In order to construct a trustworthy computing platform for the next Internet, there needs to be normalizing and promoting of autonomic elements in order to have them collaborate actively with on another. A novel penalty-incentive mechanism, named PETrust, based on a repeated game theory, is given in this paper. This paper aims at providing a set of mechanisms, which the behavior of autonomic elements is normalized and is promoted to take the expected strategy. PETrust adjusts the degree of penalty by changing the reputation status. Theoretical analysis and simulation results show that PETrust can distinguish the different features of behavior, effectively, can punish and stifle malicious behavior, can improve the system’s entire efficiency, can stimulate autonomic elements’ honest trade enthusiasm, and can provide a better capacity of resisting collusive deception. Furthermore, PETrust presents both low time complexity and few incurred packets, which is favourable for engineering deployment and implementation.