UDMA33,UDMA66,ATA100.....

[Parts]

UDMA or DMA, well...is there a new thing already or coming in after ATA 100, I don't really understand these things but I see them everytime and what are the differences:Oit's suppose to be faster right?:rolleyes:

[otheos]

Faster? NO!

The only thing that speed is dependent on is the hard drive itself.

In a simplistic way, the hard drive speed is determined by two parameters: the STR (sustained transferred rate) and the access speed.

The interface (UDMA33/66/100/133) only tells how much data can go through it.

A drive with a maximum STR of 12MB/s (typical 2year old 5400rpm drive) will have plenty of room even with the UDMA33 interface (33MB/s max) to perform at maximum speed.

A drive with a max STR of 35MB/s though will find itself being limited by the 33MB/s available to. However since drives can only perform this max STR under specific circumstances (the outer edge of the disk and reading sequentially), even UDMA 33 is still no problem, but UDMA 66 (66MB/s) is advisable.

Since current IDE drives do not exceed 42MB/s at max STR, ATA100 is absolutely useless and there is no performance difference between ATA100 and ATA66 interfaces.

Hard drives have a buffer (a small memory) that is between the disk and the interface. Everything is first writen to the buffer, and then to the disk and everything is read to the buffer and then to the interface (provided it fits). So small files that remain in the buffer can reach the max interface speed when read/written, and this is what we call burst speed. While 100MB/s over 66MB/s sounds much, burst will not be of any advantage as far as performance goes.

Now the other speed factor of hard drives is access speed. This is the speed the drive can find the data on the disk. Access speed is what gives the feeling of a fast hard drive so when buying a hard drive look for the lowest access time. (and this is also why RAID0 offers no speed advantage in non sequential operations).

HD companies have managed to sell loads of very slow drives using the marketing hype of UDMA66 and then UDMA100. People bought hard drives thinking a "100MB/s" drive would be faster than a "66MB/s", but the factors are countless and unless you take into account all of them you cannot tell which drive is expected to be faster. Nevertheless, people are unaware of this and fall into the trap.

Maxtor announced ATA133 a few months ago. Again this will be used for marketing reasons to sell more slow hard drives only this time ATA133 has a reason of existence. ATA interface limits the size of hard drives to 137GB. ATA133 overcomes this and allows >137GB drives. Unless you're after one of those drives, don't bother.

Hard drive performance is the most complex quantity to measure/benchmark objectively. But starting with the physical characteristics of the drive (rpm, platter density etc) and the numbers (access time, STR, etc) is a good point. Interface speed says nothing.

[Comage]

Very comprehensive and detailed explanation, otheos. Thanks for the information. :)

Appreciate that you took the time to type all that out. :)

[Parts]

Thank you otheos:D

[Quixotik]

otheos,
After such a complete answer, could I prevail upon you to explain why some drives are described as UDMAxx versus ATAxx? And what's the difference?!

I have tried to find info on the web and there ain't none!

[otheos]

This is just a notation differentiation.

let me try to explain:

ATA: Advanced Technology Attachment (the standard introduced by companies like Compaq Conner and WD)

This is the original IDE interface, with PIO (Programmed input output) modes 1,2,3 and 4 (up to 16.6MB/s). ATA-2 introduced DMA modes 1,2,3 and 4 up to 16MB/s again but this time in DMA (direct memory access), i.e. taking load off the CPU by accessing RAM directly.

then other companies like WD introduced

ATA3, again DMA mode but up to 33MB/s (hence UltraDMA, UDMA or UDMA33)

then Quantum introduced

ATA4 or UDMA66 (also UDMA-2)

and later

ATA5 is UDMA100 (could also appear as UDMA-3 but saw it only once)

As far as I know UDMA133 (by Maxtor) will be called ATA6 but don't quote me on it.

The fact that the transfer rates were more catchy as far as marketing is concerned, made the names carying them more popular. Also mixtures of these names appeared.

ATA33 = ATA2 = UDMA33
ATA66 = ATA4 = UDMA66
ATA100 = ATA5 = UDMA100

Confusing? Nevermind SCSI names!

[LostBok]

One comment, one question:

Mode 4 also able to use DMA transfers (Direct Memory Access) to take teh load off the CPU, but only at 16.6Mb/sec burst rate... but makes a HUGE differrence to that old Pentium 166 when it comes to running applications, especially since win95 and win98 had the DMA disabled as default.

