wild variation in RAID drive index scores - Help plz?

[cleverhandle]

Seems like lots of people are having RAID benchmark issues lately. This one seems awfully strange though. I started tweaking my rig this morning - its relevant stats:

XP 1700
Soyo Dragon Plus
2x 40GB IBM 60 GXP (RAID 0)
13GB old IBM (not on RAID controller)
2x 256 MB Crucial RAM
Dual-boot Win98/Win XP Home (both on the array, XP on NTFS partition)

Since I was just starting out, I had no unusual BIOS settings - just the optimized defaults with the 133 FSB setting, disabled onboard sound/LAN, disabled COM2 and LPT. My 4-in-1's were up to date, along with video/sound/etc.

Ran various benchmarks - the one that stuck out was the low drive index on Sandra, two runs around 32000, lower than the 36000 reference for dual ATA100's. Didn't run HDTach, since my unregistered version doesn't support NTFS. After poking around the web, I found out about the whole VIA/PCI issue. After applying the PCI latency patch, the index jumped to over 44000. Woohoo! (Probably should've repeated it, but I was getting bored)

Figuring that that problem was solved, I moved on to adjusting memory settings and other minor BIOS tweaks. Did that for a while, ran some memory benchmarks, but didn't repeat the drive bench, since I figured it would be unaffected. Then got around to increasing my FSB - started it out at 145. Ran all the benches again, and found the drive index had dropped to 28000. Okaaay... maybe the drives couldn't handle the FSB. Decreased it a bit... even worse. Back to 133... worse still. Reset the BIOS back to where I started the day... now my drive index is 27800 - more than 4000 points lower than where I started! WTH?? I didn't install any software today apart from the latency patch. The only settings I toyed with were BIOS settings, which I restored... nothing in Windows. Maybe the 44000 was just a fluke, but even so, how are my scores 4000 points lower than my starting point now?

 

[AkumaBao]

I'm starting to think Sandra's scores are like scores in Golf. The lower the better. I do not trust it's HDD bench's at all. It's done the exact same thing to me on two different mobo's with RAID 1. Please don't get discouraged about it, I just don't think they set the program up to test RAID very well.

 

[neuralfx]

yes sandra is not a very good benchmark .. the most accurate disk benchmark i have seen is probably Iometer .. storagereview uses it .. but ya sandra is known for inconsistent and strange scores ..
-neural

 

[cleverhandle]

Okay, thanks for the input. I knew the Sandra bench was a little odd, but I guess I didn't think it was that flaky. So now I've got some Iometer results (heh - not quite as easy to figure out). I tested the array using the medium load workstation configuration as in AT's IDE RAID comparison found here:

http://www.anandtech.com/storage/showdoc.html?i=1491&p=10

My results:
Total I/O's/sec: 317
Total MB/sec: 2.48
Avg. I/O response time: 202 ms

Compared to their reference (Promise controller, 64k stripe):
Total I/O's/sec: 130
Total MB/sec: 1.02
Avg. I/O response time: 245 ms

Good, I guess - a little surprising, though. The results are certainly significantly better. But my impression was that the extra power in my system over the reference shouldn't show up much in a drive bench. The onboard controller in my board is the Promise Fasttrak 100 Lite, which from what I've read is functionally equivalent to using the separate PCI card. My drives should perform a bit better than the old 75GXP's in their reference system, but I wouldn't expect the difference to be that great. Hmm...

For the moment, at least, I'm satisfied. But if any of you are familiar with RAID/Iometer and can help me interpret the results or suggest additional tests, that would be great, too.