There is a fundamental trade off between system efficiency and fairness. For instance, in wireless networks maximization of total throughput at a base station often results in user unfairness. Thus, one would like to trade some efficiency for better fairness. In the present work we argue that different applications can require different type of fairness. Real time traffic requires short-term fairness when the resource is distributed fairly on any short time interval. On opposite, best effort traffic requires long-term fairness. In long-term fairness at any given time interval the resource can be distributed unfairly but over a significant time interval the resource should be distributed fairly. Then, the streaming traffic (like the traffic originated from YouTube application) requires some intermediate notion of fairness. In the present paper we introduce a new fairness concept, multiscale fairness, which allows one to distribute the network resources fairly among different traffic classes.