Abstract: Currently, in the highly dynamic and heterogeneous P2P systems, load balancing, as one of the core issues of performance improvement, faces increasingly serious challenges.Based on Gossip protocol, we use topology maintenance messages among nodes to exchange load information periodically, present a load balancing solution based on the distributed genetic algorithm. This solution evaluates the fitness with the physical network topology to minimize the cost of virtual nodes migration. Sub\|populations are processed in parallel by P2P nodes to overcome the disadvantages of excessive computation of the genetic algorithm. The experimental results show that our approach has made great improvements in the speed and the quality of solution.