The time taken for connection establishment is monitored to aid in selecting load distribution among nodes in a data delivery system, such as a server cluster. The failure of a node to respond to a connection request may be used to identify a crashed node. The number of connections being maintained and the amount of bandwidth being consumed may also be monitored for each node, and this information may be used to determine when a node should be removed from contention for new connection requests and when a node should be reinstated to receive new connection requests.