A method for determining a load distribution for a plurality of servers is disclosed. A total user count during a predetermined interval of time is received from each server of a plurality of servers for all channel resources associated with each respective server of the plurality of servers. A present load distribution is determined for the predetermined interval of time for each respective server of the plurality of servers based on the total user count received from each server. A load gradient is determined for the predetermined interval of time from each server of the plurality of servers. A future load distribution is determined for each respective server based on the total user count for each server and each respective load gradient. Lastly, a load distribution for each respective channel resource is distributed among the plurality of servers based on the determined future load distribution for each respective channel resource. The predetermined interval of time is a sliding window of time having a predetermined number of timeslots each having a predetermined timeslot interval. Additionally, a load gradient for a server is based on a difference between the total number of users connected to the server at the end of the predetermined interval of time and the total number of users connected to the server at the beginning of the predetermined interval of time. When a new channel resource is created, an initial estimated weight is assigned to the new channel resource. The new channel resource is then assigned to a selected server of the plurality of servers based on the initial estimated weight associated with the new channel resource and a load distribution associated with each respective server of the plurality of servers.