Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

I use the newsgroups alot, and I plan on unraring the files on the SCSI then moving the completed 700MB file over to the storage, so i would be moving large files around often.

Other than read/write hits, there is absolutely nothing wrong with the speed of the SCSI drive correcT?

Share this post


Link to post
Share on other sites
How often do you copy large blocks of files?  For me, doing audio and video editing, it's quite frequent.  And so the real-world performance hit really mattered to me. If you mostly just run apps and surf the web, you're unlikely ever to notice the difference.  Keep in mind that even the slow writes we're seeing would have been "blazing" just a couple years back :)

From what I understand, the degradation in performance comes from Microsoft actually _fixing_ an old bug, whereby the OS didn't process the FILE_FLAG_WRITE_THROUGH flag correctly. Before the fix, some applications were written to use that flag, and why not if testing showed no performance problems. After the fix, those same applications would suddenly show performance problems.

Now comes the most interesting part. Microsoft said they would upgrade those affected applications that belong to Microsoft. Since the filecopy utility in Windows is obviously controlled by Microsoft, I expect to see a better filecopy in SP1.

Leo

Share this post


Link to post
Share on other sites
i'm too lazy to read the whole post, so i'm just gonna ask it now.

I'm gettinga  new computer soon and I've grown accustomed to WinXP for over a year now and i absolutely do not want to change to another OS.  Problem comes in is that i have enough money to get a SCSI HD (Fuji MAN and LSI ctrl).  Now, will those two have the read/write problems in WinXP? is it a sure fact that i will have the problmes or a luck (like i might or mihgt not have the probleM)?  I don't wanna blow my money on something thats not gonna work hehe

By all means, buy that SCSI drive if you want to. Firstly, you won't have read/write problems, but ONLY write problems, and ONLY in selected applications, and ONLY noticeable every so often. Most importantly, however, the access times (for which SCSI is famous) are entirely unaffected.

Leo

Share this post


Link to post
Share on other sites

Hello everyone,

I may be a little late finding this forum and particular Thread but I thought I would thro in my 2cents. After chating with Bob at Seagate (considered to be the resident hot dog on scsi and fibre channel) he has informed me to visit this thread to see if I could find anything to help me improve my performance of my own X15 although he has told me that my results appear to be excelent to him (althouth its much better in ME) So, instead of coming here and getting advice I will give it. First off I think most of you (a nicely as I can put this) is not installing Windows XP in the manner it should be installed particular to your system specifications and usage. Windows XP is designed and balanced for the abilty to install into as many configurations as possible and thus doing so makes it ready for the WORLD and not particularly for you and your machine. If you want Windows, Motherboards, Hard Drives, and other devices and Apps to run on YOUR machine the way YOU would like it too their are steps you MUST take in order to get the perfect balance between Performance and Stability. After setting up my BIOS and Installing Windows XP "MY WAY" and doing nearly 3 hours of Windows Optimizing I can say my 1.53Ghz Athlon XP PC would beat ANY P4 2.2Ghz that is installed in the normal manner with bios and windows at default settings...ect! Why? Its because of how and what I did before, during and after the installation that makes the system Highly tuned for my purposes. In order for all of us to get the same results we must have the same exact system, windows, drivers and installation methods...ect, but we don't we all have diferent Machines and methods of doing things and that explains most of everything I have seen in this forum. Is there a SCSI issue with my machine running Windows XP with ATTO??? Look Below.

MyX15ResultsUsingATTO.JPG

:D Yes, exactly where it should be for the 2nd Generation X15 under Windows XP. I must say it was higher in both 98 and ME but I would not use those Operating Systems ever again after such a "HUGH" UI and Stabilty Improvement like XP introduced. This test was performed 6 times with the results being nearly identicle each time. I don't see anything wrong with my SCSI under XP, do you? The numbers could be a little higher to really make me happy but thats the price I pay for using WinXP.

