Burst error correction

1. A method of burst error correction in a high-performance computing environment, the high-performance computing environment including a plurality of links and switches, the method comprising:

receiving, at a switch, one or more packets and one or more codewords associated with the one or more packets;

determining, by the switch in dependence upon a link-level replay latency algorithm, whether forward error correction has higher latency than a link-level replay; and

if link-level replay has less latency than forward error correction, requesting, by the switch, a link-level replay; and

using forward error correction if forward error correction has less latency.

2. The method of claim 1 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay includes calculating a total propagation delay of a traversal of a packet from one switch to another.

3. The method of claim 2 wherein the total propagation delay includes the propagation delay of the packet including a transmission of the packet, a time to traverse a link, a reception of the packet, and a time to land a codeword.

4. The method of claim 3 wherein the time to traverse a link is approximated with a delay multiplier.

5. The method of claim 2 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay includes calculating twice the total propagation delay of a traversal of a packet from one switch to another.

6. The method of claim 5 determining, by the switch, whether forward error correction has higher latency than a link-level replay includes accommodating a delay time for requesting a link-level replay and transmitting a link-level replay.

7. The method of claim 1 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay includes calculating a probability of one or more erroneous codewords in flight in dependence upon a bit error ratio.

8. The method of claim 1 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay further comprises comparing a probability of one or more erroneous codewords in flight with a threshold.

9. A system of burst error correction in a high-performance computing environment, the high-performance computing environment including a plurality of links and switches, the system comprising a plurality of switches and links configured for:

receiving, at a switch, one or more packets and one or more codewords associated with the one or more packets;

determining, by the switch in dependence upon a link-level replay latency algorithm, whether forward error correction has higher latency than a link-level replay; and

if link-level replay has less latency than forward error correction, requesting, by the switch, a link-level replay; and

using forward error correction if forward error correction has less latency.

10. The system of claim 9 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay includes calculating a total propagation delay of a traversal of a packet from one switch to another.

11. The system of claim 10 wherein the total propagation delay includes the propagation delay of the packet including a transmission of the packet, a time to traverse a link, the reception of the packet, and a time to land a codeword.

12. The system of claim 11 wherein the time to traverse a link is approximated with a delay multiplier.

13. The system of claim 10 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay includes calculating twice the total propagation delay of a traversal of a packet from one switch to another.

14. The system of claim 13 determining, by the switch, whether forward error correction has higher latency than a link-level replay includes accommodating a delay time for requesting a link-level replay and transmitting a link-level replay.

15. The system of claim 9 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay includes calculating a probability of one or more erroneous codewords in flight in dependence upon a bit error ratio.

16. The system of claim 9 wherein determining, by the switch, whether forward error correction has higher latency than a link-level replay further comprises comparing a probability of one or more erroneous codewords in flight with a threshold.

17. A switch with burst error correction in a high-performance computing environment, the switch including a burst error module comprising logic configured for:

receiving, at a switch, one or more packets and one or more codewords associated with the one or more packets;

determining, by the switch in dependence upon a link-level replay latency algorithm, whether forward error correction has higher latency than a link-level replay; and

if link-level replay has less latency than forward error correction, requesting, by the switch, a link-level replay; and

using forward error correction if forward error correction has less latency.

18. The switch of claim 17 wherein the logic configured for determining, by the switch, whether forward error correction has higher latency than a link-level replay includes logic configured for calculating a total propagation delay of a traversal of a packet from one switch to another.

19. The switch of claim 17 wherein the total propagation delay includes the propagation delay of the packet including a transmission of the packet, a time to traverse a link, a reception of the packet, and a time to land a codeword.

20. The switch of claim 19 wherein the time to traverse a link is approximated with a delay multiplier.