Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

Oh and santilli, maybe you could do some other benchmarks with and without SP3 so that we can confirm this. Like copying large files and such stuff... 8)

Share this post


Link to post
Share on other sites

scsi benchmark 32 first, saw the numbers, and uninstalled

SP 3.

Again, is there anything so life shaking in SP 3 that it's worth the speed hit?

I've had no problems with SP 2, and, bug fixes are great, but not at the cost of speed, and for something that I don't do.

gs

Share this post


Link to post
Share on other sites

benched with Adaptec's benchmark, and ended up with between 102-108 mb/sec, settings 256 sustained i/o.

I defragged, and ran it this morning and it's pegged on 110 mb/sec, with only an occassional drop to 108.

gs.

Share this post


Link to post
Share on other sites
Yep, I'm trying to learn about the same thing as you.

I just got a reply to a post over at 2cpus on this very issue.

It indicated that you might be able to have the OS crap out on you to the point that you would have to do fresh install that you would be able just reinstall the OS and then import your dynamic volumes. I guess that is something that you could not do in NT 4.0.

I'm trying to follow up on that issue right now. If it is true then heck with it I'll just do the conversion and go with it. I'm not looking to run any other OS right now anyway, so as long as I can get back to my files saved on different vloumes on the same drive or on different disks or arrays then I'm good.

Ture it might be wrong for MS not to fix this but I like XP and it's not that big a deal to me so I'll just do it.

"g"

I've done this importing of dynamic volumes plenty of times, never had a problem with it. The OS detects the dynamic volume, and marks the disk as 'foreign' from Disk Management. Just right click and choose 'import' (think that's what is was called, can't remember off-hand) and 2 seconds later it's done. I've even imported Win2k dynamic volumes to an XP system. Can't remember if I've tried the other way or not.

Share this post


Link to post
Share on other sites

I have recently installed SP3 for Win2K and haven't noticed any significant results except that my peak read speeds have dropped from ~19Mb/s @ 64k to 15Mb/s. I am still appalled at the poor write performance of what is quite an expensive setup. Read results are quite good. System is as the description in the image except that it was Win2K Server and not pro :oops:

atto3.gif

Share this post


Link to post
Share on other sites

I played around with a Dell Perc/2 card, and had numbers not as high as yours, but, with the same sucky writes.

I think, just my guess, that this is through design, since servers are usually used for fast read access by multiple users.

Just a guess. Could be algorythm design in the raid cards, or perhaps 2kServer is setup for that sort of use as well...??

:?:

GS

Share this post


Link to post
Share on other sites
I played around with a Dell Perc/2 card, and had numbers not as high as yours, but, with the same sucky writes.GS

Now that's weird, because on my PERC2/DC the writes look normal:

attoresult.gif

I didn't change anything, AFAIK... just a normal install of XP Pro. I'm using the standard MegaRAID driver that ships with the OS.

Share this post


Link to post
Share on other sites

Oops,

The subject of my last post indicated that the questions were for TimmyHH, but you can't see the subject line very well..... so, here we go again:

TimmyHH, can you give us a rundown of your system? I'm interested in what kind of drives, MB, etc. Thanks in advance! (Again...)

Share this post


Link to post
Share on other sites

OK, I'll try to include as many possibly relevant details as possible... if you need anything more, just ask.

Overall hardware:

MB: Asus TUSL2-C (i815EP chipset)

CPU: Intel Tualatin 1.0 GHz (not overclocked)

RAM: 512 MB PC133

VGA: Matrox G550

Sound: SB Live 5.1 (not the culprit here... but I don't have any of the LiveWare crap installed)

NIC: 3com 905C-TX

Storage setup:

SCSI: Dell PERC2/DC (AMI MegaRAID), dual U2W, 64 MB cache, write-back mode, 64K stripe size

HDDs: 2x X15-36LP 18 GB (one per channel) as RAID 0

Optical: Tosh 48x DVD, Plextor 40/12/40 on onboard IDE controller

Software:

OS: WinXP Pro with all critical updates (as of today) installed, but no SP1 beta

Partitions: 25 GB NTFS (C:), some Linux... that is, it's not a dynamic volume

Share this post


Link to post
Share on other sites
theotherphil, you need to enable write cache to get your write speeds up.