And, one last thing to anybody with a similar system could you PLEASE post your ATTO or HD Tach results so we can compare because I think I can get more from this drive but need more reference points to compare too. Also I will share everything I did to get my SCSI system to run this good on XP after more research has been completed and posted.

The Test Bed:

Windows XP Home (Installed and Optimized My Way)

Asus A7V266-E (bios 1007)

AMD Athlon XP 1800+ 1544mhz

256 Mb High Perf, Mushkin PC2100

Seagate X15 Cheatah ST318452LW on Adaptec 29160N

Share this post


Link to post
Share on other sites

By the way, I see TimmyHH has Read Write speeds higher than most of us using a PERC/2. Anybody know what his system is comprised of? Those are similar to the numbers I got on Windows ME with my X15 and my Adaptec 29160, but mine were slightly more consistent.

Share this post


Link to post
Share on other sites

See the low write scores on small block sizes? That's what probably will improve if you convert the volume to dynamic. Don't know if that was directly related to the real-world performance hit many here (including myself) had witnessed, because by the time the dynamic disc stuff came up, I was already back to W2k.

Share this post


Link to post
Share on other sites

Gee Rgray, since you are implying that those of us with low write scores simply don't know what we're doing, perhaps you might share some of your wisdom and tell us exactly what modification we might change to normalize the scores.

(Bet 'cha can't do it)

There are plenty of people with scores similar to yours right out of the box. We haven't found out why some people see the difference and others don't but I doubt it has to do with "3 hours of windows optimizing"

If you want to test this theory, benchmark your drive under a clean XP install - I would bet you will see the same numbers you are now. Perhaps you have seen the dynamic disk increase, but if you've done something else that raises them, this whole thread with over 42,000 hits is awaiting a pearl of wisdom to fall from your mouth....

Share this post


Link to post
Share on other sites

I'm sorry pprior and, you are quite right, 3 hours of Optimizing...lol is NOT gonna fix the issues here and I didnt mean to sound like it would. But I'm an accomplished tweaker and have been tweaking computers for 20 years and I absolutely know that Optimizing a system after install is faster at throughput as compared to a new computer just freshly installed with Windows and then left totally alone at all default settings, thats just common sense. Enough with that, its not the issue I'm after so let's not get into a debate about Optimization. After my Windows XP (Planet-XP) site has been designed and tweaked I will show you mods and Optimizations then, but not here. This is a "SCSI performance in XP" thread not a Windows XP Help site...

Anyway, I just got off the phone with Seagate and they tell me I'm the first person to have solved the 12mb/sec write barrier that they know of (I'm sure many others have too but just not told anyone) and they are going to post my studies and do some more testing of their own. I have been building systems for the United States Government since 1983 and I currently run a super large network. What I do and have been doing for many many years when confronted with a single SCSI device on the scsi bus is to set Enable Disconnection to NO. See, the nature of SCSI is for a device such as a hard drive to connect to the host, do its business as fast as possible and then disconnect to allow other SCSI devices to use the bus (hence multiple devices on the scsi cable) which is a common resource. But, It had occurred to me that disconnects would possibly be disallowed in newer applications like Windows XP but it was just a hunch and so I started my tests again to try and solve the 12mb/sec barrier. I know there are multiple issues here with XP and SCSI throughput in general but I will have those issues looked at later. But this thread was started by someone wanting his write speeds to more closely match his read speeds so I will focus on that. Now, if you read your SCSI manual or read up on the subject of SCSI like I have you will learn that ALL current SCSI Host adapters allow for a setting called Enable Disconnection. WHY? well thats simple taking the following notion that SCSI devices disconnect to allow other devices to use the bus, but what if you don't have other devices? Well, since ALL SCSI Host Card manufacturers claim that setting Enable Disconnection to NO or disable will increase your single SCSI device performance then most people would inheritantly do so. Reading through the SCSI cards manual you would discover that Disconnection should be set to NO for single devices and proceed in doing so just to discover weeks or months later during a benchmark that your write speeds suck. At the time you would not know why but some may discover that disabling Disconnection was the culprit and others may just come to StorageReview and post a "why is my SCSI write speed so terrible under XP" thread like this one. Most people would have forgotten that they set it to no and or simply not realize that it is set to no. DO NOT set Enable Disconnect to NO or Disable on Windows XP machines as I have seen the same results as people in this thread. Some cards may be set to NO Disconnection by default and if that is the case you will not be able to get any SCSI hard drive to write past 12mb/sec give or take a few megs. I have 5400 computers to test this theory out on and so far my results are the same every time. The following posts will reflect screenshots of having "Enable Disconnection" set to NO or disable. By the way, I will also post a screen of the Via patch and what it did for my SCSI Burst speed "Just Amazing" :D thanks to my friends at Via for fixing this but I wish I had discovered it sooner :(

