Document Home

Installing Straight to Testing (Woody) Across The Network

There is a way to do a very fast network installation straight to the testing distribution that you might want to try with one of the later workstations in your cluster. This is a far trickier installation than it might seem, and I would not suggest trying it until you acquire a fair degree of familiarity with the basic debian setup. I've done it a few times, and things can get a little tricky if anything untoward happens during the installation. At the same time, it can save a good bit of time, so it worth giving it a shot. If worst comes to worst and you find yourself hopelessly adrift, just boot the box with the rescue diskette from the stable distribution, reinitialize the partition structure, and install as discussed previously. (While there is a set of binaries for the base woody distribution on diskette, it requires nineteen diskettes, and the difficulty of maintaining a good set of diskettes with the eleven required for the stable distribution is significant enough that I would not suggest taking that approach with nineteen.)


First, get copies of the rescue.bin, root.bin, and the four drivers binaries here. (This assumes that you are using 1.44 Mb. diskettes on an Intel i386 architecture machine. If, by some quirk of fate, you are using diskette drives that use diskettes of different size, go to the parent directory at that url and you will see directories for other diskette formats.) Once you have downloaded those files, use rawrite2 to create diskettes from the binary images (type "rawrite2", when asked specify the name and location of the binary file, e.g., "c:/woody_binaries/root.bin", then type "a:", or whatever your diskette drive letter is, as the destination drive.) When you've created the diskettes, boot to the rescue diskette to begin the installation. Once you've inserted the root disk and the installation menu initializes configure the keyboard and partition the hard drive as appropriate, then load system and device modules. After the installation script load the requisite files from the rescue diskette and four driver diskettes, you should configure device driver modules. Unless you have some relatively unique hardware that is required for the system to boot, such as a relatively esoteric scsi adapter (in which case you should give second thought to following this route), I would suggest that you keep to the minimum driver configuration required to boot the machine and provide the environment you need to continue the installation. For a standard cluster workstation, with IDE drives and/or a common or generic scsi interface, that means that you only need to configure two modules, the network card and the smbfs file system. Indeed, you strictly don't need smbfs, but samba file sharing can represent enough of a configuration aid that you should not overlook it.


Once you have configured the devices, you will configure the network (assign ip address, netmask, etc.), and the next step will be to install the base system. Here the options include network as a source. As long as your network card was successfully installed and you are connected to the network, you can select the network installation option. You will be asked to specify a source for the installation - if possible select the site from which you have previously installed the testing distribution and when asked identify the address of your proxy cache if you have configured squid. This is not as big a deal with the base system as with assemblies of packages, but it does get those settings established. The installation script will then launch into pulling those packages across the network, saving you the toil of feeding diskettes, and if you install additional machines in this manner the base system will be pulled from the proxy cache, if one is present. After installing the base system you can make the system bootable, reboot, and proceed as with previous installations.