Recoverable error detection for concurrent computing programs

A method for detecting a communication error among a plurality of nodes in a concurrent computing environment, the nodes communicating with one or more messages, at least some of the messages buffered in a buffer before delivery to one or more of the nodes, the method comprising:

• executing a concurrent computing program;
• attempting to communicate with a first one of the plurality of nodes regarding data for the concurrent computing program, wherein the first one of the plurality of nodes has entered a barrier synchronization point;
• detecting a communication error related to the attempt to communicate;
• and
• flushing any undelivered messages from the buffer before resuming execution of the concurrent computing program when the first one of the plurality of nodes has entered the barrier synchronization point.

A computer-readable storage medium storing computer-executable instructions that when executed by a processor cause communication among a plurality of nodes in a concurrent computing environment, the instructions including instructions for:

• executing a concurrent computing program;
• obtaining an instruction by a first one of the plurality of nodes to enter a barrier synchronization point;
• suspending execution of the concurrent computing program on the first one of the plurality of nodes;
• resuming execution of the concurrent computing program on the first one of the plurality of nodes when all the plurality of nodes have entered the barrier synchronization point;
• detecting a communication mismatch after the first one of the plurality of nodes has entered the barrier synchronization point;
• and
• discarding any message received during suspension of the execution of the concurrent computing program before resuming execution of the concurrent computing program on the first one of the plurality of nodes.

A computer-readable storage medium storing computer-executable instructions that when executed by a processor cause a computing device to detect a communication error among a plurality of nodes in a concurrent computing environment, the nodes communicating with one or more messages, at least some of the messages buffered in a buffer before delivery to one or more of the nodes, the instructions include instructions for:

• executing a concurrent computing program;
• attempting to communicate with a first one of the plurality of nodes regarding data for the concurrent computing program, wherein the first one of the plurality of nodes has entered a barrier synchronization point;
• detecting a communication error related to the attempt to communicate;
• and
• flushing any undelivered messages from the buffer before resuming execution of the concurrent computing program when the first one of the plurality of nodes has entered the barrier synchronization point.

A method for detecting a communication mismatch between two nodes in a concurrent computing environment, the nodes communicating with one or more messages, at least some of the messages buffered in a buffer before delivery to one or more of the nodes, the method comprising:

• executing a concurrent computing program;
• and
• sending a message by a sending node to a receiving node, the message including information identifying a region the sending node is in;
• detecting a communication mismatch between the sending node and the receiving node;
• and
• flushing any undelivered messages from the buffer when the first one of the plurality of nodes has entered the barrier synchronization point before resuming execution of the concurrent computing program.