XP users w/ SCSI HDD's, look inside

[John]

Hard Disk Performance Is Slower Than You Expect (Q308219)

http://support.microsoft.com/default.aspx?scid=kb;EN-US;Q308219

 

[Trashman]

Thanx for the info John...

 

[bozo1]

Seems like they would have released the fix by now since the Q article was dated September. Hmmm

 

[BuckNaked]

Check here for some more discussion...

Dave

Edit: The above link is to a discussion in the forums over at Storagereview, and if your at all concerned with your SCSI performance under XP, its well worth the lengthy read.

 

[bozo1]

Wow - what a long read that was.

Found out that I don't seem to have the problem. I have Adaptec's release candidate drivers for XP installed so either that fixed the problem or I never had it to begin with. Wish I knew which one.

Hmm, it's been a while, maybe it's time for a fresh install.

 

[bozo1]

Well, did a fresh install and still don't have the problem.

Seems like it's hit or miss on who has the grief.

 

[BuckNaked]

bozo1,
At the begining of the thread I was wondering if it was a chipset issue (I was thinking VIA and the PCI latency issue), but repeatedly throughout the thread it was evident that it wasn't, and MS seems to own up that it seems to be an OS issue.... If in fact it is specific to XP, I am very curious why your not experiencing the problem....

Dave

 

[bozo1]

Same here. I was assuming it was the new drivers from Adaptec but I proved it wasn't the case. I'm on an AMD chipset, I've seen reports of the problem with NVidia, Via as well as Intel chipsets.

The only thing different I see between me and the guys in that thread is that I have my 29160 in a 64-bit slot.

 

[BuckNaked]

You ought to bench it with ATTO and post a screen shot in that thread, as well as your opinion, as it looks like MS is following the thread...

Dave

 

[mastertech01]

Thanks for posting this John. I had been following this concern for awhile now. I have had very similar results using my onboard Adaptec U160 controller as well as my ATTO Raid array with ATTO benchmarking. I guess we have to individually request the fix? Too bad they cant just post these things for download

Isnt this nice, you get 2 free support helps and then you pay. This for something that should be a hotfix for download at the update site. 300.00 for the crippled OS isnt enough?

 

[toant103]

no wonder, i can't even bench my scsi raid setup in sisoft sandra pro. too lazy to Call MS up

 

[Trashman]

If anyone is interested, I was able to get the so-called "hotfix" from Microsoft.....It didn't help me any, think I might re-install XP and try fix......worth a shot.....but anyway, just PM me and I can send you the file.

 

[Sundog]

<< Isnt this nice, you get 2 free support helps and then you pay. This for something that should be a hotfix for download at the update site. 300.00 for the crippled OS isnt enough? >>

:frown:

 

[Trashman]

pm's sent.

 

[mastertech01]

Thanks Richard, I will give the fix a try and let peeps know if it helped.

 

[mastertech01]

Well at least on my raid controller the small chunks are writing faster than reading with the fix, which is a substantial improvement. I still have to test with the onboard controller as of yet.

 

[NFS4]

BUMP...just got me SCSI hard drive and I'm interested in some results.

 

[mastertech01]

Well I just ran a battery of tests and the Raid array is working just fine. However the onboard 39160 with the original version X15s, and an IBM are just rediculously low. I may try using Win2k and see how it all works then. I just hate to have to look up Win2k drivers for all my stuff that XP already has running with no added driver support. I get transfers read and write nearly the same up to 142 MBPS with the raid array, which is an ATTO raid controller, U2W type dual channel with 4 36LP 18GB drives. It seems to be doing just fine. But the onboard 39160 on the Tyan Thunder K7 gives about 5-10 percent of the read performance in write performance.. in other words a read of about 10-20 times what the writes are, especially in the smallest chunks but also substandard in the largest chunks as well. Total thruput is about half what it normally is for the X15s in reads...less than 25MBPS average. I tried enabling and disabling read and write caches but to no avail. The fix seems to only affect positive performance on the raid array which has the OS installed.

 

[NFS4]

Well, I tried the SCSI patch and it didn't affect anything

 

[dababus]

does it have to do anything with 64 bit pci bus instead of 32 bit pci. i notice my 10k II doesn't perform very well under xp pro, using tekram dc 390u3d card.

 

[mastertech01]

Well they are both 64 bit busses on my system. The onboard controller with the original version X15s will give read and write equally only at 32kb chunks at approx 41MBPS... above or below that it drops dramatically and read write differentiates greatly. The worse striken drive is the IBM, which in smaller chunks hardly shows on the scale. Curiously in the lower kb chunks it will write fast and read dreadfully slow. This is with or without read caching enabled. These things are all over the charts in various bench settings. The only stable setup is the raid array. I am however not totally convinced that the ATTO benchmark properly reflects performance of all controllers or drives... since ATTO is a hardware controller with software driven raid, it may be optimized differently. Still I have used Easy SCSI Bench32, and it shows lack luster performance as well. And you might as well toss SISOFT in the dumpster for SCSI benchmarking. If I had a version of HDTACH which benches SCSI that would help. Perhaps I should try converting one of the drives to a dynamic disk as suggested in another thread.

 

[BuckNaked]

Richard,
Thanks for the hotfix, but I am still awaiting my 10K 3 from Dell to test it on...

Dave

 

[mcveigh]

interesting read, I just got my scsi drive yesterday, i'm reading all the scsi posts I can. I don't plan on using XP with this one though.

 

[mastertech01]

Ok, going with a dynamic volume does some minor improvement, mainly on the lower end. But I WAS able to get near full performance by converting to dynamic and creating a striped volume with the two early generation X15 and I mean near normal RAIDED performance up to 64kb chunks, then it begins to drop off again. Of course you will only appreciate that if you have a fast operating system on another drive or array. Single drive users will see no real improvement. I can still see write performance is lower on the dynamic array by copying a 2.61 GB file folder from one array to the other. The external array which has been performing well can copy that file from the dynamic array in 1 min 35 seconds. The dynamic array can copy the file from the external array in 2 min 5 seconds. However considering that the external array has 4 drives on dual channels that may not be a bad result at all.

Time to go to bed and be happy for what I have figured out so far. I still want to try Win2k to see if a substantial difference can be wrought from that avenue. If it isnt an XP specific problem then that should be identified.

 

[Thor86]

XP Suxors.

This just makes me happier not to upgrade to XP and stick with Win98SE/Win2k setup.