APLawrence.com -  Resources for Unix and Linux Systems, Bloggers and the self-employed

External SATA drives

Lacie is now offering an external SATA drive bundled with a PCI card for Windows and Mac. They are hardly the first to do so, but it seems curious to me that Lacie seemingly continues to ignore Linux. This made have made sense before OS X, but how hard can it be to add support Linux when you've already done OS X? OK, I know it isn't necessarily just dead simple, but Linux progress in the whole SATA hard drive area seems to be doing fine. So why does Lacie continue to ignore Linux?

Will SATA drives replace SCSI? Possibly. In theory, at least, multiple SATA disks have higher throughput than multiple SCSI drives because they aren't sharing a bus until they get back to the PCI card that's driving them. There's a discussion of SATA RAID over at Tom's Hardware that covers that. SATA can also be less expensive.

When reading about performance, watch out for Mb vs. MB. It's easy to look at SATA 150 MB/s and compare that to (for example) USB at 480 Mbps and forget that the "b" is bits and the "B" is bytes. SATA drives are fast little puppies.


Got something to add? Send me email.





(OLDER)    <- More Stuff -> (NEWER)    (NEWEST)   

Printer Friendly Version

-> -> External SATA drives

10 comments



Increase ad revenue 50-250% with Ezoic


More Articles by

Find me on Google+

© Tony Lawrence




"Will SATA drives replace SCSI? Possibly."

My opinion, is it is not likely unless SATA acquires more SCSI-like features.

SATA is, like IDE, designed to be cheap, and there are tradeoffs in performance and flexibility to go with reduced cost. Each SATA drive has to have an independent connection to the motherboard or adapter card, which is where the serial part of SATA gets into the picture. There are fewer data lines in the SATA interface, so it costs less to make. However, a lot more work goes into getting data from drive to host and back, since each byte has to be serialized at one end and de-serialized at the other. All of this has to happen in software (or firmware) and thus requires processing cycles to complete.

With SCSI's 8 or 16 bit parallel interface, a lot less hardware level activity is required to move a block of data from drive to host, and all 8 or 16 bits move simultaneously. Not to be overlooked is that one SCSI host adapter can support many devices (up to 15), which means once the initial cost of the HA has been absorbed, the cost of adding more hard disks (as well as DVD drives, tape drives, scanners, etc., none of this stuff currently being supported by SATA) is relatively modest. Contrast this to SATA, where each device must have its own port -- of which there is a very limited number on current motherboards.

Something else to consider is SCSI's ability to support a much longer bus cable (up to 25 meters). This may not seem to be a big deal to PC users, but there are some installations where aggregate cable length can be considerable due to the use of many devices on the bus -- a SCSI juke box setup is one such example, multiple RAID 5 arrays (three disks minimum per array) is another. SATA, as currently implemented, cannot support long cables, as the signaling rate on the bus is extremely high, forcing the use of short cables to avoid errors caused by waveform distortion.

"In theory, at least, multiple SATA disks have higher throughput than multiple SCSI drives because they aren't sharing a bus until they get back to the PCI card that's driving them."

In theory. However, in practice, the speed and independent intelligence of current SCSI technology substantially negate any performance limitations that might be caused by occasional bus contention. SCSI devices require a lot less hand-holding than SATA equivalents and can perform off-line functions without stalling the bus and/or the host. If a SCSI device expects to be busy for a lengthy period of time (e.g., a time-consuming seek on a relatively slow device like a tape) it will relinquish control of the bus and allow other transactions to occur.

From what I've deduced in studying SATA, the non-shared bus arrangement exists more to keep costs down than to accelerate performance. Nowadays, silicon, cables and connectors are cheap to fabricate, so creating a separate port for each drive adds little to overall implementation cost. On the other hand, the sort of programming talent required to write hardware level code of the type required to run a SCSI host adapter or embedded controller on a disk, tape or other SCSI device, is not cheap. So anything that can be done at the hardware level to reduce the amount and complexity of firmware is bound to reduce cost. Not having to execute the complex SCSI protocol sequences required of the shared bus arrangement in itself should cut out a lot of the cost.

In summary, I foresee widespread use of SATA in desktop machines and el cheapo servers, primarily because SATA is less costly than SCSI but performs much better and more reliably than EIDE. However, I don't see SATA displacing SCSI from higher end equipment anytime soon. Just an opinion from an old computer curmudgeon. <Smile>

--BigDumbDinosaur

---December 15, 2004

So what your saying is that if SATA would replace SCSI it would end up being just as about expensive as SCSI? :-P

It would probably be more accurate to say that SATA will provide a cheap replacement for SCSI for people who do not need the reliability or scalability that SCSI offers.

The way I figure it is that SCSI hardware and drives cost more simply because they are better then their ATA counterparts. The drives use better parts, are faster, and use smarter controllers. SCSI could possibly be as cheap as ATA, but it would be a waste.

