1. Diskless server installation

At the moment the installation of the diskless server basically follows the method descibed in the RHEL System Administration Guide for creation of an NFS based diskless client: install the OS on a local disk and then copy its contents to the network disk.

1.1. Install and configure packages for iSCSI boot

Install the mkinitrd-iscsi package from the RHEL3-local-extras repository. This will install its depency packages too:

# yum install mkinitrd-iscsi

Configure mkinitrd for iscsi boot:

# echo "BOOTMETHOD=iscsi" >> /etc/sysconfig/kernel

Configure grubby-iscsi by editting /etc/sysconfig/iscsi-boot:

# local
VERBOSE=1
LOCALBOOTDIR=/boot/iscsi
KERNELARGS="ETHERNET=eth0,eth1 BOND=bond0 RAID=yes"

# remote
BOOTSERVER=10.20.1.2
REMOTEBOOTDIR=/tftpboot/linux-install/$(hostname -s)
REMOTEUSER=tftpboot

# default sync commmand
SYNCCOMMAND="rsync -aH --rsh=ssh ${LOCALBOOTDIR}/ ${REMOTEUSER}@${BOOTSERVER}:${REMOTEBOOTDIR}"

	Note
	The parameters `BOOTSERVER`, `REMOTEBOOTDIR` and `REMOTEUSER` are not directly used by grubby-iscsi, their only purpose is to improve the readability of the `SYNCCOMMAND`.

	Important
	In the example config grubby-iscsi is instructed to sync the boot setup to the boot server with ssh. In this case you will need to setup ssh public key authentication to allow the rsync command to write in the `REMOTEBOOTDIR` without a password.

1.2. Adjust kernel module parameters

Remove all 'scsi_hostadapter' references from /etc/modules and if you want to bond ethernet devices configure the bonding settings. The resulting /etc/modules should look similar to this:

alias eth0 tg3
alias eth1 tg3
alias eth2 tg3
alias eth3 tg3
alias usb-controller usb-uhci
alias usb-controller1 ehci-hcd
options bonding miimon=100 mode=1

1.3. Adjust init scripts

The default init scripts interfere with the iSCSI services and therefore need to be tweaked a bit:

remove the ethernet configuration for the NICs in the storage network: These NICs already have been configured by the initrd and must not be brought up or down or reconfigured later in the startup sequence or the shutdown sequence. In short 'rm /etc/sysconfig/network-scripts/ifcfg-ethX' for the appropiate devices.
/etc/init.d/halt calls killall5. This tool kills all processes except for the current session. Unfortunately this also includes the iSCSI kernel threads. Therefore remove or comment the lines calling killall5.
/etc/init.d/halt calls /sbin/halt with argument '-i'. This argument means ' Shut down all network interfaces just before halt or reboot', which interferes with iSCSI. Remove the offending argument from HALTARGS.

1.4. Setup iSCSI to access the network disks

The config file /etc/iscsi.conf needs to contain the discovery addresses of the SANs and the targets. Technically explicit configuration of which targets to enable is not mandatory but it is safer to do so: else a configuration error on the SAN could mix up targets in a nasty way (e.g. a mixture of snapshots could be mounted).

The configuration should look like this:

#HeaderDigest=Prefer-off
#DataDigest=Prefer-off
#LoginTimeout=15
#AuthTimeout=45
#IdleTimeout=60
# don't do IdleTimeouts
IdleTimeout=0
#PingTimeout=5
#ReplacementTimeout=30
DiskCommandTimeout=15
#AbortTimeout=10
#ResetTimeout=30
#ImmediateData=Yes
Enabled=no
DiscoveryAddress=10.20.1.1
TargetName=iqn.1999-12.net.enovation:storage.esan11.appserver1-root.part0
  Enabled=yes
DiscoveryAddress=10.30.1.1
TargetName=iqn.1999-12.net.enovation:storage.esan21.appserver1-root.part1
  Enabled=yes

Parameter IdleTimeout should be set to 0. DiskCommandTimeout should not be set to 0 or else the RAID1 devices will not degrade if a SAN fails or becomes unreachable.

1.5. Partition the disks

Partition the disks of both targets:

# parted /dev/sdX mklabel msdos
# parted /dev/sdX mkpart primary 0 root-partition-size
# parted /dev/sdX set 1 raid on

Optionally create partitions for data partitions and swap.

1.6. Create the RAID1 device(s)

Create RAID1 devices out of the partition(s):

mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sd[bc]1

1.7. Copy the filesystem

Now create and mount the iSCSI root filesystem:

# mkfs -t ext3 /dev/md0
# e2label /dev/md0 /
# mkdir /mnt/iscsi
# mount /dev/md0 /mnt/iscsi

And finally copy the root filesystem to iSCSI:

# rsync -axH / /mnt/iscsi

1.8. Create initial ramdisk

The boot server needs to have the boot configuration (pxelinux.cfg) and kernel and initrd images. mkinitrd-iscsi will ensure that upon every kernel update the boot server will be updated too. For the install we will fake a kernel update with:

# new-kernel-pkg --mkinitrd --depmod --install `uname -r`

1.9. Boot server configuration

Finally, configure the machine on the boot server with iscsi-boot-admin. For example:

# iscsi-boot-admin add-server machine123 00:11:22:33:44:55,66:77:88:99:AA:BB "123 room 1.5.1 rack 12"
# iscsi-boot-admin set-active machine123 appserver1
# iscsi-boot-activate

See the documentation of iscsi-boot-admin for an explanation of these commands and instructions on how to setup the boot server.