The modeling of distributed collaborative admission control system and mechanism are studied based on the three-layer structure of mapping framework. The video service is considered as the main research object, which is assumed approximating Poisson distribution, and service time obeys exponential distribution. A resource node in the evolution of the state of the service can be described in Markov process. Accordingly, all the N resource nodes of distributed collaborative admission control can be elaborated by a Markov switching space control process. According to the proposed Markov switching system, the performance of the system is optimized from the viewpoint of sensitivity. A stochastic approximation algorithm is put forward based on the strategy of the gradient. Meanwhile, the convergence of the algorithm is proved, and an example is given to verify the effectiveness of the algorithm.