LSI U160 / IBM 36Z15 SCSI problems under WinXP

[Camisade]

I recently found the deal mentioned in these forums where Hyper Micro was offering a free LSI U160 host adapter with the purchase of an IBM 36Z15 15k rpm U160 SCSI drive. So I bought one. And the adapter/drive combo works great in a Win2k system, turning out burst read rates of 80MB/sec and suitably respectably sustained read/write rates and close-to-spec mean access times, with a CPU utilization of about 3%.

The problem is, I bought the combo for my WinXP Pro workstation. In THAT workstation, the LSI/IBM U160 combo will only turn out 46MB/sec burst read rates, only about half the sustained transfer rate that it should be generating, and spikes CPU utilization to between 8-12%! Clearly, the system has a conflict there. But how in the heck do I resolve this?

System Config:
Falcon Northwest Mach V
Athlon 1800+ processer
Asus AV7333 (w/ KT333 chipset)
nVidia Geforce 4 Ti 4600 graphics card
C: Maxtor 40GB ATA/133
D: Plextor 24/10/40A CD-RW
E: DVD drive
G: [the ill-performing SCSI drive]
H: Maxtor 80GB ATA/133

Unlike in the Win2k system, where the LSI controllers were assigned their own IRQs up in the 24-25 range, in my Win XP Pro system, they share PCI interrupts with several other devices, and nothing is assigned to an interrupt higher than 21. The LSI controller on which the IBM hard drive is positioned is assigned to IRQ 16, which is shared by the nVidia card and a USB controller.

I'm also suspicious of some kind of conflict because when I put the LSI controller in the system, Win XP takes longer to load the various utilities that auto-start upon login. It's like a DMA channel might be completely choked, but I'm not sure how to resolve this or force WinXP to allocate the resources like it should do to avoid conflicts.

Any ideas?

I've also been working on this with others, to no avail, over at Experts Exchange, in thread:
http://www.experts-exchange.com/Hardware/Q_20476432.html

So, this is stumping some pretty smart folks... anyone here up to the challenge?

 

[mechBgon]

If you haven't installed Service Pack 1 for WindowsXP, that is worth a try. WinXP "raw" can have some nasty problems with SCSI, and SP1 may or may not help.

A possible alternate or additional solution can be found here: StorageReview.com forums

Of course, you could also revert to Win2000Pro.

 

[mechBgon]

This may also be informative: Why are all my devices using one IRQ in Win2K or WinXP?

 

[Camisade]

Thanks, Mech. I'd actually come across both those threads/articles in my search for an answer. The one thing I have not tried is a fresh install with APIC disabled (and I'm REALLY hoping there's a solution that won't force an entire workstation OS reconfiguration (I don't know if you can migrate user state in XP from one disk to another on the same workstation to avoid lengthy software re-installs, etc)).

I've not found any BIOS upgrades for the LSI card. Haven't check recently to see if the nVidia card might be the culprit....

So, no joy.

 

[mechBgon]

Oh, wait a second (kicks self for not reading more closely).

I have an A7V333-RAID. I used to use it in this system: click me

With the A7V333-RAID's onboard USB 2.0 controller enabled, my Adaptec SCSI card's peak throughput was about 49Mb/second. With it disabled, either by jumper or by disabling it in Windows Device manager, throughput jumps to about 72Mb/second as measured by Adaptec SCSIBench at 128kb blocksize, same-sector read. I wrestled with this at considerable length, trying all kinds of convolutions to get the USB 2.0 not to interfere, and trying to get a higher throughput by disabling onboard Firewire, onboard sound, even the native USB 1.1 controller. Different PCI latency settings? Tried it. Different 4-in-1's? Tried it.

On the A7N266-VM board currently in that system, this same benchmark can exceed 120Mb/second. My humble K7S5A, the budget king, is not too far behind. If you wait a few days I will be able to report the results from an nForce2 board as well (EPoX 8RDA+). Suffice it to say, if your A7V333's are the RAID version that has onboard USB 2.0, disable it in Windows Device Manager or by jumper, unless you feel it's worth the cost in PCI performance.

