Sign in with
Sign up | Sign in
Your question

Low transfer rate on RAID-0 * 2

Last response: in Storage
Share
September 3, 2002 9:03:47 PM

I think I've tried everything but something is still wrong.
My System;
MB: ASUS A7V33 w Promise onboard RAID controller
HD: 2 x Western Digital WD800JB SE (stripe/NTFS)
OS: Win2k

Transfer rates in SiSoft Sandra is never higher than 26 mb/s. Raid controller reports disk running in UDMA5. I've tried to increase the Latency Timer for the RAID controller but with no success. Also changed all ATA device drivers to latest VIA drivers.
Could the reason be that the strping block size (64kb) combined with NTFS cluster size (4kb) makes a very bad combination?
Should I change to FAT32?
Could it be any other PCI-unit that slows down my transfer rate?
Is the SiSoft Sandra reliable or is there a better benchmark?

According to what I've seen in the Storage Guide I should expect rates of around 2 times a single disk and with my WD´s with 8mb buffer it should be around 40 mb/s in Sandra.

I would appreciate any answers with ideas of the reason.
Thanx!

More about : low transfer rate raid

September 5, 2002 3:48:08 AM

For some reason Promise RAID controllers and SiSoft Sandra don't click, so I wouldn't take those results seriously at all. Sandra couldn't even benchmark my RAID setup it kept telling me I didn't have any hardrives in my machine. Are you actually having noticable problems or just basing your inquiries on what Sandras Info gave you, if its just on SI Softwares Sandras info just disregard it and go on. Benchmarking programs are designed to push your machine to the max, if you run it repeatedly you can actually damage a perfectly good piece of hardware, trust me I found that out the hard way, read their disclaimers. Ryan

Details, Details, Its all in the Details, If you need help, Don't leave out the Details.
September 5, 2002 5:09:57 AM

Quote:
Could the reason be that the strping block size (64kb) combined with NTFS cluster size (4kb) makes a very bad combination?

Yes, this is a very bad combination. You can read a lot more about this here <A HREF="http://www.amdforums.com/showthread.php?s=a478935fa97e4..." target="_new">AMDForum thread on cluster & stripe size</A>
Quote:
Should I change to FAT32?

I have never actually measured this myself, but FAT32 is supposed to have a 3-5% performance increase compared to NTFS. But you will get much more from a proper cluster and stripe size. With my WD drives and a Highpoint controller I use stripe=16k and cluster=16k. Sisoft benchmarks these to 90 MB/s (a score of around 50000)
Related resources
September 5, 2002 1:03:54 PM

Strange. My setup includes 2 IBM 120GXP drives at Raid 0 and my Sisoft Sandra scores are 47000+. With the Maxtor D740's it was 28000. But I wouldnt trust Sisoft Sandra,
you can defrag your drive and that has a large affect on the Sisoft scores. Pretty stupid test software I reckon! Use HDTach.By the way Im usind the onboard promise chip on my MSI KT7 KT333 mainboard.

Computers remind me of Murphy's law:- whatever can go wrong, it's always associated with a computer!
September 5, 2002 3:12:26 PM

Quote:
You can defrag your drive and that has a large affect on the Sisoft scores. Pretty stupid test software I reckon!

Sisoft measures filesystem performance and hence clustersize and drive fragmentation will have an impact on the result. This is the way it should be, since those factor also affect your everyday usage of the drive.

Quote:
Use HDTach

HDtach measures drive performance without any concern to filesystem and clustersize. Thats why HDTach only can make write benchmarks on an empty non-pertitioned drive. If you want to make comparative benchmarks with different clustersizes and/or stripe size do NOT use HD Tach.
September 5, 2002 8:39:07 PM

Ok....
I will install Win98 on my other HD and boot from that one first....just to see if there i some fussy things with Win2K (i now there was before servicepack 2). If that doesn't help i will try to use diffrent cluster size/block size.

Actually I don't have any problems and the speed of my disks is enough right now. Just that I bought thed HD's and the MD last week and want to verify that it's that fast as i want it.

Another funny thing, my old 12Gb Quantum HD is doing 100% better now with new MD, processor & memory. At least according to SiSoft Sandra. Manage to get 16 Mb/s from that old piece.

Thanx for all help.... I will be back with my results and my conclusion. I will solve this in some way...
September 6, 2002 2:39:27 AM

yeah. dont trust sandra for HDD benchies.
even how full your drive is can effect the results ALOT.
example. 80Gb D740x maxtor drive.
empty it scores somewhere above the 30gb ata100 mark
half full it only scores halfway between ata66 and ata100. at the moment around 2/3 full and somewhat fragmented it scores below ata66.

hardly reliable.

<b>Trust Me, Trust Yourself, Trust the Hamster. Anyone else - Shoot Them. :eek:  </b> This sig was proudly brought to you by <i>Hammie</i> :smile:
September 6, 2002 3:08:22 AM

You can download hdtach, and see what it tellz you. Or you can run the nero "speed check" and get a max for the drive/partition.

Hdtach is a little more accurate, as it gives your max/min/avg and a nifty little chart to see where your speed starts really dropping off.

P@ll4dium f0r L1f3!
September 7, 2002 4:57:49 AM

Actually they both suck ( sis sandra and HDD tach). Try using the disk inspection test on Ziff Davis's media bench. Intel's IO meter is better but a bit complicated to use.

It's not what they tell you, its what they don't tell you!
!