This paper presents a dynamic hieratical task decomposition algorithm which applies for a management module of complex IT application based on multi-agent collaboration. The algorithm considers the capacity restriction of multi-agent and the dynamicity of management task caused by the variation of management strategy, IT infrastructure and service logic. Meanwhile, it also considers the balance issue of sub tasks after decomposition, which is the load balance issue of the corresponding multi-agent. The algorithm effectively improves the task executing efficiency and stability of multi-agent. Simulation and analysis results show that the algorithm in this paper is more efficient and has steadier load distribution than that of other compared algorithms.