 

[Camisade]

Okay, it would be insane that having an USB support in the system is hindering disk throughput like that. But, at this point, I'm trying about anything, so I disabled the onboard USB support on the A7V333. No difference. But, I also had a USB 2.0 PCI card in (yes, I have that many USB2 devices I use, so normally I have six USB devices connected). After removing ALL USB from the system (verified in device manager), I found the following:

(after updating the LSI U160 card bios, which made no difference. oh yeah, and my Av7333 was the without-RAID option, so there's nothing to disable, there)

(Benchmarks from HD Tach v2.61)

WITH USB 2.0 devices in the system:
Read Burst Speed: 38.7MB/sec.
CPU utilication: ~8%!!!
Random access time: 6.4ms

AFTER REMOVING ALL USB CONTROLLERS:
Read Burst Speed: 58.3MB/sec.
Max Read: 49.4MB/sec
Avg Read: 43.1 MB/sec
CPU utilication: 10.7%!!!
Random access time: 6.4ms

This *IS* --sort of-- an improvement, but at a pretty huge sacrifice and it's unbelievable that I cannot have both SCSI and USB! Also, this still falls far short of the performance I should be getting and *DO* get in another workstation, with a Dell motherboard, where I got results of
Read Burst Speed: 80.0MB/sec.
CPU utilication: 3.2%
Random access time: 3.4ms

Note the differences in the CPU and random access, too, for the same controller/drive in two different machines of about the same speed.

In fact, my ATA/133 drives get tach'd with a read burst speed of 80MB/sec, 44.5MB/sec max sequential read, and 35MB/sec avg seq. read, although they of course use more CPU at max throughput (60%) and have the expected higher random access times of ~11 of drives that are only spinning half as fast.

So, this is STILL broken, if my Ultra 160 SCSI cannot outperform my 7200 rpm ATA/133 drives, even without USB in the system!

 

[mechBgon]

Originally posted by: Camisade
Okay, it would be insane that having an USB support in the system is hindering disk throughput like that. But, at this point, I'm trying about anything, so I disabled the onboard USB support on the A7V333. No difference. But, I also had a USB 2.0 PCI card in (yes, I have that many USB2 devices I use, so normally I have six USB devices connected). After removing ALL USB from the system (verified in device manager), I found the following:

(after updating the LSI U160 card bios, which made no difference. oh yeah, and my Av7333 was the without-RAID option, so there's nothing to disable, there)

(Benchmarks from HD Tach v2.61)

WITH USB 2.0 devices in the system:
Read Burst Speed: 38.7MB/sec.
CPU utilication: ~8%!!!
Random access time: 6.4ms

AFTER REMOVING ALL USB CONTROLLERS:
Read Burst Speed: 58.3MB/sec.
Max Read: 49.4MB/sec
Avg Read: 43.1 MB/sec
CPU utilication: 10.7%!!!
Random access time: 6.4ms

This *IS* --sort of-- an improvement, but at a pretty huge sacrifice and it's unbelievable that I cannot have both SCSI and USB! Also, this still falls far short of the performance I should be getting and *DO* get in another workstation, with a Dell motherboard, where I got results of
Read Burst Speed: 80.0MB/sec.
CPU utilication: 3.2%
Random access time: 3.4ms

Note the differences in the CPU and random access, too, for the same controller/drive in two different machines of about the same speed.

In fact, my ATA/133 drives get tach'd with a read burst speed of 80MB/sec, 44.5MB/sec max sequential read, and 35MB/sec avg seq. read, although they of course use more CPU at max throughput (60%) and have the expected higher random access times of ~11 of drives that are only spinning half as fast.

So, this is STILL broken, if my Ultra 160 SCSI cannot outperform my 7200 rpm ATA/133 drives, even without USB in the system!

I am in 100% agreement! :| I sent a detailed bug report to Asus and posted a detailed bug report at VIA's forum last year about it, and never heard a peep out of them (why am I not surprised). I still have the A7V333-R at home as my IDE-based home system but I remain very disappointed in the USB weirdness.

