With the progress of the network, peoples can communicate with each other easily. The popularization of the Broadband network makes many things that could be hard to achieved becoming possible. For example, video conference, distance education and on-line games have become a part of our daily life. However, most network services are client-server architecture. This kind of service need a expensive and powerful server. The restriction will makes it difficult to improve the quality of service. P2P (peer-to-peer) architecture embodies one of the key technical concepts of the internet and many internet users was attracted by it. A peer-to-peer computer network is a network that relies on computing power at the edges (ends) of a connection rather than in the network itself. It has been used in two kinds of applications which are file sharing and distributed computing. Instead of using P2P model, client-server architecture is used in muti-user conference. A software using P2P model still needs one of the users to become a server to deal with all the messages in the conference. When the number of the users increasing, the overhead of the server will increase linearly. This makes many communication softwares to limit the number of users in a conference. Therefore, most of the companies use an expensive and powerful server offering this kind of service. In this paper, we will implement a voice-conferencing system using P2P model and the technique of distributed computing. It makes the overhead of the server in logarithmic increase. In addition, we still can configure some low-end peers with a constant overhead. No mater how the number of users increase, these peers will have a constant overhead. With the good design, we can gurantee the quality of service easily.