Node-tearing-based ordering offers more scalability than recursive spectral bisection-based ordering techniques, although the structure in the electrical power system matrices does not permit unlimited scalability. Node-tearing-based ordering produces numerous diagonal blocks that contribute to effective load balancing for four to sixteen processors, while offering the potential to generate matrices in a form that could be factored efficiently by vector processors.
Our research into specialized ordering techniques has shown that load imbalance becomes excessive beyond 16 processors, limiting scalability for a single parallel linear solver. Nevertheless, other dimensions exist in electrical power system applications that can be exploited to efficiently use large numbers of processors. While a limited number of processors can be efficiently applied to a single power system simulation, multiple events can be simulated simultaneously.