Share this post


Link to post
Share on other sites

This is what all my computers get with Enable Disconnection set at NO or Disabled including my X15 on my personal computer at home.

DisconnectionNO.JPG

This is what I get with it set to Yes or Enabled both at the office and at work

DisconnectionYES.JPG

This is what the 2nd Generation X15 does on a Windows XP machine before the Via RAID Performance Patch

2ndX15test.JPG

This is what the X15 does after the patch is applied

FixedWinXPBurstSpeed.JPG

So you see it very well could be that thousands of people all across the world have been complaining about a write speed discrepancy that was easily remedied if the device was left to Disconnect from the bus which just happens to be the default on my Adaptec 29160N. Obviously setting "Enable Disconnection" to NO has done "EXACTLY" what alot of people here including the topic starter has complained about. Now for other issues regarding SCSI throughput in Windows XP, well, learn to Optimize your Windows or live with it. Until my Web Site is released I highly recommend following Koroush Ghazi's excelent Format to Relax Guide at http://www.tweaktown.com/document.php?dTyp...e=guide&dId=324 as I basically do the same thing before, during and after my installs. If there is another reason for the write speed problem under Windows XP than I too would be affected by it.

Share this post


Link to post
Share on other sites
Guest russofris
Now for other issues regarding SCSI throughput in Windows XP, well, learn to Optimize your Windows or live with it. Until my Web Site is released I highly recommend following Koroush Ghazi's excelent Format to Relax Guide at http://www.tweaktown.com/document.php?dTyp...e=guide&dId=324 as I basically do the same thing before, during and after my installs. If there is another reason for the write speed problem under Windows XP than I too would be affected by it.

Hey guys,

Is this guy for real? I can't tell if he's joking or not.

Thanx for the laugh,

Frank Russo

Share this post


Link to post
Share on other sites

lol...No problem! :D

Yeah its a little much but hey its XP what can I say. Theres no other way to get it to run as fast as like a Windows 9X machine. It's just a resource hogging OS, Unless you take the time and Tweak the hell out of it... Personally I have zero tolerence with a slow Windows box but thats just me :x

Anyway, your welcome!

btw, everyone loves that XP Guide and its doing a lot of people some real good. Give it a try, youl like the results. Just make sure you know what each Optimization does before doing it.

Share this post


Link to post
Share on other sites
Guest russofris
btw, everyone loves that XP Guide and its doing a lot of people some real good. Give it a try, youl like the results. Just make sure you know what each Optimization does before doing it.

rg,

I took a look at the tweak guide, and found it to be a decent guide for new XP users and amature tweakers to find a large collection of mid-level tweaks. I applaud you for your efforts.

However

Neither the tweak guide nor your posting to this thread have come even remotely close to fixing the problem that some of our users are experiencing. Most of us know how our storage subsystems should perform within a 10th of a MB. We are seeing 20% of what we should be getting, and are angered and looking for an answer. Converting to DDisks gives us 70% of the performance we are expecting, and is being used as a temporary measure. Most of us are still not happy with this.

The problem lies within Windows XP, and is not something that appears to be tweakable or fixable at this point.

