Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

So at last I made tests with W2000 and DOS7. W2K installed without SP, but with all latest chipset, AGP and devices drivers. DOS7(W98SE-command.com) installed on bootable diskette with himem.sys, Emm386.exe, smartdrive.exe /X /L 2048 2048 enabled in config.sys and autoexec.bat, the file manager - Volkov Commander (analog of Norton Commander) that include very general drivers for SCSI and IDE. For DOS tests don’t made restarts and defragmentation after each copying. The rest details of my setup you can see above at this page.

ATTO SCSI drive results for W2K after installing it on IDE disk – Writes ~ 40Mb/s(curve with slope) and Reads ~ 56Mb/s. Results don’t change after installing yesterdays rev. of 4in1 VIA’s driver. Nothing changed after installing VIA’s Raid patch. But after installing of GB Latency Patch v.020 beta throughput increase for Writes to ~ 59.3 Mb/s and for Reads to ~ 59.4 Mb/s with straight curve from 4 to 1024Kb. ATTO was used with default parameters except file testing length set to 32Mb. So we come to ideal picture that people have testing performance for SCSI file transfer under W2K before installing SP3( I don’t do anything with different SP’s because don’t have them). Now we ready to look at the “real life” test results with 1Gb dvd.vob.

To remind, disk C: - SCSI, disk D: - IDE. OS is always on a bootable disk.

WinXPpro Win2000pro DOS

1. SCSI boot. 2a. SCSI boot – not tested, data

C:! to C:=3m04s (~5.6Mb/s) can be better then in case 3.

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

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

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

2. EIDE boot. 3. EIDE boot. 4. Floppy boot.

C:! to C:=3m02s (~5.6Mb/s) C:! to C:=3m14s (~5.3Mb/s) C:! to C:=1m33s (~11.0Mb/s)

D:! to D:=1m59s (~8.6Mb/s) D:! to D:=2m14s (~6.7Mb/s) D:! to D:=2m25s (~7.1Mb/s)

D: to C: =39s (~26.3Mb/s) D: to C: =35s (~29.3Mb/s) D: to C: =68s (~15.1Mb/s)

C: to D: =46s (~22.3Mb/s) C: to D: =55s (~18.6Mb/s) C: to D: =53s (~19.3Mb/s)

For the test of SCSI one disk file transfer (disk-to-self ) anyone can find out that results will be 3 times worse then for disk to disk file transfer under any windows operation systems (understandable – disk’s head cannot simultaneously read and write plus separate time needed for head’s positioning). But for SCSI case under WinXP we have 5 times worse results. And this is a problem. Then in general WinXP make SCSI self disk operation quicker then W2K. But both systems terribly slow. And as additional note IDE performance under W2k worse too in 1.28 time then under WinXP for the same setup.

Compare WinXP “normal” IDE drive disk-to-self and disk-to-disk times of copying we can deduce 3 time difference for IDE boot. Applied the same to SCSI drive we can wait for self-to-disk file transfer (20-23s)*3=~65s for SCSI boot or in every case no less then 40s*3=120s (IDE drive result) for booting OS from IDE drive. But in real we have in 1.5 time worse throughput for SCSI 15K disk compare to IDE 7.2K one. And in DOS case with non specific very general SCSI driver we have throughput 2 times better for disk-to-self SCSI file transfer and 1.55 times better then disk-to-self IDE file transfer. So hardware, Seagate don’t guilty. It is possible that in SCSI case MS sacrifice with performance of “not often utilized operation” self-to-disk BIG file copying (bigger then 1Mb) on NT core + I/O interface in name of security, reliability and ets. But if we have user with only one big SCSI disk it will be for him “very often utilized” class of operations. So all complaints to MS – not to Seagate or Adaptec. For W2k lovers – they can go bravery to WinXPpro or in worst case to SP3.

