JLdL 06Mar05.

The first two things you must do, in terms of system installation, if you
want to get a cluster up and running, is to first install the server, and
then to install or obtain the first node. Installing the server is a task
like installing any other central server for groups of users, and it will
not be further commented on here. After the server and the first node are
correctly set up and the server is running, the tools can be used to help
you to clone as many more nodes as you want, and the server can handle.

All you need is a copy of the Debian base system, so that you can copy it
into the directories of first or virtual node.  These directories will be
within the filesystems reserved for the cluster nodes, with names such as
00/ tmp/00/ var/00/ and usr/00/. Since the tmp/ filesystem has no content
you simply have to create the directory and fix its access mode. In order
to obtain the content needed for the other directories, there are several
methods that you can use:

1) If you are installing the server from scratch, you can install it up to
   the base system and pause there. Then clone the first node, copying the
   appropriate contents to the node directories, before you go on with the
   installation of the server.

2) Install up to the base system, in any spare hardware the you happen to
   have, copy the contents of each filesystem into the directories of the 
   first node in the server, and adapt the necessary files for the node.

3) Install from scratch, up to the base system, directly into the roots of
   the filesystems on the server that are reserved for the nodes, and then
   move the contents into the directories of the virtual node.

4) Get a set of tar files for a fully installed and configured node from a
   friend that already has a cluster of the correct type.  This may save a
   lot of work, possibly all of it in the case of compute nodes.

5) Get tar files with the standard Debian base system, organized according
   to the appropriate filesystem scheme, and then install each one of them
   into the corresponding directory of the first node. These tar files can
   be obtained at the site

		   http://sft.if.usp.br/cluster/

Some comments about this last alternative are necessary.  These compressed
tar files contain the base system for the i386 architecture only. The base
system is not completely configured, and the kernel version is 2.4.27.  It
will be necessary to perform a package upgrade, immediately after they are
installed. You should consider doing a kernel upgrade as well.  Before you
choose to download them, you should consider their sizes:

Sizes of the Debian-Sarge base system tars, with no compression.

-rw-r--r--  1  root  root          50001920 Mar  4 12:33 root.tar
-rw-r--r--  1  root  root          42577920 Mar  4 12:34 var.tar
-rw-r--r--  1  root  root          78387200 Mar  4 12:35 usr.tar

Total: approximately _163.1_MB_   170967040

After compression with bzip2 they reduce to almost a third of that size:

-rw-r--r--  1  root  root          20069926 Mar  4 12:33 root.tar.bz2
-rw-r--r--  1  root  root           8496323 Mar  4 12:34 var.tar.bz2
-rw-r--r--  1  root  root          30693822 Mar  4 12:35 usr.tar.bz2

Total: approximately _56.5_MB_     59260072

The tar files should be opened directly into the target directories, with
the command "tar -xpjf <file>.tar.bz2" executed within each directory:

contents of    root.tar.bz2    go into    <cluster-root>/00
contents of    var.tar.bz2     go into    <cluster-root>/var/00
contents of    usr.tar.bz2     go into    <cluster-root>/usr/00

We use here node directories with two digits, as an example. After the tmp
directory <cluster-root>/tmp/00 is created, you should fix its access mode
with "chown 1777 <cluster-root>/tmp/00".

In any case, you must fix at least the following files for the first node,

	/etc/apt/sources.list
	/etc/fstab
	/etc/hostname
	/etc/hosts
	/etc/mailname
	/etc/networks
	/etc/network/interfaces
	/etc/resolv.conf

You will probably also have to write a file /etc/exim4/exim4.conf in order
to forward all mail to some smarthost handling mail transport in the local
network. It will probably be necessary to change the timezone as well, you
can do that with the command tzconfig.  Finally, you should set e new root
password because there is no root password set in this base system.

