Sander said:
Eh? RAID 1 won't make your system much faster,
A little clarification for Finesse's and calculon's benefit here: RAID 1 can make reading substantially faster, because the reading operation will be conducted on the hard drive which requires minimum head repositioning to access the appropriate track. Writing operations may actually be
slower with RAID 1, because data needs to be written
twice - once to each hard drive. Nonetheless, I recommend RAID 1 setup for a system which is used to store a lot of large files, such as movies, MP3s, CD images etc.
that would be RAID 0, and it won't make your system *much* faster, that depends on whether the harddisk is the bottleneck or not.
A RAID 0 setup of N drives is N times as fast as a single hard drive (in an ideal case). As Sander pointed out, this performance boost will be noticeable only in situations when HDD operations are so intensive that hard drive bottlenecks the system. It probably won't affect game performance. Also note that a RAID 0 setup is unsafe as hell, and its safety decreases with each additional hard drive. If you have a RAID 0 array with N hard drives, each with MTTF (mean-time to failure) factor T, then the MTTF of the entire array is T/N. So, if you have a hard drive which is expected to give up the ghost in 5 years, a RAID 0 array of five such hard drives is expected to malfunction in only
one year! Worst of all, in a RAID 0 system all data is distributed on all hard drives and there is *no* redundancy (makes you wonder why it's called "RAID" at all), so when one hard drive breaks, you lose
all your data! Simply put, RAID 0 is an extremely unsafe array and I wouldn't recommend it to anyone.
If you absolutely
must use RAID for some reason, then go with RAID 5, because it combines performance increase of RAID 0 with security of RAID 1. Note that RAID 5 isn't as secure as RAID 1, but it is much more space-efficient, because instead of simply storing all data in duplicate, it uses one or more special parity drives for error detection and correction. Since Hamming code is used, one error per cluster can be corrected without data loss.