What do you think: Will SCSI ever replace IDE on mainstream systems?

[dullard]

Look at the results at Storage Review (click on Database at the top, choose a benchmark, and hit sort). The Western Digital WD2000JB got these rankings in the tests related to mainstream use:

SR Office Drivemark: 9
SR Highend Drivemark: 4
SR Bootup Drivemark: 5
SR Gaming Drivemark: 9
ZD Business Disk Winmark: 1
ZD High End Winmark: 8

Note: There were 18 current SCSI drives in those tests. Thus the IDE drive beat at least half of the SCSI drives in every test related to home use. On one test that IDE drive beat all the SCSI drives. Low end and middle range SCSI drives are not faster on those tests than this IDE drive. Only the very high end SCSI drive can beat this IDE drive on most tests, but still not on all tests.

So if SCSI were mainstream, would the home users be getting the low end or middle range SCSI drives? The answer is: probably. Thus on these typical home uses, they wouldn't benefit in speed. But they will get the added costs, lower disk capacity, higher heat, etc. That is why SCSI isn't mainstream. Capitalism works in most cases, and we as a society choose IDE.

Only if you go really high end SCSI can you say it will win in most tests. Look at places like Dell and see what they charge for high end SCSI. Dell on their Precision Workstation 530 under Small Business charges $399 for a 18 GB 15k rpm SCSI and $599 for 36 GB 15k rpm SCSI. They don't offer larger sized unless you want to go low end SCSI. Yes you can buy it for less and install it yourself (pricewatch gives: $213 for 18 GB, $345 for 36 GB, $665 for 73 GB.), but does the typical home user do that? Still at a minimum of $213 that is more than any component in my home system. I'm not paying $100 more just so my games load 1 second faster.

The prices are higher for a reason: the whole SCSI system costs much more to build. People hit the nail on the head in previous posts why the cost is much more for SCSI. But one thing was not mentioned in posts above, is the added heat means the platters expand more. Thus the platter changes size as it gets used. Thus the data location moves over time, and the drive must carefully compensate for that movement. There is equipment to measure the drive temperature changes and logic to compensate for that. On SCSI drives the drastically added heat means that the drives need much more sophisticated (expensive) equipment to compensate. No matter how many extra SCSI drives sell, the price will never drop to IDE prices with all this extra equipment.

 

[alkemyst]

The problem with these 'tests' is that they are testing in an 'IDE environment' with one request at a time, build up a 'SCSI environment' with multiple requests simultaneously and see where those drives stack up.

However, that said in most people's environments they do one thing at a time a la workstation, not multiple things a la server. Power users (not read as power gamer) have needs for SCSI.

That said, it can be argued once CPU's and Disk Controllers get fast enough, this may not be so important.

I still like SCSI and have ran it almost exclusively for 20 years. I now use a Ultra133 IDE drive as that's all the performance I need, but my removeable storage is all SCSI based since I do use all of those in the background alot.

 

[AtomicDude512]

Originally posted by: Sunner
One good argument for why it wont happen:
Why should it?

ATA-100/133 HD's are: cheap enough, fast enough, big enough, and reliable enough.

SCSI HD's are: expensive, partly due to supply/demand, and other such factors, but they are also more expensive to make, due to supperior motors and such, they are far faster than grandma needs, they have features that grandma doesn't need, but that still cost money, they are too small unless you wanna cough up really big bucks.

Basically ATA is good enough and SCSI is overkill for avarge users, and avarge users decide what the big OEM's make/demand.

Yes, and WDs new Raptor HD that is SATA and 10,000RPM. drool,...

 

[dullard]

Originally posted by: alkemyst
The problem with these 'tests' is that they are testing in an 'IDE environment' with one request at a time, build up a 'SCSI environment' with multiple requests simultaneously and see where those drives stack up.

However, that said in most people's environments they do one thing at a time a la workstation, not multiple things a la server. Power users (not read as power gamer) have needs for SCSI.

