Abstract: For implementing efficiently and safely cloud tasks scheduling, a task scheduling game alogrithm considering resource reliability is presented in this paper. Tasks execution cost, resource utilization cost, sacurity cost and load balance cost are considered comprehensively in the game's utility function in our algorithm. With optimizing the game's utility function as the goal, cloud tasks scheduling problem is formalized as a non-cooperative game model. This solving process is divided into two level. GA and improved MCT are used to solve the two level respectively, which enventually leads to get Nash equilibrium of non-cooperative game. Experimental results shows that our algorithm not only can guarantee task completion time and task response time, but can reduce the failure rate of task execution and improve the task execution efficiency integrally without reducing load balance degree and with a faster convergence.