Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

I disabled USB on my motherboard, under the advice of another posting on slow SCSI performance, and my burst is back up to full, and so are my transfers on my x 15 

check it out, let me know if it works for you.

Tried it, no go for me.

I have however had some success with using Dynamic Disks. I also have applied VIA RAID performance patch, George's 020 beta latency patch, and played around with the hotfixes. So who knows what's made any positive or negative changes :?

I've only just now found where to download ATTO from, and have previously been basing my results on HDTach, which for me has read speeds peaking at around 57MB/s, avg 46MB/s, but write speeds only consistent at 30MB/s. I guess it uses transfer sizes of 16k-32k or so.

Won't have time to see if the improved ATTO results with Dynamic Disks translates to real world improvements (copying files locally) until I get back home in a few weeks. I'm going to do a clean install, make an image, then try applying the different combinations of patches and test with a number of benchmarks. ATTO, HDTach, and timed file copies are on the agenda so far, anybody suggest anything else I should be testing with as well?

I'll be sure to let you all know of any findings.

Share this post


Link to post
Share on other sites

*sic*...501 posts, and almost 29,000 views...DAMN...I think that Microsoft better take a look at this.

(someone send microsoft tech support guys the link...)

Share this post


Link to post
Share on other sites

I realize this is a massive thread, but those of you posting scores on WD IDE hard drives, and IDE raid setups, please stop it. this is a thread about -SCSI- performance, and all those postings are not helpful.

Share this post


Link to post
Share on other sites

At first sorry for my English.

Have read all the posts very interresting and was not be able

to solve the problem.

I have 2 X15 36LP 18GB on a Adaptec 19160 under XP.

Now the funny thing !!!

If i disable the WriteCache on the cheetahs i have the same

terrible Write Performance like under a Basic Partition.

Maybe the WriteCache is disabled on a Basic. :idea:

Share this post


Link to post
Share on other sites
I realize this is a massive thread, but those of you posting scores on WD IDE hard drives, and IDE raid setups, please stop it.  this is a thread about -SCSI- performance, and all those postings are not helpful.

IDE RAID is also affected, because all IDE RAID controllers interface with Windows through a SCSI miniport. You are right about standalone IDE, though.

Leo

Share this post


Link to post
Share on other sites

I read about this FIX with shitty SCSI performance in WindowsXP.

Microsoft stated on TechNet: "Contact Support if you need this fix right away, other wise we suggest you wait untill this FIX is released in SP1."

This has been an issue for ever..MS is fixing the issue with the release of SP1. So most of you will all your tweaks, etc. wont fix anything. It is XP.

Well this is my first post....hiya.

Share this post


Link to post
Share on other sites

Let's paint it black: What if that fix that will come with XP SP1 is the same some folks here already tried months ago and which did not work? Wait for SP2? :?

Share this post


Link to post
Share on other sites

I ran a ATTO test on my Seagate ST318452LW drive with the disk first formated as basic ntfs and after converting the disk to dynamic. The differences were unbelievable to say the least. Before I go through the trouble of installing Win2000 Pro, can someone please tell me what the numbers should be in ATTO for the following test parameters. I can't imagine that I would benefit by going back to W2K Pro. I have folowed this thread for a while and just coverted to dynamic disk 3 weeks ago. No

troubles or glitches have been observed to date. But if there's more performance to be had I WANT IT ALL!

Total Length: 32mb

Direct I/O: checked

Overlapped I/O: Selected

Transfer Size: 0.5 to 1024.0 kb

Below are my ATTO test results using the above

parameters.

TEST RESULTS

(WINXP PRO - BASIC DISK, NTFS)

Trans.

Size Write Read

---------------------------------

.5 228 9147

1 504 16200

2 1013 27800

4 2007 40565

8 3856 41221

16 7358 43604

32 12276 44096

64 23245 44313

128 30690 43712

256 31946 44096

512 33352 45343

1024 29099 45480

TEST RESULTS

(WINXP PRO - DYNAMIC DISK, NTFS)

Trans.

Size Write Read

---------------------------------

.5 8642 9273

1 16416 17414

2 27451 30999

4 45025 47967

8 56055 57517

16 56819 58573

32 57517 58573

64 56326 58738

128 57009 58573

256 57288 58738

512 57710 58738

1024 57710 58738

My system is as follows:

Supermicro P4DC6 M/B w/ dual Channel Adaptec U160 SCSI (2

drives on each channel)

2 - Intel 1.7 Ghz Xeon Processors

1 - Seagate ST318452LW 15K 18.4GB (System 4.19GB Used)

1 - Seagate ST336752LW 15K 36.7GB (Pagefile)

1 - Seagate ST373405LW 10K 73.4GB (Pagefile)

1 - Seagate ST318406LW 10K 18.4GB

1 - 3.5" Floppy Drive

1 - 250MB Iomega Internal Zip Drive (IDE)

