GPS/MET using occultation observation has become a research hotspot in atmospheric sciences community in recently years. Much research has demonstrated that GPS occultation data can greatly improve the initial condition of numerical prediction model. The observed bending angles during occultation can be assimilated directly into model with variational technique. But the huge. amount of computation makes it an impossible mission in research and operational usage.More than 39 days CPU times will be take on CRAY C90 to assimilate 4,000 bending angles with 40 iterations. In this paper, the variational data assimilation model was not only parallelized with Message Passing Interface but also optimized for cost of input and output. 1,420 bogus bending angles are assimilated using 128 processors on PC Cluster obtains a promising parallel efficiency which is closing to linear speedup.