That said, it can be argued once CPU's and Disk Controllers get fast enough, this may not be so important.

I still like SCSI and have ran it almost exclusively for 20 years. I now use a Ultra133 IDE drive as that's all the performance I need, but my removeable storage is all SCSI based since I do use all of those in the background alot.

That is why I cleary stated that those tests were for typical home use. I left off the server tests where that drive was #26 and #41. Home use just isn't at all related to server use, so I didn't include them. Typical use is one game open, just Word open, browsing the net without anything else open, etc. Think of your grandma, does she have 10 windows open running 10 programs at the same time? For most people the answer is no. That is why the 'IDE environment' is a good test for home use. Home users only run one thing at a time for the vast majority of their computer use. There is always SCSI for the power users who need it.

High end SCSI is great. It will be the best performance solution for the forseeable future. But the costs are just too great for home use. And since all electronic things improve all the time (Moore's law), most people find it is much cheaper to buy a new IDE drive frequently then to buy one SCSI drive and hang on to it forever.

I have SCSI in a lot of my work machines. Whenever I get another used computer with SCSI I try it for a bit, then put in a new IDE drive. The new IDE drive is very noticibly faster than the 2-3 year old SCSI drives. I haven't yet had the opportunity to try a 15K SCSI drive, but I'm looking forward to it. I'm just saying this to let you know that I have lots of experience with SCSI - but I still feel IDE is better for home use.

 

[dullard]

I actually would argue for the opposite of your proposal, Chaotic42. Instead of pushing for faster drives, I think we as consumers should push for cheaper drives. It would be great to have a $30 IDE drive to put in all new computers, or a great upgrade to older computers. Most drive research is for larger drives, or for faster drives. I'd like them to spend more time researching smaller, slightly slower, but drastically cheaper drives. That would be a boon to consumers. Imagine taking a 3 year old computer with a 8 GB 5400 rpm drive and putting in a $30 20 GB 7200 rpm drive. What a great improvement in speed and storage that computer would get for spending almost nothing. This is the situation that most home users are in.

Yes this cannot happen today. But with some research into developing cheaper parts it could happen in the future. Heck by the time this research is completed, 10K rpm IDE will be standard. So imagine replacing our current 80 GB 7200 rpm IDE drives with 160 GB 10000 rpm drives for $30 in 3 years. That is my dream. Not going to happen though.

 

[vash]

The difference in performance between IDE and SCSI drives do still remain. However, the real key to SCSI taking over IDE would basically boil down to cost. IDE drives are made to be as cheap as possible, their failure rates are much, much higher than SCSI drives because IDE is all about quantity over quality. With IDE's cost being the key factor, we see their drive sizes being quite higher than SCSI counterparts. But when you look at the warranty and MTBF, you see that SCSI is far better than IDE!

SCSI, by itself, will never replace IDE. But the technology that SCSI brings to the table will create "better" IDE drives, but they will still not be "enterprise" ready with their MTBF.

vash

 

[BoberFett]

Originally posted by: dullard
I actually would argue for the opposite of your proposal, Chaotic42. Instead of pushing for faster drives, I think we as consumers should push for cheaper drives. It would be great to have a $30 IDE drive to put in all new computers, or a great upgrade to older computers. Most drive research is for larger drives, or for faster drives. I'd like them to spend more time researching smaller, slightly slower, but drastically cheaper drives. That would be a boon to consumers. Imagine taking a 3 year old computer with a 8 GB 5400 rpm drive and putting in a $30 20 GB 7200 rpm drive. What a great improvement in speed and storage that computer would get for spending almost nothing. This is the situation that most home users are in.

Yes this cannot happen today. But with some research into developing cheaper parts it could happen in the future. Heck by the time this research is completed, 10K rpm IDE will be standard. So imagine replacing our current 80 GB 7200 rpm IDE drives with 160 GB 10000 rpm drives for $30 in 3 years. That is my dream. Not going to happen though.

