> PXE supports it, for netbooting clusters of 100s or 1000s of machines. > I've never come across anybody who actually uses it. Big clusters these > days tend to have fast networks that laugh off the traffic involved in > unicasting netboot images to a few thousand machines.
You tend to get hit by timeouts on the server with booting multiple machines at once - we found about 1/2 a rack at a time was the best we could get. (but the cluster head node also acted as the kickstart server so traffic got kinda busy come upgrades) Switching to gPXE / gpxelinux and using http rather than tftp for transferring the boot image was a much better improvement than trying to sync and multicast another nice scalable installer is to use bittorrent - see the Rocks 'Avalanche' installer (on the other hand, reimaging a callcentre at once using norton ghost + multicast was waaay cool when I did it oooh 10 yrs ago) Andrew