Can anybody explain this transfer rate to me?
- Thread starter LiamC
- Start date
Prof.Wizard
Wannabe Storage Freak
- Joined
- Jan 26, 2002
- Messages
- 1,460
It is evident that this drive suffers from a spastic hemiparesis that involves the 15th Megabyte and induces the "patient" to an epileptic state of low transfer.
I suggest administrating 20mg of RMA 3 times a day.
:lol:
I suggest administrating 20mg of RMA 3 times a day.
:lol:
Cliptin
Wannabe Storage Freak
LiamC said:Hi all,
Ive been running a few tests on a few different drives for an upcoming article, and I got this sequential read map from a WD 300BB
Can anybody explain it to me? I've never seen anything like it.[/img]
Did you get that graph o more than one run?
CougTek
Hairy Aussie
It almost looks like if Western Digital placed two platters of different densities inside the drive. I don't think it is possible though. It just looks like.
Tea
Storage? I am Storage!
It'z bizzare. I await an explanation with interest.
jtr1962
Storage? I am Storage!
I would say it's some kind of strange interaction between the drive and controller. Try the drive on another PC, or on a Promise controller(if you have one handy), and see if you get the same results. I've had drives that actually benchmark faster on the inside when running off the motherboard controller of a P-100 machine. The controller in that machine could only do multiword DMA-2 and the drives were just too fast for it. A Promise Ultra 100 TX2 card fixed the problem.
It could also be a driver problem which causes the controller to operate less than optimally, or even a faulty drive. Or it could be gremlins. :wink:
It could also be a driver problem which causes the controller to operate less than optimally, or even a faulty drive. Or it could be gremlins. :wink:
jtr1962
Storage? I am Storage!
BTW, LiamC, is that Kimba the White Lion in your avatar? I remember that show from when I was a kid but it's been so long(~30 years) that I pretty much forgot what he looked like. We even named our first pet cat after him.
Yep, and it's been that long since I saw it 
I've tried the latest VIA 4in1 (4.43)
The Windows 2000 SP3 default driver and run the WD diagnostics over the drive - no faults and same result. The South Bridge is a VIA 8223
Next up is a promise and then I'll try it on my EPoX 8K7A+ (VIA 686B) and see what happens.
Does anybody know where to get updated WD firmware?
I've tried the latest VIA 4in1 (4.43)
The Windows 2000 SP3 default driver and run the WD diagnostics over the drive - no faults and same result. The South Bridge is a VIA 8223
Next up is a promise and then I'll try it on my EPoX 8K7A+ (VIA 686B) and see what happens.
Does anybody know where to get updated WD firmware?
O.K.
Tried the drive on a Promise Ultra-100 controller with latest BIOS and drivers and got the same graph.
Tried a different cable and teh drive seemed to go and extra 2 GB before the catastrophic drop, but is other wise the same graph.
I have benched a 400JB and a D740X on the exact same setup and they had no problems.
Each drive was partitioned and installed from scratch.
WD HD drive diagnostics (extended) report the drive as fine.
Don't know what else to do...
Tried the drive on a Promise Ultra-100 controller with latest BIOS and drivers and got the same graph.
Tried a different cable and teh drive seemed to go and extra 2 GB before the catastrophic drop, but is other wise the same graph.
I have benched a 400JB and a D740X on the exact same setup and they had no problems.
Each drive was partitioned and installed from scratch.
WD HD drive diagnostics (extended) report the drive as fine.
Don't know what else to do...
Since there are no errors or problems with other drives and controllers, the only explanation(s) that I can think of would be specific to the drive itself. Specificly LLF, firmware or hardware.
If you have two platters with different densities then that should show up in drive capacity: It may be close but it should not match what it is actually supposed to be. Further, the geometry of the platters are driven by firmware settings and it is very peculiar that if each platter had different geometries, the firmware would read that from the platter itself rather than having it hardwired. If it is not getting the data from the platters then what would work for one platter would not work with the other. Note it is not impossible that tyhe firmware is getting the data from the platers itself, but it would be a peculiar design.
