>
> With cache taken out of the picture, on c) I'm seeing about 25MB/s for large
> writes, and 115MB/s for reads. That's sufficient for me. All drives are
> P-IDE, and on their own controllers.

unless small files and/or random in-file writes are rare it's very good.

i don't think performance difference between raidframe/vinum and expensive 
hardware raid is really big.

i've seen some tests with FreeBSD that showed opposite conclusion, 
unfortunately i don't remember link.

NetBSD's MAXPHYS=64kB (and which can't be set higher) is probably the 
largest problem in performance.


if LFS will get really stable some day, NetBSD software RAID will be even 
better because one can actually know the stripe boundaries.

>
> On b) I've been through 3 disk failures (over 4 1/2 years). Since I'm

still much i had only one in 4 years.

> using a non-hot-swap interface, that does mean a reboot - but it's not
> an issue for my set of requirements.

for most it isn't too.

people should compare this with how much things like

1) software failure (bug)
2) other hardware failure like error in memory or in bus transmission
3) human failure of any type.

occures in 4,5 years with 3 systems. i'm sure at least 4 by average.

and for all of them RAID gives no real protection. unfortunately - it 
gives people feeling of being well protected from data loss