Abstract: In most studies, the selection of super-peer only considers whether the performance of nodes is suitable for super-peers, without taking into account the speed of the network to rebuild, when a node is out of or accession from the network .In the algorithm study, the on-line time and capacity was less considered at the same time. The algorithm was improved based on Alberto Montresor algorithm. From the concept of the self-information, combined with the node capacity and on line time, the algorithm of super peer selection was designed. The results show that the speed of building network for the self-information algorithm is similar with singlcused capacity, and selected super-peers were more stable. The self-information algorithm will effectively reduce the frequency of building network, thereby reduce the costs of the network traffic. For some catastrophic situations, for example, removing 50 0 o of nodes from the network, it can be quickly built. hhis indicats that the network built by our algorithm has robustness.

Key words: Super-peer, Node capacity, On-line time, Self-information algorithm