I'd be more interested in completely new storage technology. Hard drives using any current technology are loud and hot. I remember reading about research into some kind of crystal based mass storage that used laser to read and write data in 3D. Now that sounds cool. Data crystals.

 

[thorin]

The problem with these 'tests' is that they are testing in an 'IDE environment' with one request at a time, build up a 'SCSI environment' with multiple requests simultaneously and see where those drives stack up.

Actually I believe if you go over to SR and check their 10,000 request per second results you'll see that the WD SE drives give the smack down to most SCSI drives. (IIRC)

Edit: Oh my bad that test is CPU utilization @ 10k requests/sec in which it smacks alot of SCSI drives.
Edit2: In fact the Seagate SATA drive leads the pack (for low CPU usage @ 10k requests/sec). (IMHO we can expect the WD Raptor to perform at this level)

Thorin

 

[mechBgon]

Originally posted by: thorin

The problem with these 'tests' is that they are testing in an 'IDE environment' with one request at a time, build up a 'SCSI environment' with multiple requests simultaneously and see where those drives stack up.

Actually I believe if you go over to SR and check their 10,000 request per second results you'll see that the WD SE drives give the smack down to most SCSI drives. (IIRC) Edit: Oh my bad that test is CPU utilization @ 10k requests/sec in which it smacks alot of SCSI drives. Thorin

Meaning, CPU Utilization at 10k request/second? Lower is better, right? The only IDE drive I see in the top echelon is the Serial ATA Barracuda.

edit: oops, I see you spotted that too.

 

[Markfw]

I am going to leave out some of the details, so no flames please. I think the following is generally true of SCSI:
1) Sinc e the SCSI controller is much more advanced, and due to its design, the CPU load is much less in data retrieval, thus faster in total.
2) Since it is designed for higher end dependable systems, they are designed with higher performance, longer MTBF, and cost is not the primary consideration in design.
3) It allows for a much longer distance between controller and drives, allowing more drives inherantly, and external cases can be used.

I personally created an external case for my SCSI DLT drives, since they run so hot, and I don't allways use them, so I can turn them off when not in use. Another nice feature of SCSI.

 

[Ilmater]

Originally posted by: Hagen
The data rate of these in already 150mb/sec. and should double in the next year or so.

Who cares?! IDE drives only get, at best, around 60MB/sec transfer rates. What does it matter if you have a 10-lane highway if you only need to move 6 cars at a time?? Moving to 300MB/sec will be a moot point. SATA is better in terms of power consumption and cable size (especially cable size), but it's otherwise not that big of a deal.

BTW, why hasn't anyone mentioned Western Digital's move to 10k RPM IDE drives on their Raptor series in the near future?

 

[Whitedog]

Originally posted by: Ilmater

Originally posted by: Hagen
The data rate of these in already 150mb/sec. and should double in the next year or so.

Who cares?! IDE drives only get, at best, around 60MB/sec transfer rates. What does it matter if you have a 10-lane highway if you only need to move 6 cars at a time?? Moving to 300MB/sec will be a moot point. SATA is better in terms of power consumption and cable size (especially cable size), but it's otherwise not that big of a deal.

BTW, why hasn't anyone mentioned Western Digital's move to 10k RPM IDE drives on their Raptor series in the near future?

Have you ever heard of RAID0?

 

[BoberFett]

Originally posted by: Whitedog

Originally posted by: Ilmater

Originally posted by: Hagen
The data rate of these in already 150mb/sec. and should double in the next year or so.

Who cares?! IDE drives only get, at best, around 60MB/sec transfer rates. What does it matter if you have a 10-lane highway if you only need to move 6 cars at a time?? Moving to 300MB/sec will be a moot point. SATA is better in terms of power consumption and cable size (especially cable size), but it's otherwise not that big of a deal.

BTW, why hasn't anyone mentioned Western Digital's move to 10k RPM IDE drives on their Raptor series in the near future?

Have you ever heard of RAID0?

