A high-performance communications network, e.g., as used in a data center, can be heterogeneous in the types of nodes and the protocols between them. Detecting and zeroing in to network conditions such as under-utilized links, faulty links, excessive latencies, etc. can become an involved task requiring the particulars of the node types and their protocols. This disclosure describes techniques of link-level measurement and network problem identification that apply generically to differing node types and protocols. Hardware counters that measure the number of clock cycles spent in transmission, re-transmission, wait-states, etc. are embedded at node egresses. The counters are triggered, e.g., by specially-defined packets. Link-performance metrics such as backpressure, retransmission rate, protocol overhead, port utilization, physical-layer bit-rate, etc., are generically derived as functions of the hardware-counter readings. The techniques enable rapid identification and localization of network problems.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.