[Beowulf] Should I go for diskless or not?

[John Hearns]

2009/5/15 Jonathan Aquilina <eagles051387 at gmail.com>:
> also dont forget the more nodes you have booting off the master the
> more access time your gonna need to do a pxe boot from the one master
> node. if you plan on splitting things up once you notice a degredation
> in booting performance and putting a certain number of nodes on each
> master node that should help to disipate the pxe booting time needed
> to boot all the nodes.

This is the way the SGI ICE is architected - each rack of blades has a
'rack leader' machine, which is the machine which PXE boots the blades
in the rack, and acts as their root NFS server.
Scales very well!