Live streaming has become very popular. Many systems, such as justin.tv, have emerged. They aim to collect user live-streams and serve them to the viewers using broadcasting servers. However, the huge variation in the total number of viewers and the great heterogeneity among streams popularity generally implies over-provisioning, leading to an important resource waste. In this paper, we show that there is a trade-off between the number of servers involved to broadcast the streams and the bandwidth usage among the servers. We also stress the importance to predict streams popularity in order to efficiently place them on the servers. We propose POPS: a live streaming service using popularity predictions to map live-streams on the servers.