System and method for managing packet transmission issues in high-performance computers

A High-Performance Computer comprising:

• a plurality of nodes, wherein a node of the plurality of nodes is configured to
• wait for a first predefined time period after sending a packet to another node, and retransmit said packet if an acknowledgement for the packet has not been received in the first predefined time period, and
• periodically send a heartbeat message to each other node having a pending transaction with said node, wait for a second predefined time period after sending said heartbeat message, and abort the pending transaction if an acknowledgement for the heartbeat message has not been received in the second predefined time period, and
• when receiving a negative acknowledgement after said sending said packet to said another node, retransmit said packet after the negative acknowledgement has been received, said packet being retransmitted after a third predefined time period starting from the receiving of the negative acknowledgement, and
• wait for a fourth predefined time period after receiving said packet and drop the packet if said node is congested during the fourth predefined time period and send a negative acknowledgement message to the node having sent the packet that is dropped, and
• an interconnect network interconnecting the plurality of nodes, wherein the interconnect network is configured to transport packets between nodes of the plurality of nodes, wherein the interconnect network is a Bull exascale Interconnect (BXI) network.

A method implemented by each node of a plurality of nodes of a high-performance computer, the each node being interconnected to other nodes of the plurality of nodes via an interconnect network said method comprising:

• sending a packet to another node via a first node,
• waiting for a first predefined time period after the sending the packet, and retransmitting said packet if an acknowledgement for the packet has not been received in the first predefined time period, and
• periodically sending a heartbeat message to each other node having a pending transaction with said first node, waiting for a second predefined time period after periodically sending the heartbeat message, and aborting the pending transaction if an acknowledgement for the heartbeat message has not been received in the second predefined time period, and
• when receiving a negative acknowledgement after the sending the packet to the another node, retransmitting said packet after the negative acknowledgement has been received, said packet being retransmitted after a third predefined time period starting from the receiving of the negative acknowledgement, and
• waiting for a fourth predefined time period after receiving the packet and dropping a received packet when said first node is congested during the fourth predefined time period and sending a negative acknowledgement message to a node that sent the packet that is dropped, wherein the interconnect network is a Bull exascale Interconnect (BXI) network.