Today, parallel programming is easy and parallel computing is available 
for the masses. Distributed memory parallel computers such as Beowulf 
clusters and massively parallel supercomputers require a suitable
splitting of the data, with good load balance and minimal communication.
In this talk, we present new ways of splitting data, in the context of 
two applications: 
- Construction of low-energy atomic configurations of amorphous silicon, 
with up to 20,000 atoms, used as input to simulate e.g. solar cells.
This 3D computation is based on simulated annealing, with phases of 
local relaxation, global relaxation, and information exchange.  
The particle data are split based on sphere packing lattices. 
- Latent semantic indexing, a method used in search engines.
This requires the iterative solution of large sparse SVD problems.
A new non-Cartesian  method is presented for the distribution
of sparse matrices for such computations.