Abstract: An adaptive, symmetrically distributed load balancing system and method for resource management in a computer network. The architecture is fully symmetrical, with all nodes executing the same system code and treated as equivalent. The system is therefore inherently fault-tolerant and scalable to any degree, with a load balancing capacity that increases linearly with cluster size. An extension from single cluster operation to multi-site operation is provided, and performance is optimized through locality enhancement, by tracking latencies to select nodes offering the fastest response. Both inter-cluster and intra-cluster latency information is maintained. The system also selectively retains past states and thus has access to a rich set of recent past extents (represented in a cache) which may still be valid, and which serve as hints of good quality. Session support by all cluster members as well as support for quality of service are also provided.