I was migrating arrays the other day, and I thought Iíd play with the SiI3114 and 4 80GB Hitachi 7k250s. I had a 2 disk raid 0 array and I planned on migrating to a 4 disk raid 10 array. I made a number of arrays and ran benchmarks on each. I wanted to see the capabilities of this controller with the 4 Hitachis.

The SiI3114 is integrated onto the DK8X and connected via a PCI-X bus, so it should be running at 66 MHz on this board.

Testing Individual Disks
Each of the 4 drives produced similar HDTach & ATTO benchmarks, like these, so all disks and controller channels seem to be operating fine.
ATTO single disk

HDTach single disk

These are 2 disk raid 0 scores:
ATTO 2 Disk Raid 0

HDTach 2 DiskRaid 0

These ATTO scores were taken simultaneously on 2 2-Disk Raid 0 arrays:
ATTO Set 1

ATTO Set 2

And this is a HDTach score of one of these arrays (both scores were similar, could not run simultaneously).
HDTach 2Disk Raid 0
DisturbingÖ.

These are 3 disk raid 0 scores:
ATTO 3 Disk Raid 0

HDTach 3 Disk Raid 0

And these are 4 disk raid 0 scores:
ATTO 4 Disk Raid 0

HDTach 4 Disk Raid 0

The 4 disk HDTach score is disturbing, to say the least.

These are Raid 10 scores:
ATTO Raid 10

HDTach Raid 10

Again these drops in the HDTach scores are disturbing.

At this point I didnít know whether to question the HDTach results or question the drives/controller. The ATTO scores seemed OK, but HDTach was giving some weird results with all 4 drives in a raid setup. So I decided to test these setups with some of our own software. It was configured for heavy disk access, using a 100 MB datafile, creating 11 25 MB output files and writing the complete contents to these files, then reading them back. While a lot of things are going on during one of these runs, it does stress HDD access.

This was the result with a 3 disk raid 0 array:
FlamMap 3 Disk Raid 0

This was the result with a 4 disk raid 0 array:
FlamMap 4 Disk Raid 0

This was the result with a 4 disk raid 10 array:
FlamMap Raid 10

These runs confirm what HDTach is saying, and that thereís something wrong with the 4 disk raid 0 configuration. Iíve played with different stripe sizes and block sizes, switching channels/disks, everything I could think of. The results were always the same.

So either I have something configured incorrectly, or there are some serious limitations with the SiI3114 and 4 drives and striping. Although the ATTO scores look OK, the HDTach scores indicate a problem and our in-house software bears it out. I was disappointed in the 4 disk raid 0 performance, and expected better after having fairly good results with the 2 and 3 disk arrays.

On another note, I had trouble ghosting images to the 4 disk arrays. I could ghost images to 2 disk arrays no problem. All of the benchmarks above were done with the OS installed on an IDE drive.

If anybody has any thoughts on why this controller is acting like this, or how I might be able to remedy the situation, Iím all ears.