Abstract: Service selection is a key step to transform a user's request into an optimal composite service plan. Most existing approaches are influenced by the number of tasks and candidate services and have shortcomings of long response time for large size candidate services and without considering the critical path. To solve above problems, a peer-to-peer (p2p) service overlay was built up to simulate distributed applications. Considering the composite construction and critical path, the end-to-end QoS constraint could be decomposed into local ones by using the management of service community, which facilitates parallel service selection. Experimental results demonstrate that the composite service plan could be built up with great reduced response time and the tradeoff to a near-optimal result.