The Raptors are good disks. The reason they are good is beyond the sipndle speed. They are in fact SCSI disks with a SATA interface grafted on, much like (most of) the current crop of 7,200rpm SATA are IDE (or PATA if you prefer) with a SATA interface.
The reliability of the drives is much higher, the seek times far higher. These two are much more significant than the spindle speed IMHO. The disks also have other functions which seem to be missing from other disks.
A lot has been written about various simple RAID forms which is mis-information. For example comparing a RAID0 array with different types of disks is misleading (though perhaps makes sense on a cost basis). A stripe utilising two Raptors will be quicker than a single disk for I/O operations. I have run both on my home workstation for some time. Potential data loss doesn’t bother me as I have an additional (non-Raptor) SATA disk for data storage. I wouldn’t touch a software RAID controller where performance is an issue, it also leeches processor performance and lacks caching. Likewise I wouldn’t use a controller for RAID5 unless it was a quality item (someone mentioned hardware controllers being tied to a single motherboard/controller, a decent controller will allow a state back-up which can be transfered to another controller).
Whilst I’m at it RAID1 isn’t just about security, your read times with this array will be significantly quicker a single disk. Write times are, of course, the same. Once the array has degraded you can continue to use the remaining disk until you are able to recreate the array which makes life much easier (you can even just buy two new big disks for the array and transfer the old data if you are lacking a warranty).
Previously I ran a pair of IDE Maxtors in a stripe and then swapped back to operating them as two individual disks. Though the stripe on these disks was great for large write operations it wasn’t much better for general usage.
So what different - well the RAID controller I ran was very cheap, its controller pretty poor and hampered by the interface (PCI). I am also of the opinion that IDE disks are not suited to RAID arrays with their high seek times.
My company has deployed anumber of server platforms utilising large (non-Raptor) based RAID5 arrays and almost exclusively their performance is shocking. Replace their drives with a smaller number of Raptors and the performance increases massively.
I should also add that the failure rate on the SATA disks was aproximately 50%. They would typically fail in batches of 4-6 per server. Even running a hot-spare this was pretty disasterous as you can imagine. We have yet to see a Raptor failure.
This is a very extreme example but does perhaps give an idea of the relative drives reliability under heavy useage.
In short RAID0 isn’t as bad as it’s made out. It is more suited to the Raptors (IMHO) as they offset some of the risks whilst improving apon the advantages. There is a risk, but there’s also a pay-off!
