Method and apparatus for ordered reliable multicast with asymmetric

Abstract

Ordered machine-readable messages are reliably delivered among processing members in a multiprocessing computer system. The system includes multiple processing nodes, each having a unique source-ID and a membership view including one or more of the processing nodes with which it can nominally exchange messages. When a stimulus message is received by a first processing node, the node increments a coordinated local counter (CC). The node also sends a multicast message to all processing nodes in the first node's membership group. The multicast message includes the received stimulus message, the incremented CC value, and the first node's source-ID. The node further sets a timer, exclusively associated with the incremented CC value. When a multicast message is received at a processing node, the node performs a multicast input processing routine. The node sets its CC equal to the greater of its current value or the received multicast message's CC value. The node also sends an acknowledgement message to all processing nodes in its membership group. Also, in response to the multicast message, the node sets a timer, exclusively associated with the received multicast message's CC value. Whenever a node's timer associated with a CC value expires before messages with the same CC value have been received from each of the node's membership group, the node invokes a membership protocol requiring asymmetric safety.