Okay, I know a UDMA100 drive will perform very much the same on a UDMA66 or 100 controller, but what about RAID striping? Wouldn't the HDD cache's have enought time to fill up and flush constantly, thus enabling the drives to get much closer to the UDMA100 rates... I can't remember the exact numbers but I remember a THG review that certainly indicated that.

Granted, UDMA133 is unliekly to make too much of a difference, but might get HDD manufacturers to start producing drives that get close to pushing the UDMA100 bandwidth :)

[Quixotik]

So, would I be correct in my understanding to say that ATA refers to the connection architecture and UDMA to, essentially, an enhancement? AND, that now they are used interchangably? :rolleyes:

Reason I ask, I was given three mobile HDD racks. On their boxes there is a sticker reading "UDMA66". If I buy an ATA 100 drive and use the rack, is the rack going to cause any 'bottleneck.'

[otheos]

The racks will say ATA66 compatible referring to their cabling.

With ATA66, a new 80core 40pin cable is required (same with ATA100).

Older racks' internal cabling do not have such cabling so even if the cable from the motherboard to the back of the rack was an 80core the drive in the rack would still operate at ATA33 even if it was ATA66/100.

Newer racks have internal cabling allowing ATA66/100 drives to perform at their interface speed (again the cable from motherboard to the back of the rack has to be 80core).

Lostbok, Mode 4 can be either PIO or DMA (actually it's the other way around. PIO and DMA come in Modes 1,2,3 and 4.) Drives can be PIO Mode 4 but this does not mean they are DMA capable.

By the way. PIO and DMA are mutually exclussive. That is to say that if one device is PIO it canno be DMA and vice versa. Also it means that if a device is PIO on the cable, the whole channel becomes PIO and if you connect a second device it will be limited to PIO as well (even if it is ATA33/66/100). Newer drivers are said to overcome this but it remains to be proven to me at least.

DMA however is far more flexible. Once the channel operates at DMA, an ATA33 device say on primary master will not affect the performance of the slave ATA66/100. This is the reason why you should look to buy an ATA33 compatible CD/DVD/CDRW drive. Eventhough the extra bandwidth is not going to be used and for such small throughput the CPU does not take any serious loading by PIO, UDMA is good for keeping your other drive on the same channel to its speed.

[Cruez]

umm. speaking of which...Im pretty sure my ata/66 drive is running off a 40 wire cable. And my MB is ata66 compatible....
Does that mean Its running(transfering) slower than its supposed to?

[otheos]

It means that it is limited to 33MB/s (actually closer to 25-28MB/s with the overheads depending on chipset/driver).

If your drive can do more than 25-28MB/s in STR then its performance may be affected.

[Cruez]

Looks like Im going to go buy a new cable....especially since i just replace the 550 Athlon classic with a 750 Tbird....I guess it would make the system noticeably faster..

[otheos]

I don't know if it's going to be noticeable, but for $3 I would definitely buy one!

[Turnip12]

double post, sorry

[Turnip12]

Otheos, you mentioned that RAID 0 config offers no speed increase for non sequentional operations, can you break that down a little for me? I was under the impression that a RAID 0 setup would theoretically give you double (with two drives) the read/write speed of a single hard drive.
Was wondering if a couple 5400 hds would out perform a similar size 7200 hd if in the RAID 0 conifg, what am I missing on this?

[otheos]

I don't want to start a war on this, so I will put it as mildly as possible.

RAID0 was designed for one specific use: high sequential reads/writes.

When it does so, the theoretical maximum STR would be 2x the speed of the slowest drive.

If you got 2x drives capable of 35MB/s each, the max would be 70MB/s.

Now, to sustain such high speeds some prerequisits should exist.

have you ever seen a STR plot (this curve thats falls as you move inwards the disc due to the smaller area on the inner side...)

This is a sequential plot. What it means in real world is that if the drive(s) was to read a big file (>>1GB) it would perform according to this plot.

However such a use is specific to certain tasks that are far from common. Video/audio/image editing! I do not talk your occasional capture from your DV camera! RAID0 has way too many disadvantages to set it up at your home PC just for an occasional capture (which is only 3.5MB/s anyway). When you need to read big files (10-15GB)B) say to find a shot, to mix video etc, then RAID0 becomes handy as the disks are in constant sequential operation.

However on your home system a RAID0 array will not show any improvement in speed. Like I said the plot only shows the throughput if the read is perform sequentially. i.e if you have your idle drive and you ask to read data at a point that the plot shows it can read at 30MB/s, it is extremely unlikely that it will read at such pace. A random seek like this will be mostly dependent on the drives seek time, and while STR is still important, the lower access drive will finish faster than a higher access drive, even if the STR of the latter is higher.