I am considering looking through pages of registry dumps at this point to find the difference between users with bad configs and good configs to see if there is a difference. It's probably going to be a hack, like the HKEY_LOCAL_MACHINESYSTEMControlSet001ControlClass{4D36E96A-E325-11CE-BFC1-08002BE10318} IDE UDMA hack.

Oh well.

Thank you for your time,

Frank Russo

Share this post


Link to post
Share on other sites

Perhaps it should taken seriously..... at least carefully examined to look at effects.

For example, in one forum discussion that I ran across recently, someone mentioned our lamented performance problems with SCSI HDDs in XP and stated categorically something like "what do you expect when most people tweak their OS to give top priority to an IRQ other than the SCSI controller?"

Compare that to the common recommendation for an XP registry tweak where priority is given the CMOS/system clock, such as the recommendation contained in rgray175's link to tweaktown:

"Set the system to give priority to a major device (by its IRQ)

[HKEY_LOCAL_MACHINESYSTEMCurrentControlSetControlPriorityControl]

IRQ8Priority=1 Determine the IRQ of the device you want to give top priority. Create an entry with the IRQ number (e.g. IRQ5Priority) and give it a value of 1. Recommended that System CMOS/Real Time Clock be given the top priority to provide greatest overall system boost."

I'll be the first to admit that I don't have a clue what effect -- if any -- that would have on the Adaptec 19160 sitting in my rig, but some folks are blindly tweaking their OS, often with programs which do it automatically, without regard to whether the rig has 'peculiar' aspects such as SCSI HDDs, etc.

As to the 'enable disconnect' I think I already tried that [have two hdds on the LVD U160 cable]. However, I will go through and do some tweaking and peeking to see if I can do anything to un-dismal my transfer rates. All suggestions appreciated, especially from someone who is also stuck without the dynamic disk option [XP Home like me].

Share this post


Link to post
Share on other sites

Please do NOT start posting information to another thread. This thread is already huge, and cumbersome, but it has remained remarkably unfragmented given the size. Look at how many people are reading it. If it's off topic, of course start a new thread, but if it's about XP and SCSI performance, please keep it here.

(pretty please?)

Share this post


Link to post
Share on other sites
Please do NOT start posting information to another thread.  This thread is already huge, and cumbersome, but it has remained remarkably unfragmented given the size.  Look at how many people are reading it.  If it's off topic, of course start a new thread, but if it's about XP and SCSI performance, please keep it here.

(pretty please?)

The reason I started a new thread is because the reference to XP SCSI performance was only one part of it's purpose. I was mainly after thoughts which controller/OS combination would give me the best performance.

To anybody who has viewed my other thread (pprior included), would it be appreciated if I copy the information there to this thread? I don't think there's anything new there as far as results/resolutions go, but I'll post the info here if you feel it appropriate.

Share this post


Link to post
Share on other sites

hi all

I though I might add a little bit of info here

Little background 1st, my config is the following :

CPU : Athlon XP 2000+

Mobo : Abit KR7a-133

Ram : 512Mb (1 samsung PC2100 DDR stick)

Scsi adapter : Adaptec 29160 with bios rev 3.10.0

HDs : 2x Fujitsu MAN 3184MP 1x IBM DNES-318350W

As many of you I also had a write performance issue under XP on my Fujitsu drives, tho not as bad as some of you expecienced, it was in the area of 22Mb/s on the biggest block sizes. And I also had a burst xfer limit of 55Mb/s which was quite disappointing considering the performances of the Fujitsu HDDs and the adaptec card

Following the advices and diverse experiences written in this thread, I tried steps that proved useful for some of you.

1St : reverted back to Win2K Pro with SP3 => direct write performance increase to 36Mb/s, burst unchanged

2nd : applied the VIA Latency patch rgray175 spoke about => write performance increase to 46Mb/s, matching my read speed, burst improvement to 77.5Mb/s

here are some screenshots

ATTO :

atto.gif

HDTach :

hdtach.gif

additional info: Enable Disconnection has always been set to Yes, Domain validation is disabled and write cache enabled

Share this post


Link to post
Share on other sites

:?: Is it helpful or necessary that the SCSI card have either its own IRQ or that the IRQ receive priority [i.e. set highest priority in registry]?

