Kinetic simulation of plasmas based on the Particle-in-Cell (PIC) method, which began in the late sixties, has been further improved in recent years for the purpose of studying fusion plasmas. The new methodology called gyrokinetic particle simulation, which gives rise to orders of magnitude improvement in terms of the spatial/temporal-scale requirements for the simulation plasma, has proven to be very adaptable to the modern-day supercomputer architecture for studying magnetic fusion physics. For the present talk, we will describe our experience in using the leadership computing facilities for fusion science discovery and the plans for the future, i.e., petascale computing for the international burning plasma experiments, ITER. We will also discuss the issues such as code performance, data management, and visualization. The collaboration with the applied mathematics community for the improvement of the performance of the Poisson solver is also an important topic to be addressed.