Slashdot Mirror
Enterprise-class ATA Drives
dfung writes "This has been mindlessly discussed many times before here, but Western Digital has now introduced real enterprise-class ATA drives with SCSI-like performance specs and 30% lower price. So now you can buy a real 10K rpm ATA drive. Interestingly enough, they mention the reason for the traditional difference in price between ATA and SCSI which I never have seen mentioned here - it has to do with testing costs, not controller electronics|platter quality|etc. Another interesting tidbit is that 160 million ATA drives were sold last year. I saw about 2 million of them stacked up in the aisles at Fry's Electronics yesterday, but that sure is a lot of drives."
> The Raptor also carries a five-year warranty.
The five year warranty is a welcome inclusion. Western Digital is good about replacements.
I have a hard time believing though all my clicking-clacking(WD), and bad block (Maxtor) drives have to due with lack of testing. Testing doesn't help make the drives more reliable. Either SCSI drives have a high test failure rate, or there is more to the story.
10K drives at less than SCSI prices are a welcome addition to the low end market, but I'd only use it where reliability and high performance isn't crucial. IDE drives still don't have their own processor leaving a big advantage to SCSI, right?
-Pete
Soccer Goal Plans
SATA offers all the speed benefits of SCSI (such as command queueing and device initiated data transfer). In addition, it is one drive per channel. But "wait," you say, "servers need lots of drives in raid5 on each channel!!!" One drive per channel is a blessing in disguise.
From time to time I've seen drive logic fail (as opposed to surface errors), which often brings down the entire SCSI channel. With raid5, you can only afford to lose one drive and perhaps a couple hot spares. Certainly not 14 drives in one shot. SCSI is many pinned, and SCSI raid adapters are designed to have many drives on each channel. One drive per interface is extremely costly and impractical. In this respect, SATA is more robust.
"If one drive per channel serial interfaces are so good, why weren't the used in the first place," you might wonder. Modern high clock rate microcontroller technology permits much higher frequency twisted pair serial interfaces that can offer superior bandwidth to older parallel, ribbon cable interfaces. If SCSI were being designed today it would look something like firewire, which I'm sure you're not biased against. Don't be fooled by the ATA moniker.
That large of a cache on a HD is really just more room to quickly dump data without actually having to write it to the hard drive immediately. Helps you get fastest possible data transfer when doing I/O with your drive, and on the scale of bag for your buck it isn't that expensive over their normal 2M cache drive.
Horror & SciFi Erotic Nudes
This is a Serial ATA drive, which the article even mentions (second paragraph: "...Enterprise Serial advanced technology attachment..."), but then proceeds to call it an ATA drive (instead of SATA) for the rest of the article.
Here's a somewhat less misleading article.
Just complile yourself a Linux kernel and do software RAID. Tom's
Hardware Guide had an article on software raid performance on here (this is about Windows 2000, but anyway).
Why does the kernel go through stable and then unstable forks? Can't it always be a stable build, like with Windows?
Maxtor released their 10K drives. Rapture series drives are limited to 36GB I believe (1 plater), have a 5 year warrenty, and rated for 1.3 million operating hours (I think its 1.3 million, might be wrong). These drives are SATA, and are hot-swappable. And you too can own one for about $140-160. Which when you look at the price of SCSI, its VERY cheap.
We were all warned a long time ago that MS products sucked, remember the Magic 8 Ball said, "Outlook not so good"
i believe you are thinking of different things.
chips and tech becomes mature and their FAILURE RATE decreases. mature technology does not cost less to test. On the whole SCSI is still a more complex technology, and I would not be surprised if tested with higher margin / more thoroughly due to the "enterprise level reliability" thing.
besides, as devices gets more complex and more "mature," generally the testing costs increase because you have all these new features, plus the old features, plus the shit that keeps it backwards compatible, to test. you can do better on the profit margin / cost side by making ships that have a lower failure rate, but that does not mean chips gets tested less, or it takes shorter to test them. On the contrary, it usually goes the other way.
Anyhow, example: RAMBUS was expensive because it was a "cutting edge" manufacturing process. the output impedence of the chips had to be controled very precisely, which is difficult to do and a lot of it failed at test - driving up the cost. as process matured, less failed and price came down. but each chip still went through the same routine, and sat the same amount of time on the testers* and took the same number of pin-capacities**, so the TESTING COST stays the same***.
* as memory size increase, they sit longer, usually
** similarly, wider buses takes more pins
*** so in the end testing cost usually increases.
separate the two concepts.
My life in the land of the rising sun.
You also have to consider the extra cost of a SCSI controller. For someone who has ATA onboard, the cost of a Scsi drive + controller will be ~ $140 more than the ATA alternative.
This month's issue pits IBM's best IDE vs. a Seagate Cheetah SCSI.
The Winner? The SCSI drive by a margin of more than 30%. There is still a huge difference, especially in the random seek and file transfer areas.
I bought one when the were 130usd for a 120GB/8MB version, and I like it. If I had to get another one, and the price was more than 10usd difference , depending on capacity, I would get the normal version.
Tagged Queuing Explaination.
However, IBM's working on similar concepts for ATA.
Build it, and they will come^Hplain.
Just wanted to point out that Hitachi (Formerly IBM storage) has yet to release a SATA drive. WD's new drive is ONLY the 2nd SATA drive out there, with the first SATA drive (Seagate Baracuda V) really just a regular ATA drive that is available in SATA version, thus no real hardware improvements cause it was origionally designed with an IDE connection, and needed to still meet the requirments for a normal IDE drive.