You appear to have a similar result to mine: throughput is ridiculously low with USB controllers on the PCI bus, and still dissatisfactory without them. I'm thinking VIA just hasn't quite got SCSI in mind with these motherboards. So what next? Replace a motherboard in the Alienware, or what? Wish I had a good anwer for you. If you can get away with an AGP Pro slot, three PCI slots, two DIMM slots and support for only the 200- and 266-bus AthlonXPs (up to 2600+), then the A7N266-VM is only $72-ish at newegg.com. If you need more room than that, then EPoX 8RDA+ is an idea: AGP 8X slot plus 6 PCI, Firewire, native USB 2.0, 5.1 audio (analog jacks standard, Dolby digital header optional). Mad overclocking too

 

[mechBgon]

Oh, and one more thing: I should've said that the 49Mb/sec and 72Mb/sec result was AFTER trying a variety of PCI slots. The second from bottom was the sweet spot for my SCSI card.

 

[Camisade]

Sad news I'm afraid see--relayed to me by one of the really hopeful folks in the Experts Exchange forums:

http://www.tecchannel.de/hardware/817/5.html

and

http://www.theregister.co.uk/content/3/23502.html

I'm afraid for the time being, I'm going to have to shift back to being an Intel fan. So, since I was just going to use the SC600 for an intranet web server, anyway, I guess I'll move my graphics card and other stuff over to that, and use my Falcon with an ATA/133 drive to serve intranet web pages. ..not that I'm going to particularly mind having a desktop workstation with 2 U160 SCSI drives in it.

As we were always taught: Adapt and overcome!

Thanks for your help!

 

[mechBgon]

LOL, there are some skeletons in Intel's closet too, my friend. Look at item 5 on page 12 of this Intel chipset PDF and tell me again how Intel's so much better? 90Mb/sec maximum on the PCI bus with i850E. I'm not saying Intel's bad, just that you're not guaranteed to be free of problems simply by avoiding AMD. Then there's the AGP 3.0 bug in Granite Bay/E7205...

As I mentioned, my nForce board is hitting transfer rates exceeding 120Mb/sec, and I expect to be quite delighted with nForce2. You're $110 away from a whole lot more performance, and you can always sell the A7V333 too. Check out the performance numbers of nF2 versus KT333 and KT400 starting here (AnandTech nF2 shootout with KT333 and KT400 thrown in).

 

[Camisade]

And, how can I be certain that when I drop my Athlon 1800+ into a new board I won't run into this same problem--none of those shootout benchmarks really showed that SCSI transfer rate problem being eliminated, did it? Or am I good to go just because I'm getting away from the Via chipset and moving to nForce2? Lastly (as you can see, I'm really not deeply into hardware tweaking), do I run into any problems moving my PC2700 RAM over into, say, an Epoch EP-8RDA, which is spec'd for PC3200 DDR SDRAM?

I'm gonna have to do something... my-half-cracked plan of just using the server machine was formed without considering that the server box didn't have an AGP slot. 4 64-bit PCI slots, but no graphics would not make for a good gaming rig!

If I didn't have inarguable evidence that something was weird on the Via chipset before, it smacked me in the face yesterday when I set about re-enabling the USB in the system. With every USB device I re-enabled in device manager, the monitor screen flickered/blanked like I'd just done a screen resolution change. Zheesh.

Isn't it ironic that I was relatively happy with my Asus A7N333-based Falcon system with just ATA/133 drives in it, until I spent $220 for a U160 SCSI combo? And now I probably won't be happy until I spend a $100 or so more?

...that's gonna be hard to squeeze out of my discretionary PC-Play allowance after tapping it for this quarter by buying the SCSI combo..."Uh, honey? I know I said this was going to be the last PC-speed toy I bought because I know we're going to be buying another house soon, but after I bought that last toy (the SCSI combo)--which actually works fine, so I can't send it back--my system is actually SLOWER, so now I also need to buy a new motherboard...." <sigh>

Bah! "Damn you, Via!"

 

[mechBgon]

I'll be doing benchmarks and whatever real-world tests I can concoct on my 8RDA+ when it arrives, and reporting the results here. The primary premise in that thread has already been disproven... the PCI bus on nForce2 is not bottlenecked on the hardware side. Whatever issues people are running into are probably driver- or software-related, and I aim to test the theories I postulated.

The nForce2 will run PC2100, PC2700, PC3200, PC3500 and lets you set the FSB to whatever you want, while maintaining a 33MHz PCI speed. For instance, a person needing gargantuan amounts of RAM might opt for three Crucial 1Gb PC2100 DIMMs (ok, no one's reported trying this) and run them at PC2100 speed with a 2400+. Or with your 1800+ for that matter.

