Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

you don't have problems doesn't mean that other people don't, KCComp. You explain that you don't have problems; enough said. That's all the input should happen. If you don't have problems, you definitely can't repair someone else’s (or offer meaningful advice).

The hot fix is in the seventh stage (according to my source @MS) and is NOT the same one you d/l months ago.

Speak of what you know-not what you think your pals want to hear. Kind of a flame, but come on, you started it. BTW, MS does not change Knowledgebase numbers/identifiers unless something would cause them to release an entirely new one.

Anyway, the patch did not work for me (nine years of SCSI in my PC) and my XP box exhibits the problems. I must use XP for my testing purposes/programs so W2K is not an option.

Thanks for your time.

Share this post


Link to post
Share on other sites
If you don't have problems, you definitely can't repair someone else’s (or offer meaningful advice).

Heh, what bunk, by that statement technical support could not exist unless each support rep actually had the problem the caller was reporting.

The hot fix is in the seventh stage (according to my source @MS) and is NOT the same one you d/l months ago.

OK, first off, I never said I "downloaded the fix months ago". I said everyone downloaded the fix months ago because all MS hotfixes are cumulative and the file in question (ntfs.sys) was included in the Q315406 fix posted to Windows Update on February 10th. If you go to the Windows Update page and look at your installation history, this will be shown as "Critical Update, February 10th, 2002".

Now stay with me as I explain this in factual detail.

The fix you're concerned with, "hard disk performance is slower than you expect", is Q308219. As you can clearly see from this URL:

http://support.microsoft.com/default.aspx?...b;en-us;Q308219

the fix is in the form of NTFS.SYS v5.1.2600.14.

The February 10th update, or "Unmountable Boot Volume" fix, is here:

http://support.microsoft.com/search/previe...b;en-us;Q315403

this fix is in the form of NTFS.SYS v5.1.2600.28 which is inclusive of the earlier fix.

Because all MS hotfixes are inclusive of earlier fixes to the same component, this is probably where you misunderstood your "MS source" as saying this is in the "seventh stage" or whatever. There have been NO changes to the original performance fix, merely other changes to the same component.

the patch did not work for me (nine years of SCSI in my PC) and my XP box exhibits the problems

The patch did not work for you because A) it doesn't solve whatever problem you may or may not have, and B) installing the patch does nothing because you already had a newer version of NTFS.SYS installed on your system which the patch would not overwrite.

BTW, MS does not change Knowledgebase numbers/identifiers unless something would cause them to release an entirely new one.

No, you're misunderstanding the methodology. When you download a hotfix, you may get a newer version of the affected component than is listed in the article. That does NOT mean that the fix discussed in the article has a newer version, it means that the affected component has a newer version because there have been OTHER fixes applied to it.

Let me just say that I am certainly concerned about a potential problem and, not to put too fine a point of it without losing my ability to argue and state my personal opinions in this folder, in a position to do something about it. That is, if someone can convince me that a real problem exists, using:

- Vanilla WinXP with in-box drivers and all current WU

- SCSI Host Adapter on the XP HCL with in-box drivers (not RAID, that's another whole ball of wax)

- Dynamic disks (whether it's a Basic vs. Dynamic thing is, again, another issue, we're talking SCSI performance in XP as a general thing here).

KC

Share this post


Link to post
Share on other sites

Ah, I should note that the problem-repro system should not use the VIA chipset because of known PCI bandwidth issues.

By the way, here is an ATTO I just ran this minute on one of my 3 X15-36LPs using an LSI 53C1010 (Tekram DC-390U3W). Note this disk is full of data, is not optimized, and I had several apps and a dozen instances of IE open at the time (thus the write dip at 128, for example). This is a SiS 645DX P4 system, disk is dynamic, OS is XP with current updates.

KCAtto1.gif

KC

[/img]

Share this post


Link to post
Share on other sites

:?

I note that KCComp doesn't have "the problem" [or problems] the rest of us have been suffering and that KCComp uses dynamic disks. I have read several folks' coments in this thread to the effect that they 'cured' -- or partially cured -- their xp/scsi transfer rate problems by converting to dynamic disks from basic disks. I would be curious is anyone using dynamic disks has experienced the same rate differential between xp and 2000 o/s -- that the problem is not soley with basic disks.

