raid 0 setup

[Sam Vilain]

On Tue, 02 Dec 2003 07:59, Joachim Schmitz wrote;

  > will I be able to boot from this setup ?

No, you need to incorporate either special LILO setup (see the
raid-extra-boot option, described in lilo.conf(5)) or use the GRUB
setup script that I posted to the list a while back.

  > and why is there hda and hdc and not hdb ?

`hdb' is the `slave' on the first IDE channel.  `hdc' is the `master'
on the second IDE channel.

Generally it is best not to put more than one device on a single IDE
channel.  If you are going for high availability that mirroring helps
provide, then you may as well take advantage of the extra IDE channels
that most motherboards come with.

What follows may seem like FUD but it is my experience.  Take it with
however many grains of salt you wish.

When one IDE disk fails, in my experience brings down the entire
channel, if not the one next to it as well.  I actually tested this
for one customer who required resilient disk mirroring; we loaded a PC
(well, an Ultra 5, but that's basically a PC-class motherboard with a
Sparc CPU) with two IDE disks, one on each controller, and built it
with mirroring via Sun's Online Disk Suite.  We tested the strength of
the mirroring by disconnecting the cable from one of the drives while
the system was running (but the drives were quiet).  Under Solaris the
OS was still there, but suffering badly from the fact that *both* IDE
controllers had gone into an error state; even though only one disk on
one channel had been disconnected.  In the end we had to ship the
system with a SCSI controller and external disk pack to be sure.

On normal PCs running Linux the picture in my experience has been even
worse.  IDE drives don't seem to just fail, they go on the blink and
then for whatever reason Linux starts spending 90% of its CPU time
nursing the faulty device, or whatever it's doing.

Also, there is some fundamentally, ludicrously crippling limitation of
ATAPI, to do with the number of commands that can be active at once (I
think it's a whopping total of one) that means that the two devices
are not just contending for bandwidth, but you can only ask one of
them at a time to do anything.  Look at all the `high-end' (for want
of a better term) IDE RAID controllers - these use a single channel
for each disk.  Look at the fact that on-the-fly copying between CD
devices just doesn't work if they're on the same channel.

I don't know the exact truth behind why IDE performance is so bad -
I've not been able/bothered to find the specs to spot the flaw, but
all I know is that even on a fairly new, well supported IDE
motherboard, the performance always bites compared to even an old
(PCI-class) SCSI controller.

Apparently SATA is supposed to fix these problems, but until I see a
SATA system perform with the reliability, throughput and host
efficiency of a decent SCSI controller, I'm going to continue to avoid
ATAPI like the plague.

Sorry for the rant :-).
-- 
Sam Vilain, sam at vilain.net

A closed mouth says nothing wrong; a closed mind does nothing right.
 - anon.