If you are as impatient as me you will want to start right away, and read the following sections only when you get stuck. This section describes how to get it all working quickly. More details can be found in following sections.
Unpack and assemble all computers (nodes), switches, and all other hardware. You might have to configure BIOS on all of the nodes. If you only have one video card and one keyboard, you'll have to pull each node apart, connect the video card and keyboard, power the node up, and configure the BIOS. The most common settings which need to be configured are IDE hard disks and halt on keyboard and video errors. If your nodes do not have keyboards or video cards, they must not halt when these are not detected by the BIOS. Connect power to all nodes, and Ethernet cables between the nodes and the switch.
Install Red Hat Linux 5.2 on the server node. Please see section
Installing Red Hat Linux. Make
sure you leave enough space on the root partition for all the client
node NFS-root file systems (up to half a MB per client). Also
remember that all of the client nodes will write their logs to
server's syslogd server, hence /var/log
on the server will
require require additional disk space to store client logs as well as
its own. /var
, /lib
, /bin
, /sbin
,
and /etc
must NOT be a separate mounts for disk-less client
configuration, that is must be installed on the same partition. If
you install any of the above on seperate partition, the sdct
script used to create NFS-root file system template, will not be able
to make appropriate hard links. Make sure all network devices and
file systems are supported. You will need RARP
(CONFIG_INET_RARP
) support for answering RARP requests. RARP
is supported in the kernel shipped with Red Hat Linux 5.2
distribution.
Run the sdct
[section
sdct] script on the server to create a template
directory (normally /tftpboot/Template
). This template will
be used for the clients' /
file systems. You might have to
make few minor modifications to the template (after it is created) to
tailor it to your needs. You might even consider altering the script,
so you can reproduce the changes easily.
Compile a NFS-root boot floppy for the clients. The easiest way is to
make a monolithic kernel for your clients, making sure NFS, and
NFS-root file systems are compiled in (future versions of this HOWTO
might describe how to use modular kernels for net-booting). The
following options must be set to 'y' : CONFIG_ROOT_NFS,
CONFIG_RNFS_BOOTP, CONFIG_RNFS_RARP
Compile the kernel, change its root device NFS-root, and dd it onto a floppy disk.
mknod /dev/nfsroot b 0 255 cd /usr/src/linux/arch/i386/boot rdev zImage /dev/nfsroot
When you boot a client using this floppy it will broadcast RARP and BOOTP packets looking for it NFS root server.
Run the adcn
script on the server to add the new node to the
cluster. The easiest way to do it is by:
adcn -i 10.0.0.2 -c node2 -d my.beowulf.domain -l -D eth1Where
eth1
is assumed to be the interface connected to the
cluster. Place the NFS-root kernel floppy in the client's drive and
reboot it.