Putting that aside Linux, SATA, and file serving look like a great mix. You can get most of the nice features like hotswapable-ness and multiple rundundancy, and have long enough cords to use lots of drives... Main limitation seems to me to currently be the PCI bus. 32bits and running at 66mhz gets you around 120MB/s maximum practical speeds. And that has to be shared between all the pci devices... Nowadays with cheap HD's that can do 40MB/s worth of data movement that bus bandwidth can be maxed out without much effort.

Plus there are other limitations because of the shared bus layout. Like a practical maximum of 5 PCI slots unless you use special PCI bridge devices.

However soon we will have PCIe
for a good overview:
http://arstechnica.com/articles/paedia/hardware/pcie.ars/

This is a point-to-point serial communication bus that is backward compatable with PCI devices with a PCI-PCIe bridge.

Each device has it's own communication pathway and you can end up doing cool things like hotplugging cards. Hotplug drive controllers, hotplug sata drives. You can add capacity on your server on the fly without so much as a hiccup.

Also the bandwidth avaible is staggering compared to regular old PCI, allowing such things as Nvidia's new SLI-capable video cards (2 vid cards work together as one big card). So you can end up doing pretty massive things just with software raid setups. You can end up having a dedicated PC just for file serving up to another machine that does the actual brunt of the serving work. Use them just like a NAS.

Then on top of that you can get even more fancy with high-aviablity and load-balancing clusters. Like what google uses. Stuff that wouldn't be practical with very expensive hardware.

--Drag

"So what your saying is that if SATA would replace SCSI it would end up being just as about expensive as SCSI?"

In order for SATA to fully replace SCSI, it would have to perform and behave like SCSI in terms of features and capabilities. Most of the cost in SCSI is in the intelligent controllers and the firmware needed to run them. It is these design aspects of SCSI that make it unique among system bus implementations

A key reason why SCSI performs as it does is its highly engineered bus system (16 bit, balanced to ground in the case of ultra-160/320). Any bus system that incorporates a lot of error-checking and other stability features demands a sophisticated protocol to operate. That takes equally sophisticated programming, ergo more cost.

Although not strictly related to the design of SCSI, your point about SCSI hard drive mechanisms typically being higher performing devices compared to their ATA or SATA equivalents is correct. Most manufacturers of SCSI hard drives perceive their products to be high end and build them accordingly. The mechanism in a SCSI hard drive may not necessarily be faster than the ATA/SATA counterpart (although it usually is faster, especially in the latency area) but it is generally more rugged and the service life rating is higher.

So to answer your question, my opinion is that if SATA were to progress to the point where it was comparable in performance and reliability to SCSI, it would cost about as much. Of course, the economies of scale might mitigate that if enough SATA hardware was sold.

--BigDumbDinosaur






Fri Jun 3 22:48:22 2005: 608   anonymous


You may be wrong on one point here - serial vs parallel. According to Seagate they are working on serial attached SCSI (SAS) that performs much like SATA - (link) If you read far enough into the article you'll see where they talk about why serial interfaces are faster than parallel right now - although they do admit that if changes can be made to the way parallel interfaces work then those would again be faster.



Sat Jun 4 04:53:31 2005: 610   BigDumbDinosaur




You may be wrong on one point here - serial vs parallel. According to Seagate they are working on serial attached SCSI (SAS) that performs much like SATA...


I think you've got that a bit backward. The idea behind SATA was to attempt to aproach the transfer rates seen with SCSI, but at a much lower cost. Hence SATA and SAS are two totally different bus systems: different protocols, different signalling methods and different command sets. SAS, as I understand it, uses the same command set as SCSI and follows the same proctocol. The difference is in the physical bus design. Incidentally, a form of SAS already exists in fibre-channel SCSI, whose speed has yet to be duplicated by anything else to date.


SATA is a high speed, serialized form of the ATA bus that has been in PC hardware since the mid-1980's. SATA is a non-arbitrating bus, which means that each device must be driven from a separate port. With SATA, as with parallel ATA, much of the work gets done by the MPU. This is unlike SCSI, which is an arbitrating bus where all functions are controlled by the host adapter and target device's controller. The MPU is not involved in bus transactions at all.


SAS, which has been discussed for several years, is still mostly a gleam in some engineers' eyes. There has been progress, but I don't see any hardware shipping as yet, and I'm not even sure that the work will be worth the result. In any case, serial vs. parallel dates back many decades, and there are plenty of people to support both sides of the argument.


Probably what Seagate wants to do is try to cut out the expense of providing parallel data lines, rather than exploit new ways to boost performance. Let's face it: a serial cable with less than a dozen leads is a lot cheaper to make than one with 68, as found with wide or LVD SCSI. Ditto for the connectors, line driver silicon, etc. In other words, the trade-off becomes one of less work making the parts and more work making it run fast. Of course, cost gets added back in accommodating the extremely high switching speeds that would occur in an SAS bus trying to deliver bytes as fast as parallel SCSI.