This discussion is about why SCSI isn't mainstream. Joe Schmoe isn't going to RAID the drives in his eMachine. Believe it or not, the computer industry doesn't cater to the enthusiast. Just because a handfull of geeks on Anandtech can make use of higher speed drive interfaces isn't going to push manufacturers to develop faster interfaces any sooner.

 

[wasexton]

Ah, yes, Raid0. With all the major manufacturers reducing their warranties from 3 years to 1 year I want to entrust my data to the possibility that any of several drives could fail and I lose all my data. And lets not forget that some manufacturers are now reducing their warranty to 90 days on OEM drives. If it is a consumer system I will take my chances on 1 drive rather than several. A better idea would be mirroring...but we know what that would do to performance on IDE systems.

Just my 2 cents!

 

[thorin]

Originally posted by: Ilmater

BTW, why hasn't anyone mentioned Western Digital's move to 10k RPM IDE drives on their Raptor series in the near future?

Uhhh I did like 10 or 11 posts before you asked that......... Wakey wakey ???? Heh

Ah, yes, Raid0. With all the major manufacturers reducing their warranties from 3 years to 1 year I want to entrust my data to the possibility that any of several drives could fail and I lose all my data. And lets not forget that some manufacturers are now reducing their warranty to 90 days on OEM drives.

Ummmmm they haven't reduced them to 1yr on any drive an enthusiast would care about. All SE (8mb cache) drives are still 3yr and the new Raptor series is actually 5yrs (IIRC). And as has already been pointed out any Joe Schmoe who buys one of these cheaper 1yr warranty drives doesn't exactly RAID0 it.

Thorin

 

[Pariah]

Note: There were 18 current SCSI drives in those tests. Thus the IDE drive beat at least half of the SCSI drives in every test related to home use.

In only 2 out 4 Drivemarks did it beat a SCSI drive that was newer than 18 months old. That's not much of an accompliment. 18 months ago the 2GHz P4 hit the market. 3 months later is was $399 (couldn't find an older price). Today you can buy an XP2100+ for $78 or a P4 2.4 for $145.

On one test that IDE drive beat all the SCSI drives.

No it didn't. ZD disk marks don't count. According to WinMark99 the Maxtor DM D540X a 5400RPM drive is the 16th fastest drive ever released. Faster than multiple 10K SCSI drives and multiple WD JB lines including the WD1200JB. According to Winmark the SATA Barracuda V is 79% faster than the PATA version and faster than the 15k.3, quite an accomplishment for a drive that is slow by ATA standards. There's a reason SR doesn't factor the ZD marks into the final conclusion for any drives.

Low end and middle range SCSI drives are not faster on those tests than this IDE drive. Only the very high end SCSI drive can beat this IDE drive on most tests, but still not on all tests.

There is no low end SCSI unless you count the Seagate Barracuda ES2 that no one uses. Your statement should say old SCSI is not always faster which isn't much of a revelation.

Let's not pretend like there is some changing of the guard here. Out of the 4 workstation Drivemarks, SCSI still held 31 out of the top 40 spots. If you count the 2 server benchmarks (which it swept the top 36 spots) SCSI held 51 out of the top 60 spots. alkemyst is also right in saying that the 4 workstation tests are ideal for ATA drives as they are single tasked. Run just one disk application in the background and run the benchmarks again, and I guarantee you SCSI will dominate across the board.

But one thing was not mentioned in posts above, is the added heat means the platters expand more. Thus the platter changes size as it gets used. Thus the data location moves over time, and the drive must carefully compensate for that movement. There is equipment to measure the drive temperature changes and logic to compensate for that. On SCSI drives the drastically added heat means that the drives need much more sophisticated (expensive) equipment to compensate.

ATA drives do this too. This doesn't increase SCSI costs over ATA. IBM 180GXP was 22.1C while the 15k.3 was 25.1. That temperature difference doesn't require any sort of additional engineering to compensate.