That's what I thought, but the option to enable write cache is greyed out in the disk properties for some reason :(

Share this post


Link to post
Share on other sites
scsi benchmark 32 first, saw the numbers, and uninstalled

SP 3.

Again, is there anything so life shaking in SP 3 that it's worth the speed hit?

I've had no problems with SP 2, and, bug fixes are great, but not at the cost of speed, and for something that I don't do.

gs

Did you read my post on page 6? Again i claim that it is NOT a speed hit. SP3 shows the values without write cache(since the benchmark app forbids it)... just read some posts in this thread....

Share this post


Link to post
Share on other sites
theotherphil, you need to enable write cache to get your write speeds up.

That's what I thought, but the option to enable write cache is greyed out in the disk properties for some reason :(

RAID controllers don't usually allow you to change the physical disk write cache option from Windows directly. Try looking for the option under the RAID controller in device manager, or through the setup utility for the RAID controller.

However, I was referring to write cache for the controller. Look for a write-back option and enable that.

Share this post


Link to post
Share on other sites
RAID controllers don't usually allow you to change the physical disk write cache option from Windows directly.  Try looking for the option under the RAID controller in device manager, or through the setup utility for the RAID controller.

However, I was referring to write cache for the controller.  Look for a write-back option and enable that.

Ahhh I see now.....but that bench was with the cache enabled on the controller :( I guess to enjoy my substantial investment in U160 RAID goodness I will need to install a proper OS :x Those write speeds are not just a benchmark annomally....I get that sort of speed when copying large files.

Share this post


Link to post
Share on other sites

RAID controllers don't usually allow you to change the physical disk write cache option from Windows directly.  Try looking for the option under the RAID controller in device manager, or through the setup utility for the RAID controller.

However, I was referring to write cache for the controller.  Look for a write-back option and enable that.

Ahhh I see now.....but that bench was with the cache enabled on the controller :( I guess to enjoy my substantial investment in U160 RAID goodness I will need to install a proper OS :x Those write speeds are not just a benchmark annomally....I get that sort of speed when copying large files.

I'm gonna have to read through this entire thread again one day, but the impression I'm under now is that XP and W2K SP3 fix a bug which was preventing ATTO from disabling the cache during the benchmarks. That would suggest that your ATTO results are not a problem.

As far as copying files goes, are you talking about when copying files from the RAID drive to somewhere else on the same RAID drive? You have to expect low scores here, as the disks are reading, moving the heads to where it will be writing, moving back to the original files, etc.

For example, with a 4x 10k SCSI RAID 0 I'm getting around 18MB/s copying files to itself.

Share this post


Link to post
Share on other sites
I'm gonna have to read through this entire thread again one day, but the impression I'm under now is that XP and W2K SP3 fix a bug which was preventing ATTO from disabling the cache during the benchmarks.  That would suggest that your ATTO results are not a problem.

As far as copying files goes, are you talking about when copying files from the RAID drive to somewhere else on the same RAID drive?  You have to expect low scores here, as the disks are reading, moving the heads to where it will be writing, moving back to the original files, etc.

For example, with a 4x 10k SCSI RAID 0 I'm getting around 18MB/s copying files to itself.

I have a secondary IDE HD that I use for storing movies etc. Copying from the SCSI RAID array to the IDE HD is extremely fast, but when copying from the IDE HD to the array is substantially slower indicating that I am indeed seeing something like 14Mb/s write speed.

Here is the comparison between my SCSI RAID0 setup vs a single IBM 60GXP:

atto3.gif

atto4.gif

Although the read speed of the RAID array is good, I am very dissapointed that my single IDE drive doubles the write speed and this is noticable when copying from between the 2 sources. I know that in a server environment where SCSI is predominantly used, the read speed is an important factor but the write speed should surely be much higher than what I am seeing.

