the RSDFT program

the code is fully parallelized using the MPI message passing library for distributed memory machines. we use the LAM/MPI library on LINUX. the following very preliminary speedup tests were performed on our rather new 8 node dual-xeon cluster with GigE for the communication network. we try to cut the communication overhead by exploiting the locality of all operations in real space to allow large scale simulations on this rather cheap type of connect (compared to myrinet or sci etc.).

the graphs show the speedup for a Ga4N4 cluster and its dimer Ga8N8.

rochus (edited 22.07.03)