I have an interesting RAID setup across 2 harddrives on my home computer. I have /tmp and a mount point(/home/machines) for virtual machines (along with a few other mount points). These partitions I consider replaceable, so I put them on RAID 0 for speed and extra space. If a harddrive was to crash, I'd have to spend a lot of time reinstalling things, but I wouldn't lose things I did RAID-1 on such as /home. I can definitely tell a performance increase from having the virtual harddrives in a RAID-0 setup. Booting Windows XP in a virtual machine in 8 seconds is just beautiful.
Now, I'm looking at if I should buy another (identical) harddrive or two for either extending my RAID-0 portion, or turning it into RAID 0+1. For virtual machines, speed is my only concern. Everything in virtual machines is just a reinstall away.
I understand that RAID-0 is the fastest, yet most dangerous RAID. However, at what point does the SATA controller get maxed out and RAID 0 doesn't give a performance benefit? If I don't get a performance benefit, then I'd rather go RAID 0+1. However, if I can get a benefit, I'll go RAID-0 because I'll do anything to make virtual machines faster(except for buy expensive SSDs :P)
So basically, at what point does RAID-0 not give a performance benefit and instead just be a risky practice?
My motherboard has 8 SATA-III ports and the harddrives I use are 500G WD Black, 6 Gbit/s, 64M cache. (which I consider very fast without RAID even). I use the XFS filesystem
What I'm looking for are some general guidelines. Like for IDE drives, I know that basically bandwidth is split between two drives on the same controller. My question comes to a point as to when will I probably saturate the link, and is there any bandwidth sharing for SATA ports like there is for IDE?