Share this post


Link to post
Share on other sites

Did you install SP3 with the uninstall option? If so, try going back to SP2 or whatever you had previously to see what results that gives you.

And I'm presuming you've made sure that the write cache is set to write-back and not write-through? And checked that you're running the latest drivers/firmware?

Share this post


Link to post
Share on other sites

Although the read speed of the RAID array is good' date=' I am very dissapointed that my single IDE drive doubles the write speed and this is noticable when copying from between the 2 sources. I know that in a server environment where SCSI is predominantly used, the read speed is an important factor but the write speed should surely be much higher than what I am seeing.[/quote']

I have them same problem with my IDE RAID but then the other way around... my Write speeds are up to par but read is way to low :?:

zonderiaa.jpg

Maxtor D740x-6L 40GB, WinXP, Highpoint 370A, 16K/32K cluster/stripe

Share this post


Link to post
Share on other sites

Morninglight, I think your ATTO results are fine. The write scores are very nice, and I wouldn't expect much more than what you are getting for reads. Are these results on an empty disk? That can affect scores slightly I believe. And have you tried updating the Highpoint firmware?

Share this post


Link to post
Share on other sites
Morninglight, I think your ATTO results are fine.  The write scores are very nice, and I wouldn't expect much more than what you are getting for reads.  Are these results on an empty disk?  That can affect scores slightly I believe.  And have you tried updating the Highpoint firmware?

I use the latest firmware/drivers on a half full disk. What really puzzles my is thast my writes are higher than the reads. :?:

Share this post


Link to post
Share on other sites

RAID can be a strange beast at times. The number of combinations of hardware, firmware, drivers and benchmarking tools means that you will find something strange with any particular setup if you look for one. Perhaps the caching algorithm used on the D740X is friendly to writes on the Highpoint controller?

I think IDE RAID suffers more from this, as it's mainly viewed as a toy for enthusasts. SCSI RAID is worked on harder to get out all the kinks by vendors because simply put, it matters more.

For example, Compaq (now HP) take RAID controllers and SCSI hard disks and custom write all their own firmware. Only a handful of different model disks are sold in Compaq servers and they all use Compaq's own firwmare which goes through lengthy validation process. So you know that a hard disk sold by Compaq will work nicely on a RAID controller sold by Compaq. That's the theory anyway. I'm sure you could take the hardware and use it in a different environment that it's been validated in and find some quirks.

Share this post


Link to post
Share on other sites
Did you install SP3 with the uninstall option?  If so, try going back to SP2 or whatever you had previously to see what results that gives you.

And I'm presuming you've made sure that the write cache is set to write-back and not write-through?  And checked that you're running the latest drivers/firmware?

I was previously running Win2K Server SP2 and the only difference is slightly worse write speed (-5Mb/s with SP3) on the RAID array than before. IDE performance is unchanged. I am using write-back cache and I have the latest firmware and drivers. This is definitely an OS issue.

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

Share this post


Link to post
Share on other sites

Alpha, I'd say a fair summary of this topic is that there are a number of us seeing major scsi performance hits using XP, but also many (sometimes with very similar hardware no less) that aren't seeing it. Thus I don't think we can make any recommendations for you.

I for one enjoy my fast disks (X15 drives) more than I enjoyed XP, so I went back to win2k.

I keep making the point in this thread that in order to really appreciate the difference - copy a few large files (or a couple gigs of MP3's will do nicely) from one drive to another. If you have the SCSI performance hit, you'll know it very clearly - it was taking more than TWICE as long to do this across multiple physical drives (none of which was the OS drive) with XP when compared with 2K. It's night and day difference, not subtle at all.

It would be great if people would maybe select around a gig of files and do the copies. Run a stopwatch and report back. It's not scientific, but trust me it will impress you if you have the performance hit.

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 :)

FWIW.

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