Hello all, I'd like to introduce to the list a piece of software that just happens to do exactly what this project sounds like its trying to accomplish. The software handles the root image of any number of nodes, but allows for node/role specific behavior. It allows you to deploy a local cache of the root image onto a local disk, and a bootloader if desired. No matter what, though, the root filesystem is bit-for-bit identical everywhere. Deviations are handled by creating soft links to/from a tmpfs ramdisk. It really gives ultimate flexibility as to how/where components of the filesystem should live (localdisk/ramdisk/NFS). It can bootstrap the traditional NFSroot way (root=/dev/nfs) or use an initrd to bootstrap in any number of creative ways. The framework is in place to be able to easily boot over any network filesystem (including Lustre when it arrives), or over any network medium (including InfiniBand). It uses an initrd 'template' to create new initrds with very customizable behavior. The project is called oneSIS (http://onesis.sourceforge.net). We've used it in production for running clusters for years, but I've just recently (early this year) re-written it from scratch to make it cleaner and more flexible. Grab the CVS version from sourceforge and check it out. It makes everything the "Stateless linux" project is trying to do very, very easy. There is more information in the oneSIS software manual (http://onesis.sourceforge.net/oneSIS-manual.pdf), and I currently have a paper being reviewed for submission, or I'd send it also. -JE ----------------------------------------------- Josh England Sandia National Laboratory, Livermore, CA Visualization and Scientific Computing email: jjengla@xxxxxxxxxx phone: (925) 294-2076 Miles Sabin wrote, I think this is a fine idea. I have a few questions tho' ... Outside of largish corporates where clones are common, there are smaller environments where machines are mostly (95%+, say) similar, but where there are machine-by-machine variations which fall outside the set of things which are covered by per-user homes or hardware auto- configuration. For instance, differences in which services should be running, their configuration, or minor kernel tweaks. Are such environments outside the scope of this proposal, or is there some plan to support per-machine or per-group/class deltas from a common baseline? One thing that's always stopped me from bothering with NFS mounted homes is that for me a different machine quite often implies a different role, hence that the stuff that's appropriate to be in my home varies. This also tends to be a delta from a common baseline, so handling the variability by having distinct users for each role doesn't seem like quite the right answer. Caching doesn't seem likely to change this, and the "recreate Joe's workstation" scenario in the document seems to wander into this kind of territory. Do you have any thoughts about handling this? Slightly related to the previous two questions ... My laptop moves between my home network, my work network, business-client networks, connected by dialup ISP, and disconnected. In each scenario the services running, their configuration, and what's appropriate for my home dir vary slightly. Even tho' this is pretty clearly beyond the scope of your proposal, there seems to be at least something in common: the setup is mostly the same, but there are some small deltas. Would it make sense to have a common mechanism to handle this? I think all my questions boil down to: looks great for the 95% of stuff which stays the same, but how do you deal with the 5% of stuff which varies? And can you do it in a way which is 95% less hassle than treating the machines as completely separate entities? Cheers, Miles
