Multi-robot cooperation promises increased performance and fault-tolerance in large-scale environments. To achieve the goal with these features, dynamic task allocation algorithm is required to adapt changing environment. Whereas previous researches mostly have been done for simplifying the mission and emphasizing the task allocation algorithm, this paper addresses the needs to analyze the mission more concretely and introduces a strategy for the task allocation problem: off-line and on-line analyses. The off-line analysis takes the robot-working scenario into account statically before deploying the robots. The on-line analysis considers the real time task allocation algorithm while running the robots. Combining these two analyses, efficient task allocation is achieved with minimum cost.