I am unfortunately stuck using XP Home which does not offer dynamic disk option. I also have read some comments that indicate that not everyone is enamored of having to switch to dynamic disks if they want to effectively used scsi HDDs with xp. My transfer rates are significantly lower by measurements of time to transfer same files when I compare XP with Win98 results using Fat32 or when I compare transfer rates among EIDE and SCSI drives under XP -- in the latter case it takes almost twice as long to transfer a 1.6G file with Atlas 10KIII 18G HDDs on Adaptec 19160 as it does with Maxtor D740X 80G HDDs on my primary system. It is not simply a benchmark artifact via ATTO.

Perhaps some view the use Fat32 and WinXP Home as superfluous but it it what I have to live with -- and absolutely no one has yet to say at MS that one should expect a major performance hit using scsi hdds with XP unless you purchase XP Pro AND restrict yourself to dynamic disks.

By the way, I am pleased to hear yet again that XP isn't disasterous for those with dynamic disks. All I ask is that the remainder of us "basic-disk-by-choice" and "basic-disk-only-XPHome" users aren't left out in the cold while MS moves on to bigger and better things.

Sorry if I sound irritated . . . I'm merely p*ssed off.

Share this post


Link to post
Share on other sites

Here is a little experiment for you folks :

- Boot into Windows 2000

- Run the ATTO Disk Benchmark utility on a SCSI disk

- The numbers for reads and writes will more or less match up for all transfer sizes

- Now, open up Device Manager and double-click on the SCSI disk

- Turn OFF the write-caching and hit OK

- Go back, turn it ON and press OK

- Run the ATTO Disk Benchmark utility on the SCSI disk

- The numbers for read and writes will match the ones you see on XP

Interesting, eh? :wink:

Share this post


Link to post
Share on other sites

KCComp - I can claim to satisfy your requirements for documentation of the error under those conditions except the dynamic disk part. Why should I be forced to "upgrade" to some non-standard disk format. It may be that this could be part of the problem, but changing everyone to dynamic disk is not the solution.

Share this post


Link to post
Share on other sites
Here is a little experiment for you folks :

- Boot into Windows 2000

- Run the ATTO Disk Benchmark utility on a SCSI disk

- The numbers for reads and writes will more or less match up for all transfer sizes

- Now, open up Device Manager and double-click on the SCSI disk

- Turn OFF the write-caching and hit OK

- Go back, turn it ON and press OK

- Run the ATTO Disk Benchmark utility on the SCSI disk

- The numbers for read and writes will match the ones you see on XP

Interesting, eh? :wink:

Please Folks read the messages :idea:

I had wrote: If you disable on a Dynamic-HDD the write-cache

you have the same poor write performance like on a Basic-HDD.

Seems the write-cache is always off on a basic-disk. :wink:

Share this post


Link to post
Share on other sites

So we seem to have established that the difference in the transfer rate for write requests between Windows 2000 and XP lies in the harddisk write-cache setting

Now I am quite confident that the harddisk write-cache is really ON in Windows XP, so the problem is definitely not there

It turns out that the ATTO Disk Benchmark is creating its test file with the FILE_FLAG_WRITE_THROUGH attribute. This instructs the OS to write directly to the media. Windows XP seems to be honoring this request and Windows 2000 isn't

So really, we have all been mis-interpreting the results from this utility. There really is no drop in performance [ if there is, then its unrelated ]

Looks like the benchmark does not want to factor in the harddisk write-cache ... no idea why though :?

Share this post


Link to post
Share on other sites

Sigh...

Everytime someone tries to attribute this problem to solely a benchmark issue it's obvious that they haven't read that several of us affected by the problem have tested using real world performance and found significant drop off in write speeds and performance using XP. I've posted on this at least half a dozen times in this thread alone and about every 20 or 30 posts someone says again 'this is just an atto/benchmark problem.'

If it were, then the real world hits would not be seen as they are.