 

[Camisade]

Great. I'll look forward to seeing what kind of disk benchmarks you receive with both an ATA/133 and a SCSI drive in the system. My finger will be poised on the Epox EP-8RDA+ "buy" button, awaiting a review. I'm hoping that the review ends with a "Hoo-AH!"

 

[mechBgon]

I posted a summary of my finding, which is that my nForce board and my new nForce2 board both seem to have the same PCI performance. Now, a couple of things:

1) I run Win2000, not XP. The nVidia "Performance" IDE drivers reportedly make the onboard IDE controllers look like SCSI to Windows, and WinXP and SCSI don't seem to get along, as you have already found out. Furthermore, when I did reinstall the nForce Unified Driver Package (which enabled the USB 2.0, incidentally), it didn't offer me the choice of the SCSI-ish "Performance" driver (maybe it only offers it for WinXP??). The native IDE drivers appear as standard IDE, not SCSI.

2) I don't have a DVD drive or any video-editing software, so I can't drag the AGP card and a DVD drive into the mix to make life hard for the northbridge. There's a thread by Herkulese here which he ended with a well-documented rundown of what fixed his DVD-audio corruption problem on his A7N8X-Deluxe.

Anyway, my initial impression here is that it's performing like it should. A bit underwhelming, perhaps, since I've only got a 1700+ for an engine.

 

[Camisade]

I couldn't find your results, but it sounds like all went reasonably well, so I've gone ahead and ordered mine. If I can't get the performance I want under XP, I have no problem going back to Win2k to get a snappier system.

 

[Camisade]

Holy cow what a goat rope this turned out to be! :disgust:

Got my EP-8RDA+ nForce2 mobo in, installed it. It stopped at boot POST code 25, then began sounding an undulating siren reminiscent of a british police car. I rechecked all cable, CPU, and card seating (even though I'd been exquisitely careful the first time) and everything was good to go. Retry. Same thing. ...and again.

So I remove all components other than the IDE drives (2x HDs, 1 DVD, 1 CD-RW), the floppy, and the video card. Same thing.

So I unplug the secondary IDE controller (with the DVD/CDRW). No change. Same with the floppy, then the primary controller hard drives. No difference. So I remove the AGP Geforce 4 Ti 4600. Now the mobo boots past where it was hanging at post code 25 before stopping later in the process. Put the Geforce back in and it hangs again. "WTF," says I to myself? "That's a nearly new card and has never had any problems before!"

Put my Geforce4 in another machine. Error tone. Holy crap, my new mobo toasted my Geforce4 (keep in mind I'm am SCRUPULOUSLY careful of static and always remain grounded to the chassis and know enough not to handle the electronic parts of the circuit cards). Pull the Geforce4 our and stick in an old Geforce2 MX I have laying around.

Same problem: mobo freeze at post code 25. I pull all the cables out and boot... the mobo goes past code 25 before I get a no-floppy error. I hook the floppy back up and get to the BIOS screen. I turn the system off and plug in the CD drives. Same thing--I get to the BIOS screen. I plug in the primary IDE drives. Error 25. "WTF?" again? I swap the IDE cable (even though I know the current one is good--I was just using it!). Same thing. I unplug the CD drives again and can boot to the BIOS. I leave the CD drives out and plug the IDE drives back in, on the original cable. The system boots to the BIOS this time. I scratch my head and plug the CDs into the secondary IDE controller again and this time everything works when it wouldn't before (or at least I get to the BIOS). Zheesh. I plug in the LSI SCSI U160 card and that works, too. So, okay, I can get to the BIOS now, though I'm seeing some squirrely things here and I've toasted my Geforce4. :disgust:

Of course, XP won't boot with the new motherboard in (thought I'd try, anyway). But I'm not ready to blow my old install away yet, though, so...

So I do an XP install to the U160 SCSI drive. Three times, XP install goes through loading devices and by the time it gets to the part where I select what drive I want to install Windows on, the keyboard is locked up and I can't select anything. So now I swap out my old favorite (PS/2) keyboard for another, newer one (still using the PS/2 port, even though the new keyboard offers both). This time, the system doesn't lock up. Nothing wrong with the old keyboard, but apparently the Epox nForce has taken a disliking to it or something.

