CONGESTION AVOIDANCE FOR THREADS IN SERVERS

A method, system, and computer program product for improving the performance, reliability, and recoverability of a computer running a multi-threaded server application. The described techniques enable synchronization requirements to be minimized. A policy is enforced which limits the number of worker threads currently assigned to processing connections to a particular host. Thus if a host fails, some number of worker threads continue to actively service connections to other hosts. The policy may use static or dynamic limits. A single limit may pertain to all hosts, or host-specific limits may be used.