The most likely alternative is some HW error is occuring on the second platter that is causing a delay with every read and thus requiring two revolutions (instead of one). Thus if the interleave for this drive was actually set for higher then you would get a much better transfer rate. Note it would not be as good as a interleave of one, but it would be close because you would not be requiring two revolutions to get the data on the second half of the drive but rather it would be a revolution plus a sector or two.
The next possibile cause would be a LLF that was interrupted in the middle. a LLF typically sets the interleave. If the original interleave was wrong (causing multiple revolutions to get the data)and a LLF, that set the interleave to be correct, was initiated and interrupted half way through then you could get a transfer rate graph like what you have. However, normally an interrupted LLF will create a discontinity in the format at the interruption point that would at the least create some bad sectors. Perhaps the firware detected those and remaped them. However, that should be detected by a drop (below 1/2 speed) at the discontinity. I suppose that an interruption that was perfectly timed wouldn't cause a discontinuity.
If you have two platters with different densities then that should show up in drive capacity: It may be close but it should not match what it is actually supposed to be. Further, the geometry of the platters are driven by firmware settings and it is very peculiar that if each platter had different geometries, the firmware would read that from the platter itself rather than having it hardwired. If it is not getting the data from the platters then what would work for one platter would not work with the other. Note it is not impossible that tyhe firmware is getting the data from the platers itself, but it would be a peculiar design.
The most likely alternative is some HW error is occuring on the second platter that is causing a delay with every read and thus requiring two revolutions (instead of one). Thus if the interleave for this drive was actually set for higher then you would get a much better transfer rate. Note it would not be as good as a interleave of one, but it would be close because you would not be requiring two revolutions to get the data on the second half of the drive but rather it would be a revolution plus a sector or two.
The next possibile cause would be a LLF that was interrupted in the middle. a LLF typically sets the interleave. If the original interleave was wrong (causing multiple revolutions to get the data)and a LLF, that set the interleave to be correct, was initiated and interrupted half way through then you could get a transfer rate graph like what you have. However, normally an interrupted LLF will create a discontinity in the format at the interruption point that would at the least create some bad sectors. Perhaps the firware detected those and remaped them. However, that should be detected by a drop (below 1/2 speed) at the discontinity. I suppose that an interruption that was perfectly timed wouldn't cause a discontinuity.
Tea
Storage? I am Storage!
The days when the LLF set the interleave under user control finished a long, long time ago. Can't be that. On the other hand, the wise man, when he runs out of sensible ideas, resorts to stupid ideas. WTF? You have nothing to lose, so why not try it?
CougTek
Hairy Aussie
http://support.wdc.com/download/dlg/dlginstall_10_0.exe
I suppose it would do the trick for your drive.
I suppose it would do the trick for your drive.
Buck
Storage? I am Storage!
How would that utility do the trick?
CougTek
Hairy Aussie
This is the description of one component of DLG Diagnostic, a component part of WD's Data Lifeguard.
Unless I misread, I think Liam is trying to find a way to low level format is drive.
CougTek
Hairy Aussie
ARRGGGHHH! Preview Tabarnack! not Submit!
timwhit
Hairy Aussie
I think they will RMA a drive for any reason. I'm not totally sure about that, but it's worth a try.
blakerwry
Storage? I am Storage!
is that part of the 100% satisfaction guarantee?
ring...ring...
WD Rep: WD supp, this is <insert name here> how my I help you?
Blake: Yes, I am an owner of an aging WD hard drive that I would like to return.
WD Rep: Can you tell me what is wrong with the drive?
Blake: Well... im only 60% satisfied with this drive...
WD Rep: Can I get the serial number on that? ...your address? ... ok, your RMA number is XXX-XXXX.
ring...ring...