Here's an obvious way to show this problem: if you are having performance hit seen in the benchmark. Create a few large files (wav are great) each 500-1000 MB a piece. Copy those files from one disk to another. Do this both in XP and in Win2k. When I did this (and this is how I first realized there was a problem, even before I came to find this thread) I realized I was seeing more than double the time to copy the same files under XP as I was under W2k. As I do a fair amount of audio and video editing, this makes a big difference that is very noticable in real life. All it takes is a bit of patience, a stopwatch, and some notepaper.

Say it after me... "It is not just a benchmark problem..."

Share this post


Link to post
Share on other sites

Ok guys, I'm returning home this weekend, so over the next couple weeks I plan to do some serious testing on this matter.

Check my sig for what hardware I have available for testing. I have WinXP Pro and Win2k available. I guess the plan is to do a vanilla OS install of each on an IDE drive, and create images on to another IDE drive. So I will be able to do quick OS restores and try various individual changes, then OS restore in between each to ensure that there's no combination questions involved.

So what I'm after now is your suggestions on what configurations, etc to try, and opinions on how to test the results of each of these (which benchmarks, etc).

After I decide on all the configuration parameters to try, I'll ask you guys what you are using and what you have tried, then what symptoms you are seeing with those. Based on that I'll start testing in areas that may appear to affect the issue.

Fire away!

Share this post


Link to post
Share on other sites

pprior - I was trying to explain the anomaly that Eugene is seeing :

With the X15-36LP, reads were in line with expectations. Writes started off very poor and slowly came in line within read speeds as the write size increased. This was not the case with the WD1200JB, reads and writes of which remained pretty much in sych with each other. Note that write speeds on the X15 DO eventually reach read speeds. This is distinct from those who suffer poor write performance across all sizes. This may be a separate, unrelated issue that many folks are confusing with the one at hand.

The reason why IDE disks are not affected is because the ATA spec does not define an equivalent for ForceUnitAccess, which instructs the drive to commit the data onto the media before completing the request [ this is triggered by the FILE_FLAG_WRITE_THROUGH attribute that the ATTO Disk Benchmark sets ]

Judging from the ATTO numbers that you posted, you seem to be experiencing poor write performance even at larger block sizes. Something is definitely wrong and if you are willing to spend some time, I would be happy to help get to the bottom of it

For starters, could you post the ATTO numbers under Windows 2000? Then could you go to the Disk properties page [ via Device Manager ] turn OFF the write-cache and hit OK. Go back, turn it ON again and hit OK. Don't reboot the computer. Re-run the ATTO utility and post the numbers

One more thing, could you please post your hardware configuration again? [ controller and disk ]. I will try to reproduce those numbers out here

Thanks! :)

Share this post


Link to post
Share on other sites

I definitely was seeing problems with large block transfers - again I was seeing <50% performance under XP when copying large files (5-1000mb each).

However, I am no longer in testing mode, as I've reformatted and developed a stable productive system in win2k for quite awhile now. I'm not interested personally in doing any testing, because if I come upstairs just one more time with that "I'm going to be spending the next 5 hours reformatting and reinstalling my computer" story again, I'll be sleeping on the couch for the next couple months :(

For the record my performance was measured using Supermicro P4DC6 motherboard embedded SCSI (which is adaptec U160 dual channel based) running X15 18 & 36 gb drives as well as a Atlas 10K2 (36) and 10k3 (36). 1GB of ECC RDRAM and a Nvidia GF3 video card. (Not cheap/via based hardware).

I look forward to the day when someone figures this out, but neither my libido nor my back will allow me to sleep on the couch for as long as my wife would banish me if I treaded back into this mess again. Good luck.

Share this post


Link to post
Share on other sites

just thought id add in my results, two wd1200 8mb cache 64k raid 0 with hpt372 on an epox 8k7a winxppro

32mb length

31655 write 31805 read on the 1024 line, pretty damn poor...[/img]

Share this post


Link to post
Share on other sites

:twisted: I just spoke to my inside contact at Microsoft and I sent him a link to this thread and he sent me back the following

There is a bug that was fixed last fall in NTFS.sys. A hotfix is available but it requires special authorization to get it. It will be included in XP SP1.

