Unlike RANS method, LES method needs more time and much more grids to accurately simulate the spray process. In KIVA, spray process was modeled by Lagrangain-drop and Eulerian-fluid method. The coarse grid can cause errors in predicting the droplet-gas relative velocity, so for reducing grid dependency due to the relative velocity effects, an improved spray model based on a gas-jet theory is used in this work and in order to validate the model seven different size grids were used. In this work, the local dense grid was used to reduce the computation cost and obtain accurate results that also were compared with entire dense grid. Another method to improve computation efficiency is the MUSCL (Monotone Upstream-centered Schemes for Conservation Laws) differencing scheme that was implemented into KIVA3V-LES code to calculate the momentum convective term and reduce numerical errors. At last, Combustion is simulated using parallel algorithm for realistic chemistry mechanism that is also added into KIVA3V-LES code and named KIVALES-Chemistry-Parallel in this study. The results show that these four methods are effective ways to reduce cost and obtain well results compared with experiment data.