We were all warned a long time ago that MS products sucked, remember the Magic 8 Ball said, "Outlook not so good"
Don't be dense, to do 14 (actually 12 is the most I know of) IDE drives you get a controller that has 12 interface channels and a controller chip. The speed per channel is meaningless because with the IDE solution the only bottleneck is the host bus and the speed of the interface chip, no single drive is going to saturate it's single line. Also did you read the article and notice the waranty one these drives? Yeah 5 years just like the SCSI drives. These drives are basically SCSI 10K rpm drives with an ATA controller board, hook em up to the right IDE host controller and you have a solution that will save you 30% on your storage costs, which is substantial for many enterprises.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
"Those who consume the bulk of goods are those who make them. We must never forget this secret of our prosperity."
The Seagate Barracuda IV seems to be about the most reliable ATA drive about at the moment, with the assumption that the Barracuda V will behave similarly. While Maxtor (creeping bad sectors) and WD (crappy bearings) seem to have some common problems which crop up a lot, the 'cuda has been around for a while and I've heard nothing but good about it (aside from some RAID issues, which they fixed in the 'cuda V); from significantly smaller return rates in places which sell various makes in largeish quantities, to the simple lack of "my Seagate is failing!" posts on various forums.
Not very scientific, but certainly compelling evidence
Either way, I think a big issue with drives these days is heat; a lot of cases have the 3.5" bays in a deadzone where heat can build up quickly, and a lot of heat can massively reduce the lifetime of a drive. Either get a case where you can put them right at the bottom, near the air inlets and with lots of space around them for the air to circulate, or actively cool them. I've seen drives mounted in the top bays run at 50c which ran at 30c when mounted at the bottom; where do *you* think it's most likely to achieve and exceed it's 5 year design life?
But if none of that bothers you, there are adapters you can get that let you put a PCMCIA or CF flash card into an ordinary IDE slot.
You're not looking.
www.pricewatch.com shows 181gb scsi drives. There are also 4 x 181gb drives 1 cube away from me for our EMC.
Maxtor may make a 250 gb drive, but you can't use it in your PC. IIRC, ATA133 can only address up to 120gb.
Also, when you're buying scsi, you're not going for single drive density. You're aiming for throwing 10 drives into a RAID 1+0 config (or similar). And finally, those 250gb are new. They're not going to release it for SCSI until they have some experience with it's failure rates, and what not.
Zapman
I for one am giving ATA a shot.
To keep the costs down on a backup server, I spec'd it out with 4 200GB Maxtor drivers hanging off a 3ware RAID controller.
After all is said and done, RAID 5 and formated. I have a 568GB Array for a fraction of the cost of a SCSI equivilent.
This server is not a 99.9999 uptime server, but it needed big storage. If a drive fries...nobody is going to complain if its down while a drive is being replaced and the array gets rebuilt.
Its simple enough to pick up up another low cost IDE drive and slap it in..(By the time one does blow..of course the price point for 200GB drives will be even lower than 2day).
In a real pinch,when adding or replacing a drive, I can buy a drive at the local computer shop... This as opposed to waiting on vendor to send out their 'specific' SCSI drive with their proprietary trays (HP!)
So ya.. I see IDE having a role in the enterprise.
It remains to be seen if the 'enterpise' IDE drives live up to their billing.
Actually -- per some info I found in M$'s knowledge base, it appears that this 5 year old *bug* in FDISK (NOT in WinXP itself), limiting partition size to 32bg (or 64gb in some cases) was left in WinXP's FDISK on purpose.
Per the KB article I found, if you have a FAT32 partition greater than 32gb, XP may experience "data wrapping", which will mangle data. Therefore you should NEVER use FAT32 partitions greater than 32gb with WinXP. (NTFS does *not* have the problem.)
I'm fairly sure I've already seen this happen, with a WinXP partition that was FAT32 and 34gb. At the time I'd thought the HD had failed and RMA'd it, but after finding that KB article, now I'm pretty sure it was "confused filesystem due to this here wrapping bug" not "bad HD".
~REZ~ #43301. Who'd fake being me anyway?
Er, no. I've used both, and I am not lying.
One can cost an order of magnitude less than the other, and still performs reasonably well for small servers.
I'm talking about hardware ATA RAID, not that promise or highpoint software raid. I just tested one hardware ATA RAID-1+0 set - 75MB/sec writes, 81MB/sec reads, 640GB online.
I'll grant you this: the performance of "enterprise class" hardware SCSI raid may exceed this, unless you go fibre channel which will definitely exceed this. You may be more concerned with latency than bandwidth, in which case the faster rotational speeds of SCSI drives may be somewhat better for you (thus the point of the original post). And generally, you can cram more SCSI cards in an enterprise class server than ATA cards in a PC.
However, when looking at $50,000 storage solution vs. a $5,000 storage solution with similar reliability and performance for a particular business problem, which would you choose, if you are aren't spending someone else's money?
The way I understand it if an IDE drive get a request for tracks 1,7,3,5,4 it reads 1, crosses six track to read 7, crosses four trqacks to read 3, crosses two trackes to read 5, the crosses one track to read 4 for a total of 13 tracks crossed.
on the other hand a SCSSI drive gets the same request and reads 1, crosses two to read 3 crosses one to read 4, crosses one track to read 5 and crosses two trackes to read 7 for a total of 6 tracks crossed.
the SCSSI disk using the elevator technique allways win when all other things are equal; and usualy the SCSSI disk is built with other advantages.
Apocalypse Cancelled, Sorry, No Ticket Refunds