On Mon, 21 Nov 2005, Robin H. Johnson wrote: > On Mon, Nov 21, 2005 at 12:45:53PM -0500, Eric Thibodeau wrote: > > PXE-boot even for the Master?...so where do the images reside... > The PXE env resides on an install server, that's used for all of the > machines in the network (not just the cluster). (It serves up SUSE's > AutoYAST for our desktops, and a few different GLI profiles for our > other servers). Ayee... okay, so your setup is not _strictly_ Beowulf but more of a Beowulfish cluster. I should have mentionned that my intention is to come up with a procedure/scripts/profile(s) intended for a dedicated/classic Beowulf cluster.
> > how do you manage the slightly varying config items such as hostname > > and all? > hostname is taken from DHCP directly. dhcpcd -H. > The other slightly varying stuff is done with a postinstall script run > by the installer (the current script is under 100 lines). > > > This approach still seems a little bit time consuming since all nodes > > are still individual entities (not NFS roots to a single maintained > > image). Though granted that the nodes being all identical, emerge -K > > should in theory be a breeze....but it's not the case for maintaining > > all the config files consistent. > Use INSTALL_MASK=/etc after the install is completed and propagate > config files manually (well triggerable command connecting to a local > rsync server). > > We also make use of LDAP for authentication. I think I fogot to mention that the entire procedure has to fit within 15 pages with as little "frill" as possible. Notably, i was explicitally instructed _not_ to implement "complexe" authentication schemes 9such as LDAP+Kerberos/Windows)... > > So switching a machine's purpose/profile requires a complete > > re-install on the node? > Yes, but I'm talking about radical alterations here. > Eg we've got a binary app that needs the system compiled with gcc-3.3. > It doesn't work if glibc or a few other things are compiled with > gcc-3.4 or newer. This doesn't rule out using a collection NFS roots... Bare in mind that i am an LTSP user/implementer... so I've at leas had _some_ experience with the principle but haven't had to implement it from A to Z as of yet. > > You state 20 minutes for re-installing, is it > > a _real_ install > Full GLI install, extract stage, portage tree, emerge -K all needed > packages. If you want to make each node compile everything, it's about > an hour on the nodes (they are pretty speedy boxes). > > > or the dump of a "reference" root? (Pardon my > > ignorance of the CLI installer you are referring to... I'll read the > > http link you'll send me ;) ) > Look for GLI first of all. It's the Gentoo installer. It has a number > of frontend modes - GTK, Dialog, and CLI. > The CLI is entirely non-interactive. Hmm...will do then. I've been installing Gentoo "by hand" since 1.4 and never sought/required another technique... so how do you call the CLI gentoo Installer? Do you change the default behaviour in your DHCP server's config so that the PXE env. that is booted automatically triggers the installation of such or such profile? Thanks, Eric -- [email protected] mailing list
