In the field of “Internet of things”, area management and emergency rescue, the requirement of large-scale device collaboration is growing. Aiming at the large-scale and strict timing constraint of the large-scale device collaboration systems, in this paper, a two-level task model for large-scale device collaboration is presented. The task model is made up of collaboration task and device task called collaboration subtask. Based on this task model, a large-scale device collaboration mechanism described with Pi-calculus is given, which consists of task-level collaboration mechanism and subtask-level collaboration mechanism. In the task-level collaboration mechanism, aiming at the exclusivity of device resources, resource reservation is adopted to avoid the device access conflict among collaboration tasks. The subtask-level collaboration mechanism ensures the timing constraints among collaboration tasks and collaboration subtasks. It includes three parts: collaboration mechanism between task and subtask, collaboration mechanism among subtasks and collaboration mechanism based on time. Then the large-scale device collaboration mechanism’s accuracy is proven theoretically. Finally, a large-scale device collaboration prototype system is designed and implemented. The results of simulation experiments on the prototype system show that the large-scale device collaboration mechanism can satisfy the performance requirements of large-scale and strict timing constraint in the large-scale device collaboration.