Addressing the shortcomings of existing link congestion loss rate range inference algorithms in large scale IP network, a new link congestion loss rate range inference algorithm based on greedy heuristic method is proposed. The strong dependency on the clock synchronization of single slot E2E path measurements is avoided through using multiple slots E2E path measurements. Each congested link can be located through adopting the link congestion Bayesian maximum a-posterior (BMAP) after learning prior probabilities of the link congestion. The set consisting of paths with related congested links and similar performance is constructed. Through solving the performance similarity coefficient dynamically, loss rate range of each congested link can be recurrently inferred. The accuracy and robustness of the algorithm proposed in this paper is verified by experiments.