SAS would try to capitalize on the robustness and stability of the SCSI command set, as well as the intelligent nature of SCSI, yet use a less costly hardware interface. The question, of course, is will it perform at the level that we see today with LVD-360?




Mon Jun 20 20:03:37 2005: 673   anonymous


Ah, the fastest i've seen 2Gb FC go is ~185MB/sec i've seen Ultra-320 running at well over 250MB/sec. Ultra-320 is still the fastest per connector of any drive interface out there. Ultra-640 might be faster if anyone ever makes a device for it.

SAS is like pci-express, multiple point to point links on the same cable. The point to point rate for a single port is lower than Ultra-320. That seems sort of pointless, might just as well run SATA if the reliability ever gets good enough.







Sun Dec 18 01:22:24 2005: 1446   anonymous


After about a week of troubleshooting an SATA PCI Hardware RAID Controller on my QuickSilver G4, I switched my Mac back to Panther.



THE BEACH BALL OF DEATH:

With a ACARD SATA 6890M PCI RAID and two 250GB SATA Drives in either Mirror, Striped or Normal modes installed and during a long file copy (200MB - 2GB), Tiger and the SATA would hold up all applications. The copy would eventually finish and the system would return to normal, but the experience was painful holding up every 200MBs for about 30 seconds. Safari, Finder, and Photoshop would become non functional even while no CPU Activity was taking place. The Beach ball of Death! I tried 4 different Mac OS X Drivers for the SATA Controller and swapped all 16 possible combinations of the 4 PCI Slots. Removed old legacy Hardware, etc. Did 3 firmware updates to the SATA Controller. Still no dice. Man, I felt like I was troubleshooting a PC running Mac OS 9 (impossible, it's just an analogy). This type of troubleshooting will become more common when Mac's switch to Intel with more unsupported hardware options.



BLACK CAT SAVES THE DAY:

A friend of mine was having trouble installing Panther on his Graphite G4. I told him that I had an extra 80 GB ATA Drive, and I would give it to him with his Panther discs pre-installed. While I had his system running on my QuickSilver, I noticed better response time with my SATA Drives and less latency holdups! Only the Finder would experience some latency. All the other Apps would still function was they should on a BSD UNIX Based Multitasking/Multiprocessor system. Next, I installed a fresh copy of Panther on my own SATA RAID and updated to the latest ACARD 1.5.6 drivers, and WOW it worked even better. 1000 times better than with Tiger. I then tried a Fresh copy of Tiger on a separate partition and tried upgrading using the same 1.5.6 drivere The Latency came right back. And it didn't matter if I used the default Mac OS X Drivers or ACARD latest incarnation. Looks like the ACARD SATA PCI Card is not currently compatible with Tiger 10.4.3 on a QuickSilver Dual Ghz G4 with 1.5GB RAM.



THE BATTLE IS OVER, BUT THE SAGA CONTINUES:

Oh well, I am glad I got this hardware working and plus its bootable with Panther. (BTW, Tiger would never boot from the the SATA RAID, only PANTHER would boot correctly.)



I have a 500 GB Partition for my System, Apps, Music and some Video. I have another 300 GB Partition running on Apple's internal ATA. Both Partitions are stripped over 2 drives each; one via ACARD's SATA PCI Hardware, the other is Apple's Software RAID from the Motherboard's internal ATA. But the speed is awesome. Feels like I'm on a G5. It's nice having 800 GB of internal storage space.



The 16x DVDRW is working flawlessly too. Still burning lots of weekly podcasts with it for the morning commute.



Happy Holidays,



Goodtime

[email protected]



Thu Dec 29 19:44:10 2005: 1466   anonymous


Thanks for you thoughtful blogging on the Pather/Tiger issue with SATA and ACARD drivers. I was pulling my hair out about this trying to start the New Year with all 10.4.3 on my audio workstation.

Nothing doing. Both 120 GB Maxtor drives were flaky on my Dual 1 GHz G4 with ACARD 6880M 1.5.6 and OS 10.4.3 - luckily I had my old Panther disks around. I reformatted everything and all seems well with 10.3.9 again.

Thanks for the tips....

-John in Colorado



Thu Mar 16 18:31:34 2006: 1795   anonymous


You are very welcome.

If you have any more tips, please go to:

(link)

Thank you,

gt



Sun Feb 11 19:58:59 2007: 2860   anonymous


goodtimepcmac blog has moved to:
(link)






------------------------
Kerio Samepage


Have you tried Searching this site?

Unix/Linux/Mac OS X support by phone, email or on-site: Support Rates

This is a Unix/Linux resource website. It contains technical articles about Unix, Linux and general computing related subjects, opinion, news, help files, how-to's, tutorials and more.

Contact us





We're terrible animals. I think that the Earth's immune system is trying to get rid of us, as well it should. (Kurt Vonnegut)





This post tagged: