On my servers, speed is not the primary concern, it is reliability.
Sometimes the kernel will go crazy (as it did when running the 3c59x code
included in newer 2.0.3x kernels), and it takes long periods of time to
check these hard drives for errors (about 30 minutes to go through each hard
drive) and even then, sometimes the hard drive needs to be checked in single
user mode.
ACL support is also important feature which I'd like to se implemented.
I think the worst part of a kernel crash is that when something big crashes,
it is really hard to copy down the darn oops onto paper :)
Jordan
-- Jordan Mendelson : www.wserv.com/~jordy/ Web Services, Inc. : www.wserv.com