Does that suck or what

You think that MS would fix a hardware problem as important as this with an OS and sent it out ASAP

They suck

Share this post


Link to post
Share on other sites

well i fixed mine somewhat, using an older version of the hpt372 drivers for winxp

80MB write/85MB read using 32mb

still not where it should be (reads are slow) but so much better than the updated driver from ms!!

Share this post


Link to post
Share on other sites
well i fixed mine somewhat, using an older version of the hpt372 drivers for winxp

80MB write/85MB read using 32mb

still not where it should be (reads are slow) but so much better than the updated driver from ms!!

Can you tell us what driver versions you were using before and after? It's useful information to help us sort out who is getting the thread's XP problem and who is getting a different performance related problems such as your's.

And what's wrong with 85MB read speeds? Doesn't sound slow to me.

Share this post


Link to post
Share on other sites
:twisted:  I just spoke to my inside contact at Microsoft and I sent him a link to this thread and he sent me back the following

There is a bug that was fixed last fall in NTFS.sys.  A hotfix is available but it requires special authorization to get it.  It will be included in XP SP1.

Does that suck or what

You think that MS would fix a hardware problem as important as this with an OS and sent it out ASAP 

They suck

My guess is this hot fix is the one that doesn't help, which some of us have already tried.

While talking with an Intel rep today, I heard many good things about SP1 fixing the "holes" in XP. I hope MS has worked on the SCSI issue since last fall...

Share this post


Link to post
Share on other sites
but why would anyone with SCSI material use Windows XP instead of Win2000?

François

Why use Win2000 instead of WinNT 4? :)

OK, so the improvements going to XP aren't the same as going to 2k, but remember XP is the first in the product line that's also targeted to home users.

Share this post


Link to post
Share on other sites

After I read this post, I do a test in my own computer, and find, things are terrible, my fujitsu MAJ3182M 10k 18G harddisk with a LSI 53C1010-33 LVD card only get about 25MB/s write speed and 14MB/s read speed in ATTO bench!

when I convert the basic disk to dynamic disk, the score raise to 42 MB/s read speed and 39MB/s write speed.

But next I run PCMark 2002 pro's disk performence test, in it's detail report it shows write speed without cache even better then with cache! ( without cache, 20 MB/s, with cache , 18MB/s, also the result differ to ATTO a lot). So I guess it must be the Fujitsu harddisk doesn't enable write-back cache.

I searched fujitsu site and google, but didn't find the method to enable the write-back cache.Can any one give me a tip?

Thanks in advance!

Share this post


Link to post
Share on other sites

Maybe because one does not need to restart every 5 minutes or maybe because the drivers are more up to date or maybe because Win2000 is more stable. We cannot keep on going using NT for the rest of our lifes.

I used NT at work for several years, now i like Win2000 over NT4.

XP is more for the low budget computers, a more simple version of Win2000. Bottomline maybe, correct me if i am wrong, no games on NT, but games on Win2000 and a lot of clients are very young and like to play games on their computer.

François

Share this post


Link to post
Share on other sites
XP is more for the low budget computers, a more simple version of Win2000.

Your understanding is wrong. Just because XP combines the NT and the 9x lines, it doesn't mean that it is a simpler version of Win2k. Instead, think of it as the next step up.

NT -> 2000 -> XP

95 -> 98 -> ME -> XP

That's how the two lines merge. Oh, and by the way, you don't see Windows XP Server family simply because it will be called .NET when it is released.

Leo

Share this post


Link to post
Share on other sites

i am using Win2000 myself, but am familiar with XP as well. I liked NT4, i like Win2000, but do not like XP that much. Everytime i use XP, i am glad that at home Win2000 is installed.

It certainly is not for the power user (dixit Microsoft).

Still to each his own and i respect the opinion of others.

François

Share this post


Link to post
Share on other sites
It certainly is not for the power user (dixit Microsoft).

Could you please elaborate on that? What is there in Win2k for a power user that is not available in WinXP? Other than a bug, that is. :)

Leo

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now