I see two general advantages:
1. Scalability and robustness:
reduce load on network and server, and remove single
point of failure (server).
2. Standardization:
many cluster configurations are built around self-booting
nodes, for example, OSCAR, Rocks and HAL (High-Availability Linux).
More particular reasons for working on a local boot:
1. MPI is broken in the current configuration. I suspect it is
not compatible with a shared root partition.
2. The effort will help us understand the RLX hardware and
software. We can learn how to tailor our installation for the RLX
boards.
3. It will expand our options for configuring the cluster. I
suspect we will have enough resources for several independent
clusters configured for specific purposes.
4. It is fun!
Cheers,
Greg Walker