drives with comparable specs, right now, don't cost much more than IDE drives. If the push to serial increases the prices, suddenly, SCSI will be the bargain interface, as well as the performance interface, which eliminates the entire IDE/ATA market. In addition, SCSI to IDE adapters would give most users backwards compatibility, which would eliminate that from being a benefit to serial ATA as well.
So, it may soon be time for everyone to make the switch.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
*beats head against wall*
Who gives a crap about the PCI bus speed, or the theoretical maximum throughput of the ATA bus? The drives can't generate more than ~50 MB/s sustained transfer rate anyway. Yup, that's right! ATA-66 is fast enough for every PATA and SATA drive on the market today.
Oh, sure, you'll spike the transfer rate when reading from cache. I've done the numbers before, and it's something like a 0.1 millisecond difference between ATA-66 and ATA-133, since the largest cache is a mere 8 MB.
You are correct about SATA being faster than PCI, but it just doesn't matter. Nor do the future possibilities of SATA-300 or -600. The hardware just isn't fast enough.
And just to cover all the bases, once SATA is integrated into the south bridge chipset it won't be reliant on PCI. In the case of nVidia chipsets (and any Athlon64/Opteron chipset) it would then go over HyperTransport, which is 800 MB/s. I'm not sure what the backplane speed on Intel chips is, but I believe it's faster than PCI.
If I go look at benchmark numbers on StorageReview.com, I see that your conclusions are debatable. Let's compare the SATA Barracuda V against the Seagate Cheetah X15-36LP, which is a 36GB 15k RPM drive with a 3.6ms access time.
Price - The Cheetah is $409 on CDW. So the price comparison is 2 for $819 (plus the $150+ SCSI controller), versus $150 total for the SATA drive. So it's a 6X price multiple, not a 2X price multiple.
Transfer Rate - 160MB/sec is just what the interface is capable of, not what the drive routinely does. And where did you come up with 75MB/sec? The SATA interface is rated at 150MB/sec. In practice, the Cheetah has a read transfer rate between 45.0MB/sec and 60.5MB/sec. The SATA Barracuda is 24.7MB/sec to 43.8MB/sec. So the Cheetah has a 30% to 80% faster raw read transfer rate. Let's see if this performance benefit holds up in other benchmarks.
Real World Benchmarks - The Cheetah scores 422 on the SR High End DriveMark 2002. The SATA Barracuda scores 355. About a 19% improvement. In no test that corresponds to typical workstation usage did the Cheetah score more than 30% over the Barracuda, and the Barracuda actually won some tests, including the ZD Business Disk WinMark 99. BUT! For server usage, in the File Server DriveMark, the Cheetah scored an astounding 285% better than the Barracuda.
Conclusion - SCSI drives are a foregone conclusion for a server, but paying six times as much money for a 30% performance improvement doesn't equate to a "better buck/performance value" when building a desktop or workstation.