In theory the read/write performance increase by using RAID0 could be as high as 100% (times the number of extra drives!). In practice, you're not using an OS designed to do this, nor a filesystem layout that allows for it, nor writing in the exactly correct sized blocks to allow it to happen. You'll get an increase based on the greater number of read and write heads available during any given pass, but nowhere near that much. The increase will rise as the number of disks rises (each disk increases the number of heads available for a single read or write operation by "n*h"). The 10% figure probably isn't far off for a 2 disk setup.
Also, it's important to be aware that when dealing with more then 2 disks in a RAID config, RAID5 provides "n-1" performance increase due to stripping (same performance increase of a RAID0 with one less drive), and it's a hell of a lot safer. This is pretty much always preferable (and why I recommend using larger multi-disk arrays if you're going to use RAID at all).
Honestly, in a stock MB with 2 drive connections and an onboard RAID0/1 driver, I'm still hesitant to suggest using the RAID. Perhaps Kao has a very specific set of things he's doing that benefit from it. However, I honestly believe that in a two disk system, you're going to usually see about the same performance by simply designating one disk to be your system disk and one to be a data disk. That way, OS operations that are always going on in the background and reading and writing to disk will all occur on one drive, while data operations with whatever games and applications you're using will occur on the other. This is doubly true of SCSI or SATA drives (not so much of IDE since the controller itself gets bogged down).
I'm sure Kao can improve performance doing that. And maybe it's just that important for him to do it. But for most home users? It's just not a great idea.
Oh. And if you *really* want to maximize performance, what you want to do is format your drives to a smaller size then their maximum. In the unix world, you need to use some creative uses of newfs and tunefs (not sure how to do this in windows or even if NTFS benefits from it), but what you do is take a large drive and format it to be much much smaller then it is, and make sure that the actual formated space on the drive is all along the outside of the cylinder. Then stripe a RAID0 (or RAID5) array set up this way. The idea is that by making say a 150GB disk into a 20GB disk, you've dramatically shortened the amount of distance any head will have to travel to perform a read or write operation. Every seek operation requires that the head reset to zero (outside the cylinder), then seek inward to the correct distance, then wait for a pass to figure out where it is rotationally, then begin reading or writing the correct amount of time later along the next rotation of the disk. You can literally cut seek times by 300-500% if you shrink the size of the formated area on the disk.
I knew some guys who used to do that back in the day for DB operations. They wanted "fast", not "big". Worked amazingly well, but cost a boatload since they were using 5x as much disk as they actually needed. But if you want fast...
____________________________
King Nobby wrote:
More words please