WD Rep: WD supp, this is <insert name here> how my I help you?
Blake: Yes, I am an owner of an aging WD hard drive that I would like to return.
WD Rep: Can you tell me what is wrong with the drive?
Blake: Well... im only 60% satisfied with this drive...
WD Rep: Can I get the serial number on that? ...your address? ... ok, your RMA number is XXX-XXXX.
Tea
Storage? I am Storage!
Two platters of different densities? Fair go guyz! Can you imagine the design and engineering expense? Effectively, you'd have to engineer an entire new model of drive twice. Nope. I don't buy it.
As for LLF, nope. It ain't that either, because...
(Excuse me Tea. Can I see a long post about drive formatting and embedded servos and 1:1 interleave coming up?)
(Yez.)
(Have you looked at the clock?)
(Oh darn!)
.... it ain't LLF because we have to go to the office, that's why. :wink:
As for LLF, nope. It ain't that either, because...
(Excuse me Tea. Can I see a long post about drive formatting and embedded servos and 1:1 interleave coming up?)
(Yez.)
(Have you looked at the clock?)
(Oh darn!)
.... it ain't LLF because we have to go to the office, that's why. :wink:
blakerwry
Storage? I am Storage!
that is od... why would the transfer rate almost exactly halve itself at the 15gb mark? and since you say it seems to change... 18bg mark now?
it almost does sound like the drive is having to read each part twice....
it almost does sound like the drive is having to read each part twice....
Tea
Storage? I am Storage!
OK, let's ask another question: what is the norrmal transfer rate of that model drive? I have been assuming that the first part is normal and the second part abnormal, but is this actually correct?
Tea
Storage? I am Storage!
Further to Mark's theory, a hardware problem with the second platter should result in a saw-tooth pattern, as it is all-but universal for multi-head drives to read as below:
H0T0
H1T0
H2T0
H0T1
H1T1
H2T1
and so on. This is because a head switch operation is much faster than a even a single track seek. The only exception I have ever seen is certain old IBM notebook drives that Skallas drew my attention to, which went in the bizzare order of:
H0T0
H0T1
.....
H0T999
H1T0
H1T1
and so on
You could see this in the weird seek time graphs they produced where, quite often, a long seek was faster than a short seek.. I have absolutely no idea why they would have produced a drive with this brain-dead design, and I'd be astonished to discover that a modern Western Digital drive would use it. In any case, the majority of WD300BB drives were a three head design, in which case the break-point is in the wrong place. - half-way along the graph when it ought to be 2/3rds of the way.
Here is another theory: perhaps the drive is actually a four head or two head drive which has been artificially de-stroked to give 30GB. (I dunno what the HDD industry calls this, in the CPU game it's "downbinning".) But once again this doesn't wash, as no drive designer would be stupid enough to reduce capacity by, say, skipping every second data block on a certain portion of the platter.
Still a mystery.
H0T0
H1T0
H2T0
H0T1
H1T1
H2T1
and so on. This is because a head switch operation is much faster than a even a single track seek. The only exception I have ever seen is certain old IBM notebook drives that Skallas drew my attention to, which went in the bizzare order of:
H0T0
H0T1
.....
H0T999
H1T0
H1T1
and so on
You could see this in the weird seek time graphs they produced where, quite often, a long seek was faster than a short seek.. I have absolutely no idea why they would have produced a drive with this brain-dead design, and I'd be astonished to discover that a modern Western Digital drive would use it. In any case, the majority of WD300BB drives were a three head design, in which case the break-point is in the wrong place. - half-way along the graph when it ought to be 2/3rds of the way.
Here is another theory: perhaps the drive is actually a four head or two head drive which has been artificially de-stroked to give 30GB. (I dunno what the HDD industry calls this, in the CPU game it's "downbinning".) But once again this doesn't wash, as no drive designer would be stupid enough to reduce capacity by, say, skipping every second data block on a certain portion of the platter.