So I get XP Pro installed to the SCSI drive. I can boot, but it takes a LONG time and a check of the device manager shows that almost no devices are being detected. I have no mouse support. Of course, with no USB devices being detected, my mouse won't work. Silly, but I power down, swap out the USB mouse for an old PS/2 one. STILL no mouse support! And my Geforce is just showing as a VGA adapter and I'm stuck at 640x480. Okay, says I, maybe that's because XP has no idea what chip set this is (although neither does it prompt for any drivers). So I try to install the Epox motherboard nforce drivers, thinking this simply CANNOT be the out-of-the-box experience that Epox shipped this motherboard with!

It turns out that the Epox autorun screen has been designed so that NONE of the elements can be arrived at and selected using the keyboard. You have to click on them with a mouse and mine's not working (remember, both work just fine on other systems--as I verify twice during this process, thinking that maybe the mobo nuked them, too, along with my Geforce4). Hell, maybe it just didn't like the old PS/2 mouse like it didn't like my keyboard. Anyway, I manage to use explorer and the keyboard to get to the CD driver directory....

The drivers install and I reboot....

AND IT MAKES NO DIFFERENCE. I have a computer with a new mobo, with drivers installed and I have NO device support at all. No mouse, no other USB devices get detected, no onboard sound or network adapter are detected...nada, nothing. Zip. You might say I was fuming by this point and you wouldn't be far off. Then I get the error that drive C: (the non-boot IDE drive, right now) is not accessible. WTF? I reboot and the drive comes back, but I still have no device support. Hours and hours have gone by now and I'm losing precious work day time.

I install other drivers, Geforce, USB2, etc. Nothing makes any difference. Zheesh, says I--I think I should have stopped when I first began thinking something was wrong with the motherboard. This CANNOT be the way this product works!

Sighing, I think--well, maybe it *IS* a problem with XP and SCSI drives, though that seems farfetched at this point. So I bite the bullet, pull out the SCSI adapter/drive combo and make one last effort. I begin down the suggested path for moving XP to a new motherboard: boot from CD, and do a Repair of the (original) installation that's on my IDE drive. This removes all the old drivers and basically reinstalls XP while attempting to preserve at least SOME of the user settings (though, things being what they are, and Microsoft being Microsoft, this process is not as recommended as wiping your system and starting from scratch).

Note to self: Next time, make use of the Files and Settings Tranfer Wizard.

So XP "Repairs" the original installation, removing the old drivers and (supposedly) installling itself around the new motherboard.

The behavior is EXACTLY the same as when it was installed fresh on the SCSI drive. Takes forever to boot and no devices are detected. Installing the mobo drivers do not change this. "Screw this," I finally explode. It has to be the motherboard. If putting one of these motherboards in was supposed to be this hard, there would be more about it on the web, somewhere.

So I yank the motherboard, put in all the old components and of course, the system gives an error because the Geforce4 really did get toasted by (I can only think) the new, defective motherboard. I pick up a loaner Geforce4 Ti4200, stick it in, and the old board (A7V333), with all the old components and cabling works fine.

Zheesh. Just my luck to have gotten a FUBAR'd Epox. So I'm back on my old system while I RMA the mobo back to MWave.com, AND I have to replace a high-end graphics card. All because I just *had* to get the performance I wanted out of my U160 SCSI combo.

Unbelievable. Tell me YOUR nForce2 mobo experience went more smoothly than mine?

 

[mechBgon]

Cool, so these voodoo dolls really do work...

Just kidding! But yes, my install went smooth as pie. Bolt in board, plug things in, install Windows. I did stumble across a discrepancy in the USB3 header pinout (make sure to leave pin 10 vacant if using 3rd-party USB wiring, to make a long story short). It has been doing all the same things I used my A7N266-VM for, no problems.

I'll go over your post more carefully when I get home from work. If you happen to have low-RPM or variable-RPM fans then you would want to NOT plug them into the board itself. Also, an update to the latest BIOS would be strongly recommended if your floppy drive is working (put .bin file on floppy, press Alt + F2 at POST screen and wait). Nothing to lose at this point, if you've already got an RMA in process. VERY sorry to hear about your video card! Ok, I gotta get my show on the road here, touch base with you later.