1 - Plextor PX-W21410A CD/RW (IDE)

1 - HP DVD Writer 200i (IDE)

1 - Elsa Gladiac 920 (29.42 Drivers)

1 - Intel 10/100 Onboard Ethernet

1 - Logitech Keyboard and Mouse

1 - Hercules Game Theater XP Sound Card

4 - 256MB PC800 ECC Memory (Corsair CM618DR256-800)

19" Sony Trinitron Monitor

APC 1000XL Smart UPS

PC Power and Cooling 450XE Power Supply

Lian Li PC70 case

Digital Doc 5 (All fans connected to DigitalDoc)

1 - Sunon 120mm Fan

6 - Sunon 80mm Fans

2 - Sunon 92mm Fans

Sorry for the post with test results listed this way, but I could figure out how to get pictures in this message (Dumb ME). First time posting on Storage Review Forum.

Share this post


Link to post
Share on other sites
Guest Eugene

Hi folks,

I've been actively taking a look at this issue on the SR test bed for the past few weeks... unlike perhaps most of the machines that have put it to the test so far, this one exists solely to differentiate performance in a controlled environment. Even so, the results I've come up with parallel that of certain readers.

I utilized an X15-36LP and WD1200JB under our testbed's setup in conjunction with Atto Powertools (which, incidentally, is a "benchmark" that we've never favored and in fact owes its amazing popularity simply to one overzealous poster who evangelized its use... but that's ancient history in SR standards) to see if I could replicate problems.

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.

Utilizing Total Copy and shuffling a 600 MB ISO around these two disks as well as the testbed's own Barracuda ATA IV indicated that the X15-36LP's write speeds topped out around half the speed of the WD1200JB.

Converting the X15-36LP to a dynamic disk brings write speeds up to speed with with reads in APT as well as exhibiting an increase in Total Copy speeds. However, neither the dynamic disk nor Win2k allows the X15-36LP to hit total copy write speeds like that of the WD1200JB... peak speed was 37 MB/sec for the WD and 31 MB/sec for the Cheetah.

I apolgize for the vagueness and lack of illustrations... I'm writing this from notes and don't have access to the APT screen shots to cut and paste from as others have done in this thread... I'll hopefully be able to image them back to a partition where I can copy them in due time.

A few more random notes:

1) The testbed features a ASC-29160 controller. I haven't had the chance to control this variable yet. I also have a Mylex acceleraid 370 host adapter yet haven't been able to get it to recognize drives yet. Still working on this.

2) I've spoken to SCSI design team leaders at both Seagate and Maxtor. To my surprise, neither firm has noticed any problems. They've promised me to look into it.

3) Remember that even should write speeds be halved in Windows XP vs Win2k when utilizing SCSI drives, the difference in speed should not be as noticeable as many have stated. Analysis of traces drawn from actual desktop usage indicate that about 25% of accesses are writes. Further, these accesses, generally smallish in nature, predominately consist of actuator movement and not the actual process of writing to the disk.

As a result, its highly likely that those who report a noticeable performance increase when moving to Win2k are experiencing an unrelated performance increase. The first thing that comes to mind is the snap of a fresh installation. It may be that those who report their fresh Win2k install is much snappier than their older XP one would have also noticed the same increase had they simply formatted and reinstalled XP itself fresh again. This applies, obviously, to "typical" usage. If one's usage consists predominately of copying ISOs around (or other large files), then obviously there'll be a large difference ;).

I'll try to have more for everyone soon.

Regards,

Eugene

Share this post


Link to post
Share on other sites

Well I'm thrilled that the big guns are also seeing this problem. I also continue to be absolutely AMAZED that the companies involved here are not all over this like flies on you know what.

I have noticed huge performance differences because I often do large file transfers (several gigabytes at a time) across drives. So doubling write speeds makes a huge difference.

This stuff is not made up, folks. And remember you read it here at SR first!

Share this post


Link to post
Share on other sites

Thanks for carrying the torch, Eugene.

I've been slack on this issue, since I've gone back to Windows 2000 myself.

Share this post


Link to post
Share on other sites

I would bet that the big companies involved here haven't even bothered to test this. It is -so- outrageous that such a performance hit exists, that I bet they just assume it's a "benchmark anomoly." You've seen microsoft's replies thus far: NOTHING.

Share this post


Link to post
Share on other sites

On the contrary, from my personal contact, I was pleasantly surprised with Microsoft’s interest and immediate response. Unfortunately the temporary patch they sent me does not help. I guess we’ll have to wait for SP1.

Share this post


Link to post
Share on other sites
On the contrary, from my personal contact, I was pleasantly surprised with Microsoft’s interest and immediate response.  Unfortunately the temporary patch they sent me does not help.  I guess we’ll have to wait for SP1.

What patch was this (Q308219) ???.