Still a mystery.
jtr1962
Storage? I am Storage!
Looking at that graph a little more closely it seems that besides the other problem the drive is also stuck in UltraDMA mode 2(33 MB/sec). The STR is flat up to the 13 or so GB mark. Most drives only reach their maximum STR during the first 10% or so of the platter, not for nearly half the drive. This suggests to me that the transfer rate is limited by the interface, and since the M/B controller can do better than 33MB/sec, one of the problems is that the drive is not going higher than ATA-33.
As for the remainder of the graph, the rate appears to be exactly cut in half as some others here have noticed, but is otherwise a normal "stepped" appearance. Normally, when drives are low-level formatted at the factory each successive track is skewed several sectors from the previous one to allow the head to be at the start of the next track immediately after a track switch. If the time for a track-to-track seek increased slightly from the time the drive was factory formatted, it might just miss the start of the track after seeking, and need to wait for nearly an entire revolution before the same sector was available. This would account for the halving of STRs. It is more likely that this would occur on the inner tracks since the relative velocities of the head and platter are smaller, which means a smaller signal at the head. Therefore, the head might take a fraction of a millisecond more for a track-to-track seek but that would be enough if the skew originally had a very small margin of error. A drive manufacturer might use an "aggressive" value of skew to get a higher STR, and WD drives as of late have a higher STR than, say, Maxtor. While this is good, it also means if the drive develops a very slight change in head sensitivity, the STR drops by a factor of two, whereas it would not with a more "conservative"(larger) skew value, although at the expense of a slightly lower STR.
As for the remainder of the graph, the rate appears to be exactly cut in half as some others here have noticed, but is otherwise a normal "stepped" appearance. Normally, when drives are low-level formatted at the factory each successive track is skewed several sectors from the previous one to allow the head to be at the start of the next track immediately after a track switch. If the time for a track-to-track seek increased slightly from the time the drive was factory formatted, it might just miss the start of the track after seeking, and need to wait for nearly an entire revolution before the same sector was available. This would account for the halving of STRs. It is more likely that this would occur on the inner tracks since the relative velocities of the head and platter are smaller, which means a smaller signal at the head. Therefore, the head might take a fraction of a millisecond more for a track-to-track seek but that would be enough if the skew originally had a very small margin of error. A drive manufacturer might use an "aggressive" value of skew to get a higher STR, and WD drives as of late have a higher STR than, say, Maxtor. While this is good, it also means if the drive develops a very slight change in head sensitivity, the STR drops by a factor of two, whereas it would not with a more "conservative"(larger) skew value, although at the expense of a slightly lower STR.
Thanks for picking that up jtr (about the UDMA-33 mode).
Here's what I have still to try to form a definitive opinion
Format drive into three partitions, one either side of boundary and one crossing and test drive throughput on each with a couple of other proggy's to form a definitive opinion - it could just be a bug in WB99.
Use a standard 80-conductor cable. I've been using round ones - but in their defense, they have worked flawlessly with 400JB, D740X, 1000BB-SE and 60GXP - but still, you never know.
Try formatting disk as FAT32.
I'll keep people posted.
Here's what I have still to try to form a definitive opinion
Format drive into three partitions, one either side of boundary and one crossing and test drive throughput on each with a couple of other proggy's to form a definitive opinion - it could just be a bug in WB99.
Use a standard 80-conductor cable. I've been using round ones - but in their defense, they have worked flawlessly with 400JB, D740X, 1000BB-SE and 60GXP - but still, you never know.
Try formatting disk as FAT32.
I'll keep people posted.
Tea
Storage? I am Storage!
Sensible ideas, LiamC. JTR is having useful thoughts also.
Tannin
Storage? I am Storage!
In the old days, LiamC, all this stuff was under user control. You would buy a hard drive and a hard drive controller card. Then you would format the drive using the BIOS on the card. Mostly you would have to get at the BIOS format routine by using DEBUG. Different brands of controller card had different BIOS format routine start points, so you had to the address to jump to.
