UDMA tranfer rate problem

[pehache]

Hi,

Motherboard MSI VIA KT4AV (MS-6712), latest driver, BIOS 5.2
HDD IDE Seagate ST380021A
Win XP Home up-to-date

My problem: a pretty low transfer speed to/from the HDD. Here are the
perfs get with HDTune:

Average transfer rate: 12.9MB/s
Access time: 15.5ms
Burst rate: 15.2MB/s
CPU usage: 4.8%

In the device manager, the disk appears with an active tranfer mode
"U"UDMA mode 5", which sould ensure a much higher burst rate (100MB/s).


I did not check the IDE cable, but I could read that a 40-wires cable
is fine up to UDMA mode 3. So even if I have a 40 wires cable, I should
have the UDMA 3 perfs.

In the event manager, I can see no disk error.

What could be the origin of the problem ?

Thanks,

 

[Bob Willard]

Hi,

Motherboard MSI VIA KT4AV (MS-6712), latest driver, BIOS 5.2
HDD IDE Seagate ST380021A
Win XP Home up-to-date

My problem: a pretty low transfer speed to/from the HDD. Here are the
perfs get with HDTune:

Average transfer rate: 12.9MB/s
Access time: 15.5ms
Burst rate: 15.2MB/s
CPU usage: 4.8%

In the device manager, the disk appears with an active tranfer mode
"U"UDMA mode 5", which sould ensure a much higher burst rate (100MB/s).


I did not check the IDE cable, but I could read that a 40-wires cable
is fine up to UDMA mode 3. So even if I have a 40 wires cable, I should
have the UDMA 3 perfs.

In the event manager, I can see no disk error.

What could be the origin of the problem ?

Thanks,

ISTR that a 40-wire IDE cable will limit you to 33 MB/s peak, which is
roughly 20-28 MB/s sustainable. But, you aren't getting even close to
that, with a HD that can sustain very roughly 50 MB/s. And, even your
measured access time is ~1.5 times slower than the HD specs.

It might be that your HD is running in PIO mode - that would produce
about the results you report.

A simpler explanation, maybe: if your PC was busy -- particularly if
what it was running used virtual memory -- then the data reported by HDtune
might be terribly misleading. If this is a possibility, repeat the HDtune
runs with the PC as close as possible to standalone.

 

[pehache]

It might be that your HD is running in PIO mode - that would produce
about the results you report.

At least the device manager doesn't report PIO, but UDMA 5 as the
selected transfer mode for the disk.

A simpler explanation, maybe: if your PC was busy -- particularly if
what it was running used virtual memory -- then the data reported by HDtune
might be terribly misleading. If this is a possibility, repeat the HDtune
runs with the PC as close as possible to standalone.

One of the runs was shortly after a reboot, and I have plenty of RAM
(1GB), so I guess it was not the problem. I will try again, anyway,
while monitoring the memory usage.

If it can help, my external USB2 HD Drive (Western Passport 80GB) is
also terribly slow (burst rate of 7MB/s)...

 

[Eric Gisin]

Motherboard MSI VIA KT4AV (MS-6712), latest driver, BIOS 5.2
HDD IDE Seagate ST380021A
Win XP Home up-to-date

My problem: a pretty low transfer speed to/from the HDD. Here are the
perfs get with HDTune:

Average transfer rate: 12.9MB/s
Access time: 15.5ms
Burst rate: 15.2MB/s
CPU usage: 4.8%

It is not in PIO mode, cpu is low.

In the device manager, the disk appears with an active tranfer mode
"U"UDMA mode 5", which sould ensure a much higher burst rate (100MB/s).

I did not check the IDE cable, but I could read that a 40-wires cable
is fine up to UDMA mode 3. So even if I have a 40 wires cable, I should
have the UDMA 3 perfs.

No, the 40-pin cable limits you to udma2, 33MB/s.

In the event manager, I can see no disk error.

What could be the origin of the problem ?

Most likely mwDMA mode2, 17MB/s.

You probably got UDMA CRC errors, causing atapi.sys to drop to DMA.
Start by deleting the IDE controller in Device Manager and let it reinstall.

 

[Folkert Rienstra]

HDTune:

So then, what does the undisputed HD Tach say.

Well, "average" doesn't say much, does it. Just (max + min)/2
One spike to the minimum by a bad sector and the result is ****ed.

Close to 16MB/s so could be MW-DMA mode 2 though a bit highish for that

So obviously not PIO.

UDMA mode 2.

So even if I have a 40 wires cable, I should have the UDMA 3 perfs.

Half of that then.

ISTR that a 40-wire IDE cable will limit you to 33 MB/s peak, which is
roughly 20-28 MB/s sustainable.

I would say 30.

But, you aren't getting even close to that, with a HD that can sustain

very roughly 50 MB/s.

If you want to call 41MB/s max 'very roughly' 50 MB/s.
The average is 35MB/s for that drive.

And, even your measured access time is ~1.5 times slower than the HD specs.

