A method and system for balancing load across a set of target endpoints available on a server, and initiator endpoints accessing the server. The method including starting rebalancing of target endpoints at a defined interval, receiving monitored load data for a set of target endpoints, disqualifying target endpoints in the set of target endpoints that have a low load, selecting a next most busy target endpoint, marking the selected target endpoint as disqualified, classifying alternate paths of virtual connections assigned to the selected target endpoint according to busyness of endpoints of the alternate paths, load order and load imbalance, examining a load reduction offered by the alternate paths in order of classification, and returning an alternate path that has a highest load reduction for target endpoint.