The controller card would either have a fixed default interleave value (really old ones, such as the Xebecs that came with the IBM XT), or else you had to specify your desired interleave, which you chose by guessing what might be the best. If you guessed too high and, say, ran a 5:1 interleave-capable drive and controller card combination at 6:1, you lost a little performance. If you guessed too low, trying for 4:1 on that same combo, you lost a massive amount - a slipped rev on a 3600 RPM drive is serious stuff!
After that, you used FDISK to partition it (32MB maximum partition size in those days before DR-DOS came along - so you had to split a 40MB drive), rebooted once more and did the high-level format.
Then, after a few years of this stuff, along came Spinrite and a couple of other similar but inferior tools. With Spinrite you could re-do the low-level format on the fly, and change the interleave. In fact, you could get it to test various potential interleaves and figure out the optimum one for your drive and controller card. (I seem to remember doing this with some other program well before I got Spinrite. Was it the Norton Utilities, back when "NU" wasn't a synonym for "pox"? Or PC Tools maybe?)
Believe me, getting the interleave right was a major, major boost. Taking an existing system with a 5:1 interleave and tuning it to get 3:1 or even 2:1 made a huge difference that even your granny could detect. Take a K6-2/350 and a 6GB Fujitsu slug, and replace it with an Athlon 900 and a 7200 RPM ripsnorter - that's the rough order of difference. No need for benchmarks, you could see it instantly.
Err ... What was the question?
The controller card would either have a fixed default interleave value (really old ones, such as the Xebecs that came with the IBM XT), or else you had to specify your desired interleave, which you chose by guessing what might be the best. If you guessed too high and, say, ran a 5:1 interleave-capable drive and controller card combination at 6:1, you lost a little performance. If you guessed too low, trying for 4:1 on that same combo, you lost a massive amount - a slipped rev on a 3600 RPM drive is serious stuff!
After that, you used FDISK to partition it (32MB maximum partition size in those days before DR-DOS came along - so you had to split a 40MB drive), rebooted once more and did the high-level format.
Then, after a few years of this stuff, along came Spinrite and a couple of other similar but inferior tools. With Spinrite you could re-do the low-level format on the fly, and change the interleave. In fact, you could get it to test various potential interleaves and figure out the optimum one for your drive and controller card. (I seem to remember doing this with some other program well before I got Spinrite. Was it the Norton Utilities, back when "NU" wasn't a synonym for "pox"? Or PC Tools maybe?)
Believe me, getting the interleave right was a major, major boost. Taking an existing system with a 5:1 interleave and tuning it to get 3:1 or even 2:1 made a huge difference that even your granny could detect. Take a K6-2/350 and a 6GB Fujitsu slug, and replace it with an Athlon 900 and a 7200 RPM ripsnorter - that's the rough order of difference. No need for benchmarks, you could see it instantly.
Err ... What was the question?
Don't know what the question was, but you did remind me of the solution - SpinRite. To my knowledge it still optimizes interleaves/LLF's.
The problem with partitioning and formatting is that what is needed is to fix the Low level (or intermediate level for those that believe low level involves puting in timing marks) formatting not the high level formating which simply places the OS structure on the drive.
The problem with partitioning and formatting is that what is needed is to fix the Low level (or intermediate level for those that believe low level involves puting in timing marks) formatting not the high level formating which simply places the OS structure on the drive.
Tea
Storage? I am Storage!
Alas, Tannin cut off his potted history of the low-level format at around 1989 or so. (Which is probably the most recent thing he can remember with any clarity.) After that, several things changed. First, the very, very best drive and controller combinations of that era became capable of the ultimate performance tweak: the 1:1 interleave. It was quite a talking point in those dayz, having a 1:1 interleave drive. Of course, all modern drives have a 1:1 interleave as a matter of course. Have had since, oh, 1991 or so, I guess.