I notice that my Adaptec 19160 is assigned IRQ 9 along with my soundcard [santa Cruz], NIC, and others. Can or should I seek to assign a separate IRQ to that slot on my CUSL2-C?

Merely curious -- my ATTO write / read scores start at 85 MB/s / 85 MB/s for small blocks and cap [at 64 and larger] at 12000 / 11000, which is miserable for the Maxtor 10KIII compared to my Maxtor d740-X which renders write/read scores beginning 2400/7100 small block to 40000/41000 at 64 and above.

What is crippling my transfer performance [by tested file transfers as well as ATTO] for both writes and reads? I recognize I may have a separate or even different problem than the majority of those posting on this thread.

PIII 933/512M SDRAM/XP Home

Share this post


Link to post
Share on other sites

Croc, I would have to say no to that as I have tried that very thing with XP but I was not able to assign an IRQ to a particular slot on my board as Microsoft explains Windows XP will ignore any settings the board tries to impose regarding irq's. But, MS does give a workaround if it is absolutely necessary to give a specific device its own irq. It requires that you reinstall Windows and during setup force a HAL (Hardware Abstraction Layer) using Standard PC instead of using the default ACPI configuration. Then and only then will Windows XP accept the IRQ that the board assigns it.

Q299340 How to force a HAL http://support.microsoft.com/default.aspx?...b;EN-US;q299340

And Arkos Reed, I did not mention the PCI Latency Patch, I mentioned the Via RAID Performance Patch for SCSI Card performance improvement under Windows. Not bad at all an over 100% increase in write speed and 34% in burst :D Its a shame that so many people don't realize that Via has implemented the RAID Performance Patch that addresses the Burst speed issue for most SCSI cards and their chipsets. I know atleast a dozen people with Via Chipsets, SCSI systems all running XP or 2K and all of them have SCSI performance issues and none have installed the Via RAID Performance Pack. BTW, dont let the word RAID confuse you as its intended for RAID chips like the ones we are using here and it works weather your using it in RAID or not. But, I must say that the performance patch is just for Burst speed and makes up only a small margin of total performance. Given that the market is saturated with Via chipsets its my guess that their must be thousands of people all over with the same Terrible SCSI performance in Windows XP issue you and I had as this thread is littered with them. Now I just wonder what the diference would be with the same system running XP with the Via Performance Pack http://www.viaarena.com/?PageID=66

Share this post


Link to post
Share on other sites
Croc, I would have to say no to that as I have tried that very thing with XP but I was not able to assign an IRQ to a particular slot on my board as Microsoft explains Windows XP will ignore any settings the board tries to impose regarding irq's. But, MS does give a workaround if it is absolutely necessary to give a specific device its own irq. It requires that you reinstall Windows and during setup force a HAL (Hardware Abstraction Layer) using Standard PC instead of using the default ACPI configuration. Then and only then will Windows XP accept the IRQ that the board assigns it.

Q299340 How to force a HAL http://support.microsoft.com/default.aspx?...b;EN-US;q299340

I would not recommend it, though, unless you have a serious problem (such as a device not working unless you disable ACPI). Do it only as a last recourse.

Leo

Share this post


Link to post
Share on other sites

I have to agree with leokor. Unless you have other issues like the ACPI sound issue or Network Card problems I would also recommend not messing around with IRQ assignments as XP does a good job of handling the hardwares resources. The main problem is the automatically chosen ACPI mode during installation. W2k/XP will use only one IRQ (9) for all PCI devices. Normally this should be no problem, but in these cases the IRQ sharing suffers from bad performance. The computer won't crash, and everything works, but not as well as it should. Two examples: When using a Hammerfall USB/MIDI operation will cause audio stuttering even at highest latency. Data transfers via a network card in the background will disturb audio playback significantly. The remedy is to change from ACPI to Standard-PC mode and then assign IRQ's to your liking, but like I said before, I suspect assigning a lone irq to your SCSI card would not fully help the performance issue if at all but you have to experiment to solve issues like this, not just assume it won't help.