Yes this cannot happen today. But with some research into developing cheaper parts it could happen in the future.

Do you think there has been a lack of research by drive makers in the cost reduction area? If there was a cheaper way to build drives they would be using it. The major way drive costs are driven down today is by reducing platters and read heads. With single platter 80GB drives and single read head 40GB drives, the cost reductions are probably going to slow down in the near term. Once we are down to single head 100GB drives there isn't going to be much room for cost reduction, though ATA makers have already come up with one with the reduced warranty length.

It's unlikely it costs much more to physically build a SCSI drive vs an ATA. The major cost increase comes from the greater required R&D that needs to be recouped and the much longer warranty. It costs significantly more to support a drive for 5 years than it does for 1 or 3 years. These 2 factors will not allow SCSI to ever drop to ATA prices.

 

[dullard]

Originally posted by: BoberFett

I'd be more interested in completely new storage technology. Hard drives using any current technology are loud and hot. I remember reading about research into some kind of crystal based mass storage that used laser to read and write data in 3D. Now that sounds cool. Data crystals.

I know a professor here that works on that. They can easilly write the data in 3D. Unfortunately as far as he knows, no one has ever been able to read the data...

 

[dullard]

Let me start this again by saying yes, the best SCSI is better than the best ATA. Don't forget this when reading this post. But remember my point: the main SCSI benefits aren't what home users need.

Originally posted by: PariahIn only 2 out 4 Drivemarks did it beat a SCSI drive that was newer than 18 months old. That's not much of an accompliment.

I'm just saying that many people think all SCSI drives are faster than all ATA drives. They are flat out wrong. Tons of ATA drives can beat tons of SCSI drives in many benchmarks. I'm just trying to point out the flaw in their logic. SCSI drives themselves are not released on a 3 month schedule like processors, or a 6 month schedule like video cards. 18 months is just about top of the line still when it comes to SCSI drives. Those other SCSI drives can still be bought, and so why should we ignore them? Just going to ATA to SCSI won't necessarilly be faster - we'd have to go from ATA to the best of the SCSI to see the difference.

ZD disk marks don't count.

We can go on and on with benchmarks and just pick some to throw out (example comparing a Mac to a P4 then tossing out all benchmarks but a few Photoshop filters). By throwing them out, suddenly we come up with a result that is no where near reality. The only reason to throw a benchmark out is if it didn't represent use that a HD will see (then it is just a meaningless benchmark - like testing memory bandwidth with different HDs - that would be a test to throw out). ZD disk marks represent some hard drive uses. In those uses, it does a good representation of performance. But to please you, lets throw it out. It still beat at least half of the SCSI drives on all of the other tests I linked.

There is no low end SCSI unless you count the Seagate Barracuda ES2 that no one uses. Your statement should say old SCSI is not always faster which isn't much of a revelation.

By low end I basically meant anything but the 15K rpm drives. Your statement then basically means that everyone who uses SCSI uses the latest and greatest of the 15K rpm drives - and that statement is laughable. I've got multiple SCSI computers in this room, not one of which has a 15K rpm drive.

Run just one disk application in the background and run the benchmarks again, and I guarantee you SCSI will dominate across the board.

Right SCSI is great for that. But the home user rarely ever runs a disk intensive program in the background while doing something else that requires lots of speed. About the only good example where it will occur is if someone scanned for viruses during a game, I've had that happen and yes the game slows a bit, but it didn't affect me in any serious way - and that is on an old 5400 rpm drive (and all I have to do is wait 5 min for the scan to end and get back on with my game). I again will state the obvious: most home users aren't running multiple disk intensive programs at once, so while SCSI does much better here it doesn't mean much to them.

ATA drives do this too. This doesn't increase SCSI costs over ATA.

I said they both have to compensate. But the SCSI drive creates more heat, and thus yes it does need to compensate with more technology (which comes at a cost). Ok I admit I was wrong a bit here (but you didn't say where your numbers came from). At StorageReview, the averge SCSI drive produces a 28.5°C temp gain, the average ATA drive has an 18.3°C temperature gain. Thus the average SCSI drive is 55.7% hotter. 55% is not negligible.

