Methods and systems for load balancing using forecasting and overbooking techniques

Methods and systems for establishing user sessions between a client and a server or server farm can be carried out by a load balancing agent executing on a computer in communication with the client, server or server farm. The load balancing agent can intercept a connection request generated by an application executing on a client and responsively select a server on which to establish a user session. Selecting the server can be based on overbooking a particular server and choosing a server with a predetermined amount of forecasted load. Forecasted load can be determined by summing the current load on a server with a current load calculated by multiplying: the average load on the user sessions hosted by the server; by the number of cached user sessions stored on the server; and by the probability that a cached user session reconnects.