Communication link fault tolerance in a supercomputer

摘要

A method of operating a supercomputer having a plurality of computing elements each connected to a fast communications link is disclosed, the method comprising the steps of: scheduling specified elements to perform computing tasks in specified cycles of a computing operation; in the event of failure of a fast communications link in a given cycle, transferring state from a disabled element no longer able to communicate as a result of the failure to an idle element not scheduled to perform a task in the given cycle; operating the idle element to perform any uncompleted tasks scheduled for the disabled element remaining in the cycle.