Do you think there has been a lack of research by drive makers in the cost reduction area?

Yes. All the majority of home users need is 10GB drive. And that is being generous. But they keep spending all sorts of money researching 100 GB, 200 GB, etc drives. This R&D drastically adds to the cost of the 10GB drive we need. The excesses in data density comes at a loss of reliability. Thus dump tons more money into R&D to attempt to allieviate this problem. Add more cost to the drive to give excess capacity that most home users never need. I'd rather see R&D used for making platters at 1/10th the cost they already are, etc, instead of making platters 10 times as large as the vast majority of users need.

It's unlikely it costs much more to physically build a SCSI drive vs an ATA. The major cost increase comes from the greater required R&D that needs to be recouped and the much longer warranty.

If as you argue there isn't any significant difference in temp and other factors between ATA and SCSI why then did you argue that SCSI has greater required R&D? You cannot argue both ways. I feel the intense density required to make drives as big as possible is one major culprit to reliability. Thus the excess capacity that most home users don't need has caused the warranty to be so expensive. Lower density drives that are more reliable, and more fit the needs of consumers, will drastically reduce warranty costs.

 

[Pariah]

Tons of ATA drives can beat tons of SCSI drives in many benchmarks.

Only one ATA (2000JB) drive managed to crack the top 10 in all 4 WS benchmarks. 2 drives (1200JB,180GXP) made 2 top 10's, and all 4 of them were 10th place. The DM+9 scored 8th in the only top 10 it achieved. So basically we have one ATA drive that finished better than 8th in any benchmark. Let's not exaggerate here and say tons of ATA drives.

Just going to ATA to SCSI won't necessarilly be faster - we'd have to go from ATA to the best of the SCSI to see the difference.

No, you should say even the best of ATA to the best of SCSI will still yield difference in SCSI's favor. The oldest ATA drive to make any top 10's was the year old 1200JB which was 10th twice. If you have any other ATA drive that isn't current gen you are no where near even 2+ year old SCSI. And let's face it, who is going to buy a drive and then upgrade it 2 months later. Someone who is upgrading from older ATA doesn't have to have the latest SCSI to notice a major improvement.

ZD disk marks represent some hard drive uses.

For all of us that use 5 year old versions of office and Adobe apps. ZD diskmarks have proved bogus in past versions, and this one is starting show signs of the same. Old versions would run out of cache for ATA drives while SCSI wouldn't resulting in odd results where ATA would beat SCSI by a large margin. This version is starting to look like it is doing the same thing. The highest 2MB cache ATA drive (besides 2 5400RPM drives which is also odd) is the 120GXP all the way down in 24th place. Sure looks like another caching bug to me making the benchmark worthless. Edit: the 2000BB is ahead of the IBM, but the bogus looking descrepancy still exists.

Your statement then basically means that everyone who uses SCSI uses the latest and greatest of the 15K rpm drives - and that statement is laughable. I've got multiple SCSI computers in this room, not one of which has a 15K rpm drive.

Anyone buying SCSI today should not be buying 2 year old gear. That's dumb. It should also be noted that 10K drives occupy more top 5 positions than 15k, so no, I wasn't implying that.

I don't know exactly how they were measured, but lets assume they came from a computer that was at 20°C (roughly room temperature).

So basically you're going to form an argument around numbers that you don't understand and decided to not take the time to research. Good strategy. And by "research", I mean read the heading of the chart:

"The difference between the drive's highest operating temperature and the room's ambient temperature."

I'll let the rest of the people on this forum decide if 150% extra is negligible.

I'll let the rest of the people on this forum decide if you know what you are talking about and if 14% == 150%.