Wrong again. average measured access time for that drive is ~12ms, according to c't.
According to manual the average seek is 9.5 ms so average access (in theory) is
9.5 ms seek + 4 ms latency = 13.5 ms access time.

It might be that your HD is running in PIO mode - that would produce
about the results you report.

A simpler explanation, maybe: if your PC was busy -- particularly if
what it was running used virtual memory -- then the data reported by HDtune
might be terribly misleading.

If this is a possibility, repeat the HDtune runs with the PC as close as possible
to standalone.

Run HD Tach.

 

[pehache grmpf]

Most likely mwDMA mode2, 17MB/s.

You probably got UDMA CRC errors, causing atapi.sys to drop to DMA.
Start by deleting the IDE controller in Device Manager and let it
reinstall.

Wouldn't the device manager rate the disk as "DMA transfer mode", if the
system had drop it to there ?

I already deleted the IDE driver a few days ago, because I saw that the
displayed active transfer mode was "PIO". Now it is displayed "UDMA 5".
Could it be DMA despite of what is displayed ?

Is there a log somewhere, where the possible CRC errors would be reported ?

 

[Rod Speed]

Wouldn't the device manager rate the disk as "DMA transfer mode", if
the system had drop it to there ?

I already deleted the IDE driver a few days ago, because I saw that the
displayed active transfer mode was "PIO".

Thats more evidence of UDMA CRC errors, thats likely
what caused Win to run it in PIO mode, those errors.

Now it is displayed "UDMA 5". Could it be DMA despite of what is
displayed ?

UDMA CRC errors, not the same thing.

While superficially that would normally make you consider
a bad or out of spec ribbon cable, like one of those round
cables or longer than legal etc, that doesnt explain why
the external drive has lousy performance too.

Is there a log somewhere,

Yes, if you're running XP. Right click on My Computer,
select Manage in the popup menu, Event Viewer/System.
Forget whether CRC errors show up there or not tho.

where the possible CRC errors would be reported ?

They should be visible in the Everest SMART report for the drive.
That would be worth checking anyway, because the lousy thruput
may just be due to sectors that can be read successfully with
multiple reads, but not perfectly the first time.
http://www.majorgeeks.com/download.php?det=4181

Again tho, that doesnt explain why you get lousy performance
with the external hard drive too. Maybe its something more basic
like the power supply isnt putting out one of the rails within spec
or its got lots of noise on some of the rails etc.

 

[pehache grmpf]

Thats more evidence of UDMA CRC errors, thats likely
what caused Win to run it in PIO mode, those errors.

Not necessarily, because of:
http://support.microsoft.com/default.aspx?scid=kb;en-us;817472

Before SP2, it seems that Win was anomalously reverting to DMA, PIO,... even
without CRC errors.

UDMA CRC errors, not the same thing.

What do you mean ? Can the device manger display "UDMA 5" if Win has
actually reverted to DMA ?

While superficially that would normally make you consider
a bad or out of spec ribbon cable, like one of those round
cables or longer than legal etc, that doesnt explain why
the external drive has lousy performance too.

Yep. But they may be two different problems...

Yes, if you're running XP. Right click on My Computer,
select Manage in the popup menu, Event Viewer/System.
Forget whether CRC errors show up there or not tho.

No visible CRC errors there...

They should be visible in the Everest SMART report for the drive.
That would be worth checking anyway, because the lousy thruput
may just be due to sectors that can be read successfully with
multiple reads, but not perfectly the first time.
http://www.majorgeeks.com/download.php?det=4181

I have OKs everywhere in the SMART report.

Again tho, that doesnt explain why you get lousy performance
with the external hard drive too. Maybe its something more basic
like the power supply isnt putting out one of the rails within spec
or its got lots of noise on some of the rails etc.

I guess I will have to live with that, anyway )

 

[Rod Speed]

Rod Speed wrote

What do you mean ?

I just meant that the problem with the lousy thruput could
be just be UDMA CRC errors. Said that poorly there.

Can the device manger display "UDMA 5" if Win has actually reverted to
DMA ?

I didnt mean that. And that shouldnt be possible.

Yep. But they may be two different problems...

Yeah, not common that you do get two separate
problems with the same result tho.

No visible CRC errors there...

I'm not sure that they do get listed in there tho.

I have OKs everywhere in the SMART report.

That doesnt prove much, post the actual report.

I guess I will have to live with that, anyway )

No you dont, you could try another power supply.

 

[pehache grmpf]

That doesnt prove much, post the actual report.

http://cjoint.com/?elxgyTMoaE



No you dont, you could try another power supply.

If I was sure that the power supply was poor, I would try...

The voltages doesn't seem to drop when the HDD is in heavily used.

 

[pehache grmpf]

Well, "average" doesn't say much, does it. Just (max + min)/2
One spike to the minimum by a bad sector and the result is ****ed.

