|
As long as the blades are able to PXE boot, it is easy to reprovision them from a PXE server on the same network. You will be providing a kernel and an initrd file (containing the files of the initial ramdisk the kernel will boot from) as well as an NFS repository for the actual Linux distribution.
We have done this for both Red Hat- and Novell- (SuSE) based Linux distributions. If you are not hung up on what Linux distribution you need, and just need "Linux," then check out the commercial distribution "Scyld Beowulf." This particular distribution would take a different approach that requires all of this manual work from you: one server is installed as the headnode and the compute nodes are being reprovisioned through PXE without actually pushing any data onto the local harddisks. This enables you to use all local disks for scratch or even use diskless compute nodes to minimize potential failing parts.
A system is reprovisioned by booting it, and is ready to run programs within seconds or minutes. This is mostly used for HPC computing and clusters but would also be an excellent platform for Web servers. You can manage it as if it was one machine, rather than individual servers. Also, blades often use newer drivers than your distribution provides so you might need to use a driver disk to install.
|
