Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

Don't know if this has been mentioned before (a search didn't bring it up), but the current Maximum PC magazine has a small article on this issue, in which they say that Microsoft's fix will be out "soon". They also gave mention to this Storagereview.com discussion board.

Just thought I'd mention the article.

Share this post


Link to post
Share on other sites

I am downloading SP1 build 1106 now, albeit at a lowly 5K a sec. It will be interesting to see if it helps any. The early beta I'm currently running didnt make much diff.

Share this post


Link to post
Share on other sites

1106 is the final build which was recently leaked to the web.

Rumor is however the final packaged versions (along with multilanguage support) will be build 1110. Whatever it is... The official package comes out Sept. 9

The current 1006 that was leaked (and is confirmed as 1106) has some queer versioning numbers though.

Ripped from elsewhere also:

"The code in this release is said to be final, however the build numbers

are not yet final build numbers, so you'll see some goofy looking text in

the about boxes. Anyone remember the 2600.xpclient... build numbers of

the final Windows XP versions about a year ago ?"

B3G.jpg

Share this post


Link to post
Share on other sites

Just got my new Maximum PC and they have a short mention in WatchDog about this problem as they ran into when making the Dream Machine for 2002. They say that SP1 for XP will not fix the problem but that Microsoft shares the blame for the problem and will have a fix out in the next 3 months. They state that it's because they (Microsoft) told developers to set FILE_FLAG_WRITE_THROUGH and FILE_FLAG_NO_BUFFERING to obtain better perfromance with older OS versions (Win2k's supposed to have the problem but it doesn't show unless you disable write caching). Microsofts Goldner also states that they have seen up to a 10X performance hit depending on packet size.

Goes to show we are all not Crazy!!! And a fix is on the way!!! :D

Pete

Share this post


Link to post
Share on other sites
I am downloading SP1 build 1106 now, albeit at a lowly 5K a sec.  It will be interesting to see if it helps any.  The early beta I'm currently running didnt make much diff.

Have patience, people! The final SP1 is out in a mere week. Why this urgency to try it now?

Leo

Share this post


Link to post
Share on other sites
Apparently XP = Xtra Patience.  :cry:

Well...it's a good thing Microsoft is adressing the problem and doing something about fixing it. I can wait another three months to install XP on my X15-36LP/Tekram DC-390U3W SCSI storage system :D . I did notice that some people don't see the performance hit, like SR and a few forum members. That still perplexes me :? .

Share this post


Link to post
Share on other sites

I think Microsoft's fix might come in the form of a compatibility shim. You see, the Win95, Win98, and other compatibility modes you see in the Properties are actually each a collection of shims. There is a special Application Compatibility Toolkit on the Windows XP CD (in SUPPORTTOOLS) that allows to pick and choose any combination of shims. A new shim that tricks an application into thinking the right through flag is set, without actually setting it, could just do the trick.

Leo

Share this post


Link to post
Share on other sites

I heard about this and decided to check it out.

System Specs:

P4 1.8a

782mb ram, PC2700 DDR

Asus P4S533

Adaptec 29160 BIOS v. 3.10.0, Disconnect Yes, Write back cache N/C

Seagate 9.1GB U160 7200RPM (model ST39236LW)

Windows XP Pro Upgrade. File system Fat32.

I've run the Atto benchmark and it seems that I am close to being within specs of the drive. Seagate specs

Internal transfer being 24.7 to 39.4 MB/Sec, unless I'm misinterpreting them.

csseagate.jpg

Let me know if I'm off or not. I've not seen anyone else using smaller drives.(Heck I might have missed it theres alot of info in this thread.) Most of what I've seen on this thread have been the 18GB and larger drives.

Share this post


Link to post
Share on other sites

A follow up to the description in the pic. I tested it with and without the write back cache and there was very little difference. I also ran the test several times and the results were always close to being the same.

Share this post


Link to post
Share on other sites

I thought my SCSI performance still felt slow even after replacing a controller I thought had gone bad.

So started looking for a fix and found this thread.

I looked for where do download a copy of atto and didn't find one as quickly as I had hoped.

But I do have some large files so I did test with those.

My system is:

AMD 1900 XP

MSI K266 7t MB (Something like that, the book isn't handy)

1 gig ram PC2100

Adaptec 29160N

WD 18310 U80 18gig 10k Drive

IBM DDYS-T36940M U160 36gig 10k Drive

I have Win XP pro and RH Linux 7.3 running on the box. All the latest patches as of last night applied to both.

My NTFS was optimized before the test. Linux is running on a clean ext3 (Journaled filesystem). I am using the linux driver that ship with the OS, not the one from the Adaptec website.

I have a 699 meg data file on a linux server, so I downloaded it over a 100mbps switch using scp and started playing with it.

The scp download took (min:sec):

XP: 5:47

Linux: 4:20

(scp does do encryption and I think the sending machine may have been a botltle neck on that test.)

Then I used xcopy on XP to duplicate the file to the same directory under a different name. (i.e. xcopy a.dat b.dat) On linux I used cp to do the same thing. (i.e. cp ./a.dat ./b.dat)

XP: 8:45 (2 tests, ~20 min blown)

Linux: 1:15 (4 tests, ~4 min blown)

I then copyed the file from the U160 drive to the U80 drive.

XP: :57 sec

Linux: 30 sec.

Someone mentioned that MS should hire some coders, maybe they should just copy the implementation like they did the TCP/IP stack from BSD. It is pathetic that they would be this much slower.

I paid $200+ bucks for XP pro. Linux is free.

You people are VERY nice being content with the thought that "at least they are working on it now." Maybe it will be ready in a few more months.