The fact is that the the transfer rate is constant over the whole disk.
Apart from some a few peaks, the rate is always between 12 and 14 MB/s

Run HD Tach.

I'll try as well...

 

[Rod Speed]

Rod Speed wrote

http://cjoint.com/?elxgyTMoaE

Its clearly not UDMA CRC errors, those are C7 in
the top table, and there arent any of those reported.

If I was sure that the power supply was poor, I would try...

Presumably that means you dont have another
in a different system you can try for free.

The voltages doesn't seem to drop when the HDD is in heavily used.

Sure, but that doesnt prove anything about the noise tho.

Not that likely tho with no UDMA CRC errors being reported.

Could just be HD Tune getting it completely wrong.
Certainly worth seeing what HD Tach has to say instead.

 

[Michael Hawes]

If I was sure that the power supply was poor, I would try...

The voltages doesn't seem to drop when the HDD is in heavily used.

I used to have KT7A motherboard. I used SCSI drives because of slow HD
performance. Now I have Nforce 4 motherboard, same IDE drives run much
faster, near SCSI speed. Oldrt chipsets just do NOT perform as expected. I
work on a wide range of systems, down to Celeron 550MHz. The older the
system, the slower the transfer rate on a standard diagnostic (DOS based)
that I run as a test of drive condition.
Mike.

 

[pehache grmpf]

Presumably that means you dont have another
in a different system you can try for free.

Correct

Could just be HD Tune getting it completely wrong.
Certainly worth seeing what HD Tach has to say instead.

I've run HDTach, and indeed it reports much higer burst rates than HDTune
(!!), though similar sequential read transfer rates.

For the internal drive: burst=44MB/s, average=9MB/s
http://cjoint.com/?emvevnkDKc (red curve, against blue curve for the same
drive in the HDTach DB)

For the external drive: burst=26MB/s, average=8.9MB/s
http://cjoint.com/?emvfHsDyPH

So it looks now ~OK for the bursts, still not for the average...

Is HDTune known to report wrong rates ?

 

[Rod Speed]

Rod Speed wrote

I've run HDTach, and indeed it reports much higer burst rates than
HDTune (!!), though similar sequential read transfer rates.

For the internal drive: burst=44MB/s, average=9MB/s
http://cjoint.com/?emvevnkDKc (red curve, against blue curve for the same
drive in the HDTach DB)

For the external drive: burst=26MB/s, average=8.9MB/s
http://cjoint.com/?emvfHsDyPH

So it looks now ~OK for the bursts, still not for the average...

Those charts are completely ****ed, they should look
like the blue one for the Barracuda with the internal.

Is HDTune known to report wrong rates ?

Nope, you've definitely got a problem.

Have you got one of the accelerators installed ?

 

[pehache grmpf]

Nope, you've definitely got a problem.

My computer instead

Have you got one of the accelerators installed ?

? accelerators ?

 

[Rod Speed]

Rod Speed wrote

My computer instead

? accelerators ?

Like the intel accelerator.
http://www.pcworld.com/howto/article/0,aid,112382,00.asp

If not, I'd try a completey clean install of XP and see
what happens with the HDTach charts with that.

 

[Folkert Rienstra]

Correct


I've run HDTach, and indeed it reports much higer burst rates than HDTune
(!!), though similar sequential read transfer rates.

For the internal drive: burst=44MB/s,

Well, that certainly rules out PIO and MW-DMA.

average=9MB/s

(From experience) this looks like it is a physical/mechanical problem.
The spike rules out an electrical problem, assuming that the graph is reproduce-
able. Have no idea though how that results in a flat rate, just that I have seen it
before for a drive with excessive shock error.

 

[Rod Speed]

Well, that certainly rules out PIO and MW-DMA.

(From experience) this looks like it is a physical/mechanical problem.

Unlikely given that he gets the same result with
one internal drive and another external drive.

The spike rules out an electrical problem, assuming that the graph
is reproduceable. Have no idea though how that results in a flat rate,
just that I have seen it before for a drive with excessive shock error.

Again, unlikely given that he gets the same result
with one internal drive and another external drive.

Much more likely to be something stuffed up in the OS config.

 

[Rod Speed]

Rod Speed wrote

This gave me the idea to test it under an

unused W98 that I still have on a small partition.

I had to get HDTach2.7, and to run the test several times to have a
full report without any crash, but it was worth the effort! For the
internal drive:
Average transfer rate 42MB/S
Burst Rate 60MB/s
Access time 11ms
http://cjoint.com/?enuxykaEnk

Yeah, very decent result.

(only the first 8GB of the disk are tested, but I guess that this is "by
design")

Dunno, thats not how I recall it worked with 98.

Cant easily test it here tho, I havent gotten around to reconfiguring
the PVR/test machine to allow quick reboots to 98.

Well, it seems that the hardware is OK and that the problem is somewhere
in WinXP...
Yep.

(I didn't test the external drive; is W98 supposed to manage USB2 ?)

Yes, with the appropriate drivers.