Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

pardon me if i'm mistaken

but are you guys referring to the system files that windows uses for the scsi control?

it was stated somewhere (not sure if it was here) that winXP protects certain system files, and re/moving or overwriting does nothing as winXP will just put back the original files

sorry if i'm off topic :D

Share this post


Link to post
Share on other sites

That's exactly what we're doing allnighte. What we've done is worked a way around XP's protection of the system files. It makes no difference though as far as we can tell.

Share this post


Link to post
Share on other sites

So it sounds like all we have to go on was that one MaximumPC article in September (?) where a Microsoft guy said they expected to release a fix for the XP-SCSI problem within about 3 months from that time?

I've been assuming that he was NOT talking about the non-fix that came out in SP1 (referred to in the KB article referenced above) because it came out a relatively short time after said article. I have been expecting something closer to the end of the year. Doesn't anyone who reads SR have an ear closer to MS to give us a scoop on this?

Share this post


Link to post
Share on other sites
Guest Eugene
Doesn't anyone who reads SR have an ear closer to MS to give us a scoop on this?

There are a few. They've been ignored.

Share this post


Link to post
Share on other sites

Shucks, that implies that I haven't paid enough attention to them myself... or that I'm not experienced/well-informed enough to know who they are. Care to be more specific?

I'm well aware of a good number of industry experts here and I read everything they post with great appetite -- but I don't recall anyone saying "here's the word from somewhere close to Microsoft about what is being done about this." I guess it just was more subtly delivered than that. I wish I had a spare 6 hours to comb every post and find it.

I'm also aware of the "it's not a real problem" theorists but I assume that's not what you're referring to.

Okay, I'll just keep my eyes open and keep learning all I can.

Share this post


Link to post
Share on other sites

Pardon my ignorance, but I have some questions. This is my first post in this lenghty thread.(Didn't read the whole thing)

1-Does the file system make a difference at all?(Fat32 vs NTFS)

2-So, win2000, 98, ME, and NT's had no problem with SCSI performance?

Share this post


Link to post
Share on other sites

A quick, poorly explained and probably slightly inaccurate answer:

1-Does the file system make a difference at all?(Fat32 vs NTFS)

Not to this problem, though Microsoft fixed a seperate NTFS issue in SP1.

2-So, win2000, 98, ME, and NT's had no problem with SCSI performance?

Win2000 acquired the problem in SP3, but W2K SP2 or earlier, and other non-XP OSs do not suffer from this problem.

My understanding of the problem is, that it's actually a sort of bug (a no-cache flag being ignored) that MS discovered in their SCSI implementation and fixed in XP, and then in W2K with SP3. The fix causes this poor (but technically more reliable) performance. So they don't want to just undo the fixes they made.

It sounds like they (MS) also issued incorrect advice about the flag settings for best performance with SCSI, and lots of program vendors used it. Even if they tested it, there was little performance loss because the bug was ignoring the flag. Then when the bug was fixed, all these programs were suddenly giving poor performance. Even Windows Explorer seems to have this problem.

IIRC, cas wrote a simple copy program (cascopy) that set the relevant flag correctly, and this seemed to work much better. So it's possible to fix the problem, if every program gets patched. But that won't happen, so everyone's waiting for MS to come up with a miraculous solution. They're working on it, but can't say if/when they'll fix it. No idea why even Explorer doesn't work as it should, though.

It's more complicated than that, but those are the basics of the problem. I'm not an expert, I just read and (mostly) understood what's on these forums. Hope this is useful for someone.

I also have a question:

Despite the reduced write performance, is a good SCSI drive still much faster in XP than a good IDE drive for general Windows use and gaming?

Share this post


Link to post
Share on other sites
I also have a question:

Despite the reduced write performance, is a good SCSI drive still much faster in XP than a good IDE drive for general Windows use and gaming?

DEAR GOD YES

well, for my X-15 36LP that is.

the access times aren't affected, so everything still feels very snappy and quick compared to my IDE drives

Share this post


Link to post
Share on other sites

When I first installed my SCSI drive I got seq read 44MB/s, seq write 16MB/s while my IDE drive performed 35/35. I read somewhere in this huge forum thread about converting to NTFS and dynamic disks. After

doing that I got read 46 and write 45 MB/s which I am happy with for the time being (measured with Sandra Prof benchmark).

I have not had time to follow the whole thread so I might repeat what others have said. but I still want to share my experience with you. :D

Share this post


Link to post
Share on other sites
Where can I download that cascopy? I'd like to try it in XP.

The original uuencoded version of cascopy can be found here.

My comments on this issue can be found here.

Although Spod suggests that cascopy may yield similar results on w2k and xp, to my knowledge, no one has posted their results.

Please, use a very large file, which may be created using a utility found in the same thread as cascopy. Also, whatever you can do to ensure a proper apples to apples comparison would be appreciated.

Share this post


Link to post
Share on other sites

Well, what the heck does this mean then? I got the cascopy program and ran it on an 800 MB file.

Now, my story is that in XP Pro my IDE is aprox twice as fast as my 15K Cheetah when doing a same disk copy. It's embarrasing for the SCSI drive.