So for a desktop drive, access times are more important that STR. As for your comparison, the RAID0 of two 5400rpm drives will never much the performance of a single 7200rpm drive. Even the newer 5400rpm drives with 30MB/s max STR (i.e. 60MB/s theoretical in RAID0) when compared to a single 7200rpm with say 35MB/s have a huge disadvantage: not only the access time is higher (usually around 10ms vs. 8.5ms of the 7200rpm) but the have a higher latency! Seek time is the sum of access time and the latency. Add to it the inherent latency of the RAID controller..... you get the picture.

Now 2 5400rpm drives will be faster with RAID0 than without, but not for mission critical data!

People have suggested a 2x 7200rpm RAID0 setup can be as good as a 10K rpm SCSI drive. But unless you use tasks that require STR (after all this is what RAID0 was designed for) the 10K will outperform the RAID0 anytime (seek times are about 80% higher for the RAID0). Let's not bring 15K drives into equation.

Something that RAID0 has over SCSI is price/MB but then again you only buy what you pay (usually).

As for more than 2 drives in RAID0, (assuming 4 channel controllers) the benefit is that the curve that used to dive near the centre before, now stays flat. Such setups are required for applications that demand a minimum STR over the whole disk (like 80MB/s for 180GB). What do you do with it? raw AVI capture is one thing.

What is good about RAID0 is that it gives life to old drives (that's what I use it for). I have 2 4300MB drives in (soft) RAID0 on my linux box and I can get another year out of them (but back them up every night) say to find a shot, to mix video etc, then RAID0 becomes handy as the disks are in constant sequential operation.

However on your home system a RAID0 array will not show any improvement in speed. Like I said the plot only shows the throughput if the read is perform sequentially. i.e if you have your idle drive and you ask to read data at a point that the plot shows it can read at 30MB/s, it is extremely unlikely that it will read at such pace. A random seek like this will be mostly dependent on the drives seek time, and while STR is still important, the lower access drive will finish faster than a higher access drive, even if the STR of the latter is higher.

So for a desktop drive, access times are more important that STR. As for your comparison, the RAID0 of two 5400rpm drives will never much the performance of a single 7200rpm drive. Even the newer 5400rpm drives with 30MB/s max STR (i.e. 60MB/s theoretical in RAID0) when compared to a single 7200rpm with say 35MB/s have a huge disadvantage: not only the access time is higher (usually around 10ms vs. 8.5ms of the 7200rpm) but the have a higher latency! Seek time is the sum of access time and the latency. Add to it the inherent latency of the RAID controller..... you get the picture.

Now 2 5400rpm drives will be faster with RAID0 than without, but not for mission critical data!

People have suggested a 2x 7200rpm RAID0 setup can be as good as a 10K rpm SCSI drive. But unless you use tasks that require STR (after all this is what RAID0 was designed for) the 10K will outperform the RAID0 anytime (seek times are about 80% higher for the RAID0). Let's not bring 15K drives into equation.

Something that RAID0 has over SCSI is price/MB but then again you only buy what you pay (usually).

As for more than 2 drives in RAID0, (assuming 4 channel controllers) the benefit is that the curve that used to dive near the centre before, now stays flat. Such setups are required for applications that demand a minimum STR over the whole disk (like 80MB/s for 180GB). What do you do with it? raw AVI capture is one thing.

What is good about RAID0 is that it gives life to old drives (that's what I use it for). I have 2 4300MB drives in (soft) RAID0 on my linux box and I can get another year out of them (but back them up every night)

[otheos]

^

[gpapad]

It is not true that ATA100 is useless for todays drives. Two ATA100 drives with Maximum serial read speed of 43 MB/s (raid or WITHOUT raid) will raise the use of the IDE channel of an ATA100 controller to its maximum pottetial (this is about 90 MBs or less ??). So an ATA66 IDE channel (two IDE devices) is not sufficient for todays drives, it will partly degrade performance.
ATA133 is not a waste, considering also the fact that platter
dencity is getting higher (so is the maximum serial speed.)

In general we need:

Interface speed > 2 * drive_speed.

:)

[otheos]

NO.

each IDE channel provides 100MB/s max.

Unlike SCSI, no two drives on the same IDE channel can use it at the same time.

