Towards efficient power system state estimators on shared memory computers

We are investigating the effectiveness of parallel weighted- least-square (WLS) state estimation solvers on shared-memory parallel computers. Shared-memory parallel architectures are rapidly becoming ubiquitous due to the advent of multi-core processors. In the current evaluation, we are using an LU-based solver as well as a conjugate gradient (CG)-based solver for a 1177-bus system. In lieu of a very wide multi-core system we evaluate the effectiveness of the solvers on an SGI Altix system on up to 32 processors. On this platform, as expected, the shared memory implementation (pthreads) of the LU solver was found to be more efficient than the MPI version. Our implementation of the CG solver scales and performs significantly better than the state-of-the-art implementation of the LU solver: with CG we can solve the problem 4.75 times faster than using LU. These findings indicate that CG algorithms should be quite effective on multicore processors.