Thx,

Toni

Share this post


Link to post
Share on other sites

If M$ is so darn interested in a fix to this problem, why is it taking more than 6 months to fix a major file subsystem bug? Think about that - their flagship OS has a glaring performance problem and 6 months later they haven't fixed it? One would think they might use part of their 40 billion in cash reserves to hire a programmer or two to fix this problem?

Absolutely amazing.

Share this post


Link to post
Share on other sites

(wish editing posts was allowed)

I re-read my last post and it sounds a bit more like Microsoft bashing than I intended. I think MS has done a -great- job improving their OS in the last two upgrades (win2k and XP) and I hope they continue to do so. It's just frustrating to not be able to use it because of such a big performance hit and I can't understand why they aren't all over it like you know what on a stick.

Anyway, I just wanted to make that clear. I sure hope they do fix it, and soon. When are the .net server packages supposed to be coming out? Aren't they based on the xp system (i.e. XP server)? If so, it will sure be interesting to see if this bug exists in that package. What a disaster if it does!

Share this post


Link to post
Share on other sites

I remember posting about this issue months ago. I first noticed it when I first installed XP. Everyone said I was crazy, including 3ware and Microsoft. Eventually I got nowhere and gave up on it.

Under XP my disk speeds are abysmal. I think there's two main reasons for this:

1. XP has some kind of issue with SCSI devices. I believe the main one is that being an NT base, XP likes to force write cache to DISABLED. Then it grays out the option to adjust it so you can't turn it on. Hence poor write speed on any device that even IDENTIFIES as SCSI, such as IDE RAID controllers. I believe I saw an MS knowledge base document on this but I have been unable to track it down. When I first mentioned this problem to MS their reaction was "Why would you ever want to cache writes on SCSI? You might lose data if the power goes out."

2. The default filesystem read/write functions only like to write one disk cluster at a time. Thus if you're using an NTFS with 4k clusters, the OS wants to read or write 4K before it reads or writes the next 4K. How do you stripe a single 4K cluster to a 4 drive 64K stripe RAID? You don't. You need at least 256K of data per read/write to keep the drives humming at peak speed. You never get it. Some apps like video capture aps use write routines that are faster. Don't believe me? Check the ATTO results on RAIDs, it uses different write sizes. I get miserable speeds even on reads until I get above the stripe size of my raid, which is the default 64K.

For anyone needing more proof that this is an OS issue:

Windows XP ATTO peak read: 72MB/s

Windows XP ATTO peak write: 19MB/s

FreeBSD Bonnie read test: 80MB/s

FreeBSD Bonnie write test: 62.5MB/s

Timed file copy (I broke out a stop watch and timed a file copy from one partition to another on the RAID array using explorer and cp):

Windows XP 700MB in 1:53 average 6.2MB/s

FreeBSD 700MB in 11.2 seconds average 62.5MB/s

Note I dual booted the OS on the same machine so the hardware specs are identical:

Controller: 3Ware Escalade 6400, latest firmware and driver

Drives: 4x IBM 75GXP 46.1GB 2MB cache 7200rpm

RAID type: RAID 0

Stripe size: 64K

XP was using a default NTFS filesystem.

BSD was using a FFS with fast writes enabled.

Both partitions for read and write were identical formats for each OS and both files were read from and to the same location on the physical discs (I reformatted the two partitions for each test).

Share this post


Link to post
Share on other sites

My times have been way off for the longest time...I thought that it was my controller w/XP but now...

Come on, Microsoft!!! :evil: I called, got a password, and d/l the hotfix (Q308219). Same crappy performance...

:?

Share this post


Link to post
Share on other sites

*sigh*

I've posted before in this thread, everyone already has that hotfix, it was posted on Windows Update in February.

I have three X15-36LPs on Windows XP and do not have any performance problems whatsoever.

This thread has degenerated to the point of uselessness. People ask why people like Microsoft aren't doing anything about it -- it isn't clear that there's anything to do something about, or if there is, what in fact needs to be done.

There are possibly one or two people here with some familiarity with rigorous, scientific testing, but the rest of this is just noise. All different benchmarks, IDE drives, drives with data and without, host adapters, RAID host adapters, IDE RAID adapters.... yeesh.

KC

[/b]

Share this post


Link to post
Share on other sites

The noise level has indeed gotten quite high in this humongous thread (I think it's up there in competition with that "Apology to China" thread :) ), but the fact does remain that there is clearly a problem.

In my own experience, copying large (1+ gb) files and blocks of files, I saw more than 100% speed improvement going from XP back to win2k and this agreed with the poor benchmarks seen under winXP. In other words, this is not only a a benchmark issue, this is sitting with a stopwatch and counting time during real-world use. It is -not- a subtle difference.

Couldn't agree more about all the IDE folks, but I'll guarantee there is something buggy in the SCSI/XP interface here.

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