Paper_Abstract: 
  The paper analyzes various parallel matrix-vector multiplications 
  with different matrix and vector types resulting from a non overlapping 
  domain decomposition. Under certain requirements to the f.e. mesh all 
  given matrix and vector types can be used in the multiplication. 
  The general framework is applied to the investigation of the preconditioning 
  step in cg-like methods. Not only the well-known domain decomposition 
  preconditioners fit into the concept but also parallelized global 
  incomplete factorizations are feasible. 
  Additionally, those global incomplete factorizations can be used as 
  smoothers in global multilevel methods. 
  Numerical results on a SPMD parallel machine show the good speed up 
  of the preconditioner proposed.