A Parallel WLS S tate Estimator on S hared Memory C omputers

This paper descr ibes our exper ience with developing a par allel weighted- least-squar e (WLS) state estimation (SE) pr ogr am for shar ed-memor y par allel computer s. Since the key computational ker nel of the WLS algor ithm based on the Newton-Raphson appr oach is a solver of spar se linear equations, a significant par t of our effor t was focused on selecting, implementing and evaluating this algor ithm. An optimized shar ed memor y ver sion of the conjugate gr adient (CG) algor ithm was found to be competitive to state-of-the-ar t implementation of LU solver s for the SE pr oblem on the SGI Altix, a shar ed memor y ar chitectur e. We also por ted the full SE algor ithm including CG to the Cr ay MTA-2 shar ed memor y multithr eaded ar chitectur e and investigated its per for mance.