A conference system includes a controller and clusters of media nodes. Each media node performs media packet processing operations and connects with other media nodes and client devices. The controller sends to a client device contact information for candidate clusters. The controller receives from the client device a measure of reachability to each candidate cluster as determined by the client device using the contact information. The controller receives from the client device a request to join a communication session. Responsive to the request, the controller determines a best cluster among the candidate clusters to which the client should connect for the communication session based on the measures of reachability to the candidate clusters, and selects a media node in the best cluster. The controller sends to the client device contact information for the media node to enable the client device to connect to the media node for the communication session.