Systems and methods are presented herewith for selecting a preferred route for routing a packet from a first network node to a second network node. A set of possible routes is maintained, with each route having am associated weight value. A random subset of routes is then selected based on the weight values. Each route of the subset is then probed to determine its gain value. The preferred route is selected based on the gain values (e.g., by selecting the highest gain value). Then, all weight values are updated based on the respective gain values. The steps are periodically repeated. Then, whenever a packet needs to be routed, the route currently designated as preferred is used.