What I was expecting was this thing to fly when the -writecahe option was on, right?? On thing I noticed was my drive got really loud when I used that option.

27,596,379

4,628,519 (with -writecache)

??????

Share this post


Link to post
Share on other sites

If you read the cascopy thread, it should shed some light on why cascopy often performs better than the built in copy routine (which is what –withcache uses).

I am a bit confused though. You say that your ATA drive is twice as fast as your SCSI drive, but that is not borne out by your numbers. A single disk copy speed of 27.5MB/s is very good. Remember that you must read and write to complete a copy. Of course, a disk to disk copy should be more than twice as fast, assuming sufficient IO bandwidth.

What we need now, are your numbers from win2k.

Share this post


Link to post
Share on other sites

Ah, sorry, I incorrectly assued the option was to enable the option which I thought XP was ignoring. The cascopy program definately shows the kind of performance I expect from the SCSI drive. Fantastic demonstration of this xp flaw, thanks.

I can provide Win2000 scores as well, but I'll need to partition my IDE drive and install it :-). Will post later.

Share this post


Link to post
Share on other sites
Guest Eugene
Fantastic demonstration of this xp flaw, thanks.

Has a flaw actually been demonstrated yet? Unless I missed something I believe cas is pressing so hard for you to run it on 2000 to compare or contrast it with the xp results and thus confirm or deny the alleged "xp flaw."

Share this post


Link to post
Share on other sites

Exactly right.

A significantly better score from w2k indicates a problem. Although a similar score doesn’t rule out the possibility of inefficiencies in SCSIPORT or the class driver, it suggests that they are not as serious as some have claimed.

In this case, additional cascopy controls might help in narrowing down the issue.

Share this post


Link to post
Share on other sites
Guest Eugene
Exactly right. 

A significantly better score from w2k indicates a problem.  Although a similar score doesn’t rule out the possibility of inefficiencies in SCSIPORT or the class driver, it suggests that they are not as serious as some have claimed.

In this case, additional cascopy controls might help in narrowing down the issue.

Alright, I bit the bullet and did what I should have done a long time ago when cas was willing to devote his time the SR community by programming his copy utility.

The following tests were run utilizing cascopy.exe in default mode copying a file 674.496,512 bytes in size from drive F: to drive G: partitioned to maximum sizes utilizing NTFS. Figures are an average of three trials which all presented reasonably precise results:

DM+9 200 GB to DM+9 200 GB, WinXP: 46.8 MB/sec

DM+9 200 GB to DM+9 200 GB, Win2k: 44.6 MB/sec

Delta moving to 2k from XP: -4.7%

X15-36LP 36 GB to X15-36LP 36 GB, WinXP: 45.9 MB/sec

X15-36LP 36 GB to X15-36LP 36 GB, Win2k: 47.2 MB/sec

Delta moving to 2k from XP: +2.8%

These figures were drawn utilizing all standard SR Testbed3 controls. Can anyone imagine a more controlled testing environment? :)

Serious food for thought for many of this overly long thread's participants.

Share this post


Link to post
Share on other sites

Eugene, thanks for taking the time to do that.

Kudos to logicprobe and cas 2 and everyone else too, of course.

But which SP of 2K and XP were you using?

Are we to assume it was 2K SP2 vs. XP SP1? or SP3 vs SP1? or...?

Share this post


Link to post
Share on other sites

Eugene, if possible could you please also do a test copying a file to another directory on the same drive? I've found that my X15-36LP takes nearly 40 secs longer to do this than my WD1200AB (1GB .vob). This is under winXP SP1. It's not possible for me to try it with win2k at the moment.

Thread with results is here:

http://forums.storagereview.net/viewtopic.php?t=6983

Share this post


Link to post
Share on other sites
Guest Eugene
But which SP of 2K and XP were you using?

Are we to assume it was 2K SP2 vs. XP SP1?  or SP3 vs SP1?  or...?

this test was run under TB3's default XP installation (which does not include SP1) and under Win2k with no service pack. As long as what cas has said about cascopy is true, these results will hold up in both operating systems regardless of what combination of SPs are used.

Eugene, if possible could you please also do a test copying a file to another directory on the same drive? I've found that my X15-36LP takes nearly 40 secs longer to do this than my WD1200AB (1GB .vob). This is under winXP SP1. It's not possible for me to try it with win2k at the moment.

I may be able to try that test when I have some time. However, its important to understand that the results, whatever they may be, are irrelevant to the supposed xp flaw. Intra-drive file copies rely heavily on caching strategies found at the drive as well as OS level and can vary significantly when contrasted with typical application level performance. Intra-drive file copies (as well as inter-drive ones for that matter) are not good benchmarks to run on drives and do not measure anything except how well a drive copies file to itself. I've been trying to explain this for over four years now to readers who insist on its use as a "real world" benchmark. Back in 1998, for example, the DiamondMax 2880 was faster than the Cheetah 9LP when it came to file copies. This was not borne out in WinBench99 nor in subjective use of typical applications on both drives.

Share this post


Link to post
Share on other sites

So, if I can get my puny little brain round this...