 

[mechBgon]

All righty then...

For starters, the Ti4600 is known to have a physical clearance issue with this board, so what happened to your video card may have been this:

Use caution when installing long video cards.
Use extreme caution when installing long video cards, such as Geforce 4 TI4600, to avoid damage to the video card. Install your memory modules before installing the AGP video card. Avoid bending or inserting your video card into the AGP slot at angles and watch that the video card does not contact the white memory clips. Quickly secure the AGP card to the case after inserting into the motherboard.

from EPoX's FAQs It's not the only board where the DIMM slots are really close to the AGP slot, but it's the one that happens to coincide with the location of sensitive bits of Ti4600's.

Secondly, the IDE weirdness... no idea why it behaved quite like that.

Thirdly, your PS/2 issues... if there was one thing I disliked about my 8RDA+, it was the uncooperative I/O shield, which was reluctant to pop into place (a little too springy for its own good). Its spring contacts against the I/O blocks on the board were darn stiff too. I wonder if the I/O shield was simply keeping your motherboard out of reach of your PS/2 devices' pins. I taught mine not to interfere in a hurry...

The 1/29 BIOS update is probably a good idea if you haven't gotten around to that yet. You might want to try the board out of the case with the keyboard that didn't work before, and see if my theory is correct regarding the I/O plate.

 

[BShaunM]

Here are some links to discussions on the StorageReview boards

From what I understand (and i am no expert) scsi performance suffers in windows xp with certain benchmarks & when using windows explorer to copy/move files.

for a in-depth discussion, go here:
Final Word on SCSI performance in WinXP

Here is a link to a workaround, BUT HEED THE WARNINGS BEFORE USING IT!
Driver Filter


Hope this helps

 

[Camisade]

Yeah, I came across that filter driver in my troubleshooting searches before I gave up on this Via-chipset A7V333 (that I'm still running, while I await replacement of the EP-8RrDA+). It's apparently primarily for write performance, and my beef is read performance. The filter doesn't help much at all with that.

 

[mechBgon]

I don't know how relevant it may be, but I recently began testing Service Pack 3 for Win2000 at work. Just like I've heard, read speeds dropped to near-invisible levels in ATTO on small block sizes (however relevant that may be). Changing the drive to a Dynamic Disk in Windows Disk Management restored it to normal and actually it ended up benching a bit higher than before. So if you do revert to Win2000 and go to SP3, it may be worth doing that (right-click the drive's symbol and the option to Upgrade To Dynamic Disk will appear).

 

[Camisade]

Interesting... I wonder if there is a similarity with respct to a Via / Asus (A7V333) chipset board if the disk is changed to dynamic instead of simple? Or whether they're just hopeless when it comes to high performance SCSI drives anytime you have USB enabled. Certainly can't hurt to try....

 

[zayened]

.

 

[zayened]

Originally posted by: zayened
.

i dont remember posting this???

 

[Camisade]

Heya, MechBgon -

Well, I finally got the replacement motherboard (8RDA+) and installed it, and now everything seems to be working fine. I've reverted to Win2k (without updating to any service packs, yet). But what I'm finding is that the LSI U160 adapter and IBM 36Z15 SCSI drive combo that turned in 80MB burst reads 57MB/sec avg sustained reads on my Dell Win2k server system will STILL only churn out 73MB/sec burst reads (using HD Tach) and 43MB/sec avg sustained reads on the 8RDA+ motherboard. For reference, using HD Tach, my ATA/133 drives are showing 61 MB/sec burst reads and 35MB/sec avg sustained read rates with an FSB setting of 133 (the board defaulted to 100, so I have bumped that). Those ATA/133 results are slower than I was seeing on my old A7V333 motherboard!

I wonder if there are some motherboard settings that I need to tweak? How in the heck are you getting 90-100+MB/sec burst read rates on your system? Can you post some of your key BIOS parameters? I recall you're running a 1700+, and I'm running an 1800+, so there has to be something I haven't set correctly...doesn't there? If you're going to recommend settings to try, I'm running PC2700 RAM.

-Camisade