Share this post


Link to post
Share on other sites

Hi Guys,

Firstly U must forgive my basic understanding (if that) at this issues, but it is something that i am suffering from and have been realy annoyed since i spend a few bob on my SCSI (mini)system, only to find XP, aparently, has a non-BUG related problem

I have a Raid 5 setup, 6 drives, with 1 as hot spair using Seagate Cheetah 18XL 18GB drives connected to a Mylex AccelaRaid 250 (Ultra2Wide SCSI Raid Controller)

Now i understand that MS have rectified a problem with integrity, and have stumbled upon another, system performance using SCSI, however would it not be possible to provide an option to allow users to choose which they would sacrifice. I used to use the exact same setup within Win2k before SP3 and i was happy with the performance, both speed and stability, so surely it would be possible for the MS guys to allow users to sacrifice reliability over performance until MS find a workaround for the problem.

I just did a very basic test, and i know that this does not take into any account all the technical issues that you guys are aware of, and if anyone want to inform me of were my serious downfall is with this comparison, please go ahead, as i welcome all comments to further my understanding of this issue.

But I copied a 584MB Bin file from the Scsi Raid drive to a WD WD200EB (a 20 GB, 5400 IDE drive that i use for backup) and obtained the following time, using Explorer as the means to copy:

WinXP Sp1

SCSI > IDE : 38Sec

IDE > SCSI : 4min 10Sec

Can someone please tell me why the read from the scsi is apparently fine however the write it this bad. Surely a reliable write is only as good as a reliable read, and if so then the read from the scsi should have also been affected.

I am open to all comment, but please be pateint, as i'm a newbie and only trying to help.

Share this post


Link to post
Share on other sites

This is easy - Raid-5 always has very poor write times relative to read times: This is characteristic of raid-5 arrays. Typically a raid-5 array has slower writing times than single drives. Writing to a raid-5 array requires a verify to make sure the data was written correctly. reading does not have this difficulty.

If you need better write times use raid 1+0. there is redundancy and speed. both the reading writing will operate at raid-0 speeds. The problem with raid-1+0 is that it uses more hard drives than raid-5

Share this post


Link to post
Share on other sites

Sorry for the bad view of test’s data.

Don’t have editing here, so need to organize another way and repost my mixed results to make them readable.

WinXPpro

1. SCSI boot.

C:! to C:=3m04s (~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)

Win2000pro

3. SCSI boot – not tested,

data can be better then in case #4.

4. EIDE boot.

C:! to C:=3m14s (~5.3Mb/s)

D:! to D:=2m14s (~6.7Mb/s)

D: to C: =35s (~29.3Mb/s)

C: to D: =55s (~18.6Mb/s)

DOS

5. Floppy boot.

C:! to C:=1m33s (~11.0Mb/s)

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

D: to C: =68s (~15.1Mb/s)

C: to D: =53s (~19.3Mb/s)

Share this post


Link to post
Share on other sites

Please read the first post here.

http://forums.storagereview.net/viewtopic.php?t=5734&

and reference Eugene's comments in several threads:

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

As a follow on to the tests you ran with cascopy, could you also perform the same tests with Explorer? If my understanding is correct, not all configurations have the apparent performance-hit in XP. It would be worth seeing how Explorer performs before we draw any solid conclusions.

Share this post


Link to post
Share on other sites
Guest Eugene
As a follow on to the tests you ran with cascopy, could you also perform the same tests with Explorer?  If my understanding is correct, not all configurations have the apparent performance-hit in XP.  It would be worth seeing how Explorer performs before we draw any solid conclusions.

I have already demonstrated that our testbed suffers from the explorer problem several months ago.

Regards,

Eugene

Share this post


Link to post
Share on other sites

Has anyone had a chance to try the New Adaptec SCSI Service Pack, whether or not that would make a different, or if Adaptec has resolved the issues, I am not sure, but if anyone has please let us know the results.

Here's a link to were i saw it:

www.warp2search.net/article.php?sid=9093

Share this post


Link to post
Share on other sites

fa147, I installed Adaptec's SP4 U160 drivers when they first went "final" around November 16th, and haven't noticed anything better or worse in my system performance. However, I currently have only one hard drive attached to my 29160N, plus a CD burner.

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