Eugene's tests suggest (prove?) that all XP needs is patched/better written applications (including Windows Explorer). The SCSI subsystem is fine, it's the way all the programs use it that is at fault. So they just need to patch their own programs to use the flag correctly, and tell everyone else who writes programs for XP to do the same.

Is it that simple?

If so, MS and Windows programmers everywhere, sit up, take notice and GET ON WITH IT!

Please?

Share this post


Link to post
Share on other sites

SPOD - great summary of the very question I have. If it's true that the programmers at the biggest software company in the world can't write their software to do a basic file copy without taking a 50% performance hit with their latest flagship OS, I'm sorry but WHY are they making so much money again????

Maybe Norton can market an add-on product to allow XP users to "improve performance by 100%" and charge yet more money.....

Share this post


Link to post
Share on other sites
Guest Eugene
Eugene's tests suggest (prove?) that all XP needs is patched/better written applications (including Windows Explorer). The SCSI subsystem is fine, it's the way all the programs use it that is at fault. So they just need to patch their own programs to use the flag correctly, and tell everyone else who writes programs for XP to do the same.

No... what they indicate is that XP explorer has a patched copier (like that of 2k's SP3) and copies files with the integrity that Microsoft originally intended. Applications have nothing to do with it. The vast majority write to the hard disk at the same speeds in XP or 2K.

If so, MS and Windows programmers everywhere, sit up, take notice and 

GET ON WITH IT!

Please?

Actually, what this means is that hardware enthusiasts here at SR and everywhere else must sit up, take notice, and get on with it. There may be reasons not to use XP, but poor SCSI performance is not one of them.

Share this post


Link to post
Share on other sites

I have the next configuration - VIA KT333 chipset on Gigabyte 7VRA MB. I have also Adaptec 19160 SCSI adapter (310 BIOS, tried latest & previous Adaptec drivers) with Seagate 15K 18Gb ST318452LW drive(15% full) attached to 16bit wide SCSI bus and IDE IBM 7200 80Gb 40Gb-platter “Vanquver” (80% full) drive, 2x512MB DDR 3000CL=2 Corsair memory, Athlon 2100XP CPU, RadeOn 8500 video. IOMEGA ZIP drive and TEAC CD-RW attached to narrow (8bit) bus of the same SCSI adapter.

The same two WInXP pro SP1 systems installed on bootable SCSI and IDE disks. Choosing of boot need to be made from BIOS. At first SCSI drive was NTFS formatted and IDE – FAT32. Now both drives have FAT32 partition. SCSI adapter is in PCI2 slot with separate IRQ. In adapter’s BIOS: “domain validation” – dis., “write back cache” – en. for devices “00” and “07”, “parity” – en., “bus scan” – en. only for connected devices. In OS “enable write cache” is checked for both drives. I have installed after SP1 last VIA 4in1 drivers, 020B latency patch and last VIA raid patch.

Both disks – basic, don’t want to go for the dynamic. And I have system swapping – disabled. Of course we need to reboot after each file transfer to clear memory and swap file (tweak in registry, but better to switch off system swapping). So detailed description made to eliminate further questions – it’s terribly hard for me to write in English.

ATTO data close to every one results for basics disks – “writes” ~ 13.5Mb/s and “reads” ~ 58.5Mb/s

Tried to change SP1 WinXP scsiport.sys file with the same one from SP2 W2K but receive the same results.

Want to discuss some “real life” tests. Copying (using FAR file manager) 1Gb (1073741824bite) DVD.vob file from C:! to C:, D:! to D:, C: to D: and D: to C: - booting in turn from SCSI (C:) and IDE (D:).

1. SCSI boot.

C:! to C:=3m04s (1024M/184s~5.6Mb/s)

D:! to D:=2m25s (~7.1Mb/s)

D: to C: = 23s (~44.5Mb/s)

C: to D: = 20s (~51.2Mb/s)

2. EIDE boot.

C:! to C:=3m02s (~5.6Mb/s)

D:! to D:=1m59s (~8.6Mb/s)

D: to C: =39s (~26.3Mb/s)

C: to D: =46s (~22.3Mb/s)

So what we can deduce from this don’t taking into account memory paging, swaping (have LargeSystemCache – disabled). We have terrible read-write SCSI performance ONLY when copying the file on SCSI disk (disk-to-self copy).I think throughput ought to be near ~ 14Mb/s for boot SCSI drive and near ~ 11mb/s for non booting SCSI drive. And we don’t see any drops in read-write SCSI performance under WinXPpro for disk to disk file transfer data independent of SCSI or IDE boot. The other peoples file transfer results from one SCSI 15K disk to another SCSI 15K disk attached to two separate same SCSI adapters, to two independent channels of one adapter or in worst case to one channel of adapter don’t have any performance hits. So some program block for disk-to-self copy management in WinXP OS have flaw with SCSI file transfer. And we will need only to copy 1Gb file from some folder to the root of SCSI drive to know about if MS applied some fix to solve this problem or not. And if we will have time of file copying more then approximately ~ 1.5min for 15K disk – problem not fixed. We need also to try the same test under W2000 SP2 where peoples feel themselves good testing with ATTO.

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