:x

If this isn't fixed Monday with SP1, I think we need to start putting their heads on the Linux Journal Pike and get this published everywhere we can. Being nice and quite since March hasn't gotten this fixed. $ matters to M$. Being nice isn't getting this fixed.

I bought a new controller thinking my old trusty Tekram was the problem. My new controller is no faster in XP though it is a very fast controller.

Sandra put my performance on par with a U66 IDE drive.

I am not at all happy with this. I will test again with the new numbers as soon as I can get SP1 installed.

:x

Share this post


Link to post
Share on other sites

that's not a remotely reliable metric

ntfs has acls, ext3 doesn't

scp implementations vary widely between platforms, and pscp (of the putty suite) is not known for its speed - most win32 ssh-related things run under the cygwin compatibility dll or use the putty source as a reference

i agree that the xp/scsi issue is an ISSUE though

Share this post


Link to post
Share on other sites
that's not a remotely reliable metric

ntfs has acls, ext3 doesn't

scp implementations vary widely between platforms, and pscp (of the putty suite) is not known for its speed - most win32 ssh-related things run under the cygwin compatibility dll or use the putty source as a reference

i agree that the xp/scsi issue is an ISSUE though

Agreed on the scp. this was done with pscp on windows and openssh on linux. The test was more to show that it was faster to transfer the file over the network than it was to copy with it locally on XP.

As far as acls, I don't what the acryonym stands for. Access Control List??? If that is it, ext3 does maintain ownship and permissions lists. It is also a journaled file system so it should do an advanced write/commit phase that NTFS does not have to do.

Either way, there is NO excuse of a file system to delay this operation 8x. I wouldn't even be happy with a 50% decrease. I bought fast hardware for a reason.

So far, today, Sept 9th, sp1 does not show as downloadable.

Share this post


Link to post
Share on other sites

SP1 is out now, anyone done any benches to see if basic disk performance on SCSI controlers is any better.

I understand that they might have not fixed it yet but was there ANY improvement at all?

"g"

Share this post


Link to post
Share on other sites

My 8 minute copy of a 699 meg file has drop to 3:50 minutes, seconds. That is better, but still a far cry from 1:10 or so in linux.

I am considering setting up a FAT and doing a similar test. My fat that I have now is smaller, but Linux and XP both support fat drives...

Sandra now reports it as a bit slower than the 5400rpm ATA100 drives.

Not good for a 10,000 RPM U160.

Better, but not good enough.

How do I complain to MS?

Share this post


Link to post
Share on other sites
that's not a remotely reliable metric

ntfs has acls, ext3 doesn't

Shouldn't ACL's only effect (slightly) the initial access to a file, while the system performs security checking?

Share this post


Link to post
Share on other sites

[rocketmanx]

> Microsofts already stated SP1 WON'T fix the problem.

[isochar]

> [sP1] does not fix this problem

Are you SURE?

From http://support.microsoft.com/default.aspx?.../SP1FixList.asp , the list of fixes in SP1, there is:

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

> Hard Disk Performance Is Slower Than You Expect

> The information in this article applies to:

> Microsoft Windows XP Home Edition

> Microsoft Windows XP Professional

> SYMPTOMS

> After you install Microsoft Windows XP, hard disk performance may be slower than you expect.

> NOTE: Hard disk performance may be even slower when your computer performs many small

> hard disk read/write operations.

> CAUSE

> This behavior may occur in the following situation:

> You use Small Computer System Interface (SCSI) hard disks in the computer.

> -and-

> The hard disks are formatted as NTFS.

...

> STATUS

> Microsoft has confirmed that this is a problem in the Microsoft products that are listed at the

> beginning of this article. This problem was first corrected in Windows XP Service Pack 1.

That sure looks like the issue being described, doesn't it?

I have an XP Pro/X15-36LP/NTFS system (my Northwood), but I won't have physical access to it until Sept. 19, so I can't confirm this firsthand.

--

Stephan T. Lavavej

Share this post


Link to post
Share on other sites

I'll forgive you since you don't frequent the forums often. :) As seen in a recent Maximum PC article, which can be seen in this thread, Microsoft has acknowledged the problem and know what it is. However,

a previous NTFS fix doesn't solve the problem, nor does the upcoming Service Pack 1. But don't start selling your equipment on eBay, just yet. The Dog has learned Microsoft that a fix is on the way.

Share this post


Link to post
Share on other sites

The f*ck XP-SP1 has now my cool running , as Dynamic, X15-36LP

lowered to write 8,5MB/s.

Read is ok at 58MB/s

Sure ?

YES I AM SURE.

Dino

[rocketmanx]

> Microsofts already stated SP1 WON'T fix the problem.

[isochar]

> [sP1] does not fix this problem

Are you SURE?

From http://support.microsoft.com/default.aspx?.../SP1FixList.asp , the list of fixes in SP1, there is:

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

> Hard Disk Performance Is Slower Than You Expect

> The information in this article applies to:

> Microsoft Windows XP Home Edition

> Microsoft Windows XP Professional

> SYMPTOMS

> After you install Microsoft Windows XP, hard disk performance may be slower than you expect. 

> NOTE: Hard disk performance may be even slower when your computer performs many small

> hard disk read/write operations. 

> CAUSE

> This behavior may occur in the following situation: 

> You use Small Computer System Interface (SCSI) hard disks in the computer.

> -and-

> The hard disks are formatted as NTFS.

...

> STATUS

> Microsoft has confirmed that this is a problem in the Microsoft products that are listed at the 

> beginning of this article. This problem was first corrected in Windows XP Service Pack 1.

That sure looks like the issue being described, doesn't it?

I have an XP Pro/X15-36LP/NTFS system (my Northwood), but I won't have physical access to it until Sept. 19, so I can't confirm this firsthand.

--

Stephan T. Lavavej

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