Yes. All the majority of home users need is 10GB drive. And that is being generous. But they keep spending all sorts of money researching 100 GB, 200 GB, etc drives. This R&D drastically adds to the cost of the 10GB drive we need. The excesses in data density comes at a loss of reliability. Thus dump tons more money into R&D to attempt to allieviate this problem. Add more cost to the drive to give excess capacity that most home users never need. I'd rather see R&D used for making platters at 1/10th the cost they already are, etc, instead of making platters 10 times as large as the vast majority of users need.

More of the I don't understand how HD's are constructed and how that relates to cost argument.

If as you argue there isn't any significant difference in temp and other factors between ATA and SCSI why then did you argue that SCSI has greater required R&D? You cannot argue both ways.

Not really sure what's so tough to grasp that making a drive spin at 15k requires more R&D than a drive that spins at 7200RPM ignoring the fact that 7200RPM has been around for 10 years and is very mature at this point. At one point 7200RPM was state of the art. After the SCSI drives recouped the R&D costs and the technology is mature and cheaper, it drifted down to ATA, just like 5400RPM before it. Now we're just about to see 10k drift down to ATA, though it still isn't targeted at the home consumer and it will be years before we see them in major OEM computers.

 

[sharkeeper]

Only one ATA (2000JB) drive managed to crack the top 10 in all 4 WS benchmarks. 2 drives (1200JB,180GXP) made 2 top 10's, and all 4 of them were 10th place. The DM+9 scored 8th in the only top 10 it achieved. So basically we have one ATA drive that finished better than 8th in any benchmark. Let's not exaggerate here and say tons of ATA drives.

The WD2000JB is a fast drive. However, I have one in my personal system and it just seems quite slow when the OS (XP) first comes up. I guess I'm spoiled from high end SCSI RAID, but it's ridiculous to have to wait a few seconds after the desktop comes up before you can launch a program. :| That's where SCSI really shines in the IO department. Put any ATA drive there and you will see they far fall short of SCSI drives.

Where I can really see the sluggishness is notebook drives. NEVER convert a FAT32 disk to NTFS because you'll wind up with 512 byte allocation units and this results in a really slow notebook especially when starting up! :Q I'll have to reload and use NTFS from start with 4096 BYTE ALU. Hopefully that will be better. SCSI systems don't seemed bothered by the small ALU's nearly as much however the IO activity is definitely higher with NTFS with 512BALU.

Cheers!

 

[Hazer]

One thing to point out is this: No matter how fast any drive gets, it will always be limited by the PCI bus. 33MHz x 32 bit bus / 8bits per byte = 133 MB/s.

SATA has a few advantages. You can daisy chain them, since they are serial. This means that the PCI does not have to 'take commend' of only one drive at a time.

Physical limitations on platter rotation and density still havent reached the barriers of the PCI bus. Very sad.

What really needs to be done is increase PCI spped. Maybe have it clocked at 66MHz, but have legacy support for 33MHz components. Or better yet, make special paths to system memory (where else does HDD info go?) so that the IDE interface does not get limited by the ATA controller.

 

[Chaotic42]

Originally posted by: Hazer
One thing to point out is this: No matter how fast any drive gets, it will always be limited by the PCI bus. 33MHz x 32 bit bus / 8bits per byte = 133 MB/s.

SATA has a few advantages. You can daisy chain them, since they are serial. This means that the PCI does not have to 'take commend' of only one drive at a time.

Physical limitations on platter rotation and density still havent reached the barriers of the PCI bus. Very sad.

What really needs to be done is increase PCI spped. Maybe have it clocked at 66MHz, but have legacy support for 33MHz components. Or better yet, make special paths to system memory (where else does HDD info go?) so that the IDE interface does not get limited by the ATA controller.

Serial ATA can't be limited by the PCI bus. They're talking about rolling out 1GB/s and faster versions of it. I think that it, along with SCSI on-board controllers, are seperate from the PCI bus.