A little later we saw the rise of the embedded servo drive. These drives do not allow a user LLF. (Again, all modern drives are made this way - since about 1994 or '95, if Tannin's mumblings are to be trusted.)
And in-between those two times, we say the development of the integrated controller (AKA IDE) which takes a great deal of the tweaking potential away from the user (which is bad) and delivers the maximum possible performance right out of the box (which is good).
Spinrite was best known for its ability to adjust interleave, but even Spinrite can't do it on modern drives. Rule of thumb is, if it's IDE, you can't do it. (Not a 100% rule, there is no cast-iron connection, but it's close enough for our purposes.) Spinrite still calculates skew for its own internal use, but I don't think it can alter that either.
But I would run Spinrite on the drive in any case, as there is something screwy going on and maybe it could shed some light on it. Also, I'd run a progressive seek time test such as vseek, just to se if it shows us anything to give a hint.
A little later we saw the rise of the embedded servo drive. These drives do not allow a user LLF. (Again, all modern drives are made this way - since about 1994 or '95, if Tannin's mumblings are to be trusted.)
And in-between those two times, we say the development of the integrated controller (AKA IDE) which takes a great deal of the tweaking potential away from the user (which is bad) and delivers the maximum possible performance right out of the box (which is good).
Spinrite was best known for its ability to adjust interleave, but even Spinrite can't do it on modern drives. Rule of thumb is, if it's IDE, you can't do it. (Not a 100% rule, there is no cast-iron connection, but it's close enough for our purposes.) Spinrite still calculates skew for its own internal use, but I don't think it can alter that either.
But I would run Spinrite on the drive in any case, as there is something screwy going on and maybe it could shed some light on it. Also, I'd run a progressive seek time test such as vseek, just to se if it shows us anything to give a hint.
yeti
What is this storage?
funny STR graph
Looking at the referenced graph from the 400BB (from the SR site, I still haven't figured how to include a graph in my posts), it seems that at least the first part is identical to that in this thread.
http://www.storagereview.com/articles/200009/20000921WD400BB_STR.html
This drive is then probably a based on the 400BB (7,200rpm 20 GB/p), where, around the middle something makes it lose one rev every time. I like the "too tight a skew" explanation, which would play just right. Also it cannot be an issue of different densities/capacities on different platters as the LBAs usually cover all platters pretty much evenly, the STR would show waves otherwise.
Looking at the referenced graph from the 400BB (from the SR site, I still haven't figured how to include a graph in my posts), it seems that at least the first part is identical to that in this thread.
http://www.storagereview.com/articles/200009/20000921WD400BB_STR.html
This drive is then probably a based on the 400BB (7,200rpm 20 GB/p), where, around the middle something makes it lose one rev every time. I like the "too tight a skew" explanation, which would play just right. Also it cannot be an issue of different densities/capacities on different platters as the LBAs usually cover all platters pretty much evenly, the STR would show waves otherwise.
Bozo
Storage? I am Storage!
Dumb question. Can you run this drive on a motherboard with an Intel chipset?
Bozo
Bozo
OK, this is just plain weird.
Deleted the NTFS partition, (as an aside, whenever I re-installed W2K I deleted the partition and started again) re-created the primary and formattd as FAT32. The mysterious drop disappeared! But the drive is still stuck in UDMA-33 mode.
I had changed the round 80 wire cable for a ribbon, but that made no difference. I also set up the drive as single (was on C/S), and again it made no difference. I'm going to repartition and try NTFS again.
Deleted the NTFS partition, (as an aside, whenever I re-installed W2K I deleted the partition and started again) re-created the primary and formattd as FAT32. The mysterious drop disappeared! But the drive is still stuck in UDMA-33 mode.
I had changed the round 80 wire cable for a ribbon, but that made no difference. I also set up the drive as single (was on C/S), and again it made no difference. I'm going to repartition and try NTFS again.