I may just give this a try the next time I feel like installing my Windows XP and then re benchmarking my drives to see what if any improvement has occured in Hard Drive performance. I bet it does not help any tho...

So far Optimizing Windows XP and installing the Via SCSI Performance patch has given me acceptable SCSI Hard Drive Performance on my Adaptec 29160 and X15.

Share this post


Link to post
Share on other sites

Well -- it was a long shot but I thought I'd better ask! Thanks for the prompt responses and input. I don't see that it would be worth it to try the HAL workaround, at least at this point.

rgray 175 : I went through the tweaking process as suggested, managed to pick up some overall performance after three days of messing with it as I had a chance. Many of the tweaks done earlier. However, still no joy with the transfer rates.

Cannot for the life of me figure why both reads and writes are so dismal in ATTO -- but HDTach says reads are average 41,9xx kps [64,9xx max] for Maxtor 10KIII 18G and 21,6xx kps ave [26,6xx max] for Quantum 10K 18G. I assume from earlier post here or somewhere that HDTach shows function without regard to O/S but that ATTO tests WinXP functionality.

Share this post


Link to post
Share on other sites

I don't know if this has any relevence at all, but in this thread over at 2cpu forums, there is description of a new beta adaptec SCSI driver for XP that is stated to fix some performance issues with SCSI CD burners.

It might be worth some checking out. I'm running only 2k anymore, so I can't.

Here is the direct link to the adaptec link for the 29160 card.

I'm not sure whether this has been discussed on this thread before or not - the file is actually dated 3/27/2002, so it's not a new file.

Share this post


Link to post
Share on other sites

Prior: Yeah it was discussed earlier in the thread. It did not help.

Everyone:

Windows XP SP1 going gold any day now!

Might not help a damn thing... but it will elimnate a nice nest of bugs eitehr way. It also contains that NTFS performance fix that has been reported to be unrelated (and does not fix the problem). Regardless everyone can start testing with WinXP SP1 as a base for easy of comparing results (no wondering what hotfixes are in).

Share this post


Link to post
Share on other sites

I'm Running Win2k(SP3, sp2 gave Identical results) With A Tyan TigerMP motherboard and a Mylex 352 with 4 x Maxtor Atlas III 18.2 GB drives, And I have some major problems.

Check this crap,

Copy C:div20vts_01_1.vob to c:vts_01_1.vob

Size: 1073565696

Time: 62734 ms

Transfer Rate: 16.320 MB/s

Copy C:div20vts_01_1.vob to c:vts_01_1.vob

Size: 1073565696

Time: 65359 ms

Transfer Rate: 15.665 MB/s

Copy C:div20vts_01_1.vob to c:vts_01_1.vob

Size: 1073565696

Time: 62125 ms

Transfer Rate: 16.480 MB/s

Copy C:div20vts_01_1.vob to c:vts_01_1.vob

Size: 1073565696

Time: 64578 ms

Transfer Rate: 15.854 MB/s

Done in Diskbench, My atto mark looks fine.

Write Cache enabled:

atto3.jpg

Write Cache Disabled:

atto4.jpg

HDTach, as usual has NO idea what my array is doing.

http://homepages.ihug.com.au/~pgoodwin/hdtach.jpg

More information:

My benchmarks to other drives are fine With:

Array to IDE going a fine 25 meg sec(the drives limit)

however, from IDE to C: of the 1gb file still sits at about 16 meg a sec

Another point of interest, if I transfer a 250meg file, the first time it transfers it does it at the usual 16 meg sec, but after that it seems to cache and gives me the expected 114meg a sec(and continues to do so)

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