 

[mechBgon]

Originally posted by: Hazer
One thing to point out is this: No matter how fast any drive gets, it will always be limited by the PCI bus. 33MHz x 32 bit bus / 8bits per byte = 133 MB/s. SATA has a few advantages. You can daisy chain them, since they are serial. This means that the PCI does not have to 'take commend' of only one drive at a time. Physical limitations on platter rotation and density still havent reached the barriers of the PCI bus. Very sad. What really needs to be done is increase PCI spped. Maybe have it clocked at 66MHz, but have legacy support for 33MHz components. Or better yet, make special paths to system memory (where else does HDD info go?) so that the IDE interface does not get limited by the ATA controller.

To be precise, SATA II will be daisy-chainable. Present SATA is a one-device-per-cable setup.

I was working after-hours today on an IDE-equipped AthlonXP 1800+ system at the office, re-installing some software and installing some Windows patches. After working all day with my SCSI-equipped AthlonXP 1700+ system, I can say it is very different to watch the IDE drive ineffectually thrashing along where the SCSI drive would be bZZZT - DONE! Granted, it was only equipped with a garden-variety 40Gb Maxtor 740DX 7200rpm drive, versus my X15-36LP, but it does make me appreciate what I've invested in.

 

[thorin]

Originally posted by: Hazer
One thing to point out is this: No matter how fast any drive gets, it will always be limited by the PCI bus. 33MHz x 32 bit bus / 8bits per byte = 133 MB/s.

Actually future SATA controllers and SATA Enabled chipsets (ie: those coming on the next round of chipsets) will have the SATA link seperate from the PCI bus for that exact reason. The SATA standard's bandwidth requirement is beyond what the PCI bus can provide (Generation 1 is 150MB/Sec vs PCI in it's current consumer form which is 133MB/Sec , not to mention Gen 2 @ 300MB/Sec or Gen 3 @ 600MB/Sec). This has nothing to do with the old ATA problem of only one drive being able to 'talk' (for lack of a better word) at a time (and that has/had nothing to do with the PCI bus).

SATA has a few advantages. You can daisy chain them,

Actually that's in SATA II and it isn't chaining, you can put them on a hub in an extension (subset) of their point-to-point (star) topology (ie: a hub has one SATA II connection coming in that all the device connected to the hub share ..... which actually leaves lots of bandwidth for earch drive given that SATA II has 300MB/Sec available bandwidth).

since they are serial.

Uh you can implement chaining and hubing on parallel interfaces as well (though it would be extremely complicated but definately not impossible by any means).

This means that the PCI does not have to 'take commend' of only one drive at a time.

I have no idea what you're trying to say with this statement but my guess and answer are above.

What really needs to be done is increase PCI speed. Maybe have it clocked at 66MHz, but have legacy support for 33MHz components. Or better yet, make special paths to system memory (where else does HDD info go?) so that the IDE interface does not get limited by the ATA controller.

Have you been away or something ..... a replacement for the aging PCI bus (and AGP) has been in the works for years, and has been getting tonnes of press for at least the last full year. Check out PCI-X and PCI-Express . (Not to mention that there ARE current PCI standards that implement a 66MHz clock and a 64bit bus width).

Thorin

 

[sharkeeper]

Have you been away or something ..... a replacement for the aging PCI bus (and AGP) has been in the works for years, and has been getting tonnes of press for at least the last full year. Check out PCI-X and PCI-Express . (Not to mention that there ARE current PCI standards that implement a 66MHz clock and a 64bit bus width).

What is all this bandwidth going to be used for? It will be quite a while before ANY HDD will have STR challenging even PCI 32/33. Where the bandwidth is needed (and used) is the SCSI domain. I cannot see using 4 or more flakey (READ NON SCSI) disks in RAID 0 or even RAID 5. There are a few ATA controllers (3Ware comes to mind) that are actually quite good, but ATA disks have too many inconsistencies between samples for critical ops IME. They just don't make the cut for our operation. I guess we're extremely demanding but the Cheetah has never let us down!

Cheers!