So if you have two drives on one channel, each doing 43MB/s, they will NOT give 43+43=86MB/s at once as they cannot use the interface both at the same time. They will give a max of 43MB/s as they only read/write one at a time at their max speed. So, even 66MB/s is enough, and hence 100MB/s is excessive and hence "useless".

The only time that two IDE drives read/write at max speed (of say 43MB/s) are when the two drives are on [bdifferent channels[/b]. And in that case they each have 100MB/s to themselves or 66MB/s for ATA66, which is still plenty.


As for ATA133, give the fact that the PCI bus is limited to 133MB/s and with driver/controler overheads this is around 110MB/s the advantage in bandwidth that ATA133 provides over ATA100 is only 10-20MB/s. As explained above, hard drives will take a long time to saturate the 100MB/s so an increase by 10-20MB/s nowadays is really a joke.

ATA133 however overcomes the 128GB limit of the IDE interface and the only reason it offers the extra 33MB/s (in theory) is as an "extra bonus".

By the time drives reach 100MB/s IDE will be long forgotten anyway.

Finally remember that the Max STR values can only be obtained on the outer edge of the disks and can be sustained for a limited period of time unless performing a real sequential operation which is not often (and hence the non-benefit of RAID0). Finally, for two drives to do max STR at the same time.... well this is completely operation specific and definitely not common.

Hope this answered your question.

[eladijaz]

Ciao,
I'll give a brief -
Raid-0 - can transfer at upto double speed (2 identical drives) once it finds the info it's after. It is of course limited to the same access speed of the original drives. Interfacing with a raid controller doesn't make the drive heads move faster or the disks spin quicker (if you wish hard enough ...)
So to answer your specific question - if the data for which you yearn is not in one big file, placed sequentially on the disk - the heads must go in search of and you still have the same search time ( or sometimes more ) as having one drive. So lots of files/searches means data comes out slower (than the theoretical max).
But yes, for larger files, 2*5400 drives would probably be faster than one 7200 drive, but for smaller files the 7200 would be faster.
EG. I have 2 drives on my system.
(these are the 'safe' averages)
C: 7200 rpm maxtor 10gig
- Data transfer for large files ~17.5mb/s
- Data transfer for small files ~6mb/s
D: 5400 rpm seagate 40gig
- Data transfer for large files ~7mb/s
- Data transfer for small files ~6mb/s

So Double the data rate of the 5400 drive you get 14/12 mb/s
Take into account that raid gives better performance when dealing with larger files - you can see that it might be possible to get a safe average of 14mb/s max. But the 7200 drive already works faster than this. And as far as the small files go - well - you're much less likely to get near the theoretical maximum.
Elad.

As a side note - I only recently tried out the 80wire cable on these drives (So going from a 33MB/s interface to 100MB/s) and noticed no difference in Win98se's performance. I benchmarked and got .... the same scores. Except of course for burst speed - which is, of course, coming from the drives cache ALL 2MB of it! I presume that someone is making use of these speeds but not I.

[fasttoon]

Hi all,

Otheos, You seem to know your stuff, good info.

I want to add my 2 cents worth.

I run win2k and its limited to ata66 transfer speeds (don't ask me why, thats what the specs are) and I do not notice any significant speed difference between 5400 rpm & 7200 rpm drives 2mb buffers, both are ata 100 and the 7200 rpm benchmarks as faster, yet in real life performance its hardly a noticiable difference. I do see that the benchmarks are virtually the same on my ata 33, ata66, & ata 100 drives for uncached speed but the buffer makes a difference, seems like the biger buffer 1mb or larger really makes a difference.

I would say to run all your drives on ata 100 cables, they are cheap and even if your drives are not up to speed it will help in sending clearer signals. I have one on my ls 120 drive and it is faster in transfers by about 2 minutes for a 100+mb floppy, so its an unscientific way of saying they do help at least in my case.

I have a maxtor ata 100 card, and there seems to be a difference between that and the ata slot on the motherboard. I have heard about the ata 5 mode and it does not run at that mode on the motherboard connector, on the ata card it runs at ata 5 mode, during benchmarks I notice the motherboard IDE runs at about 101~130 mb/s avg, on the ata card it runs at up to 150 mb/s, in real life the ata card seems a little faster but not by much. (by the way this is the same drive 7200rpm ata 100 maxtor)

So if you can answer a couple of question and/or point me in the right direction I will appreciate it.

Why does it show up as ata4 on my motherboard vs ata5 using my pci card?

Why in benchmarks does it exceed its rating? I also had a samsung ata66 that benchmarked at 95~101 mb/s

Thanks for the help!