Network Performance Anomaly Detection with In‑band Telemetry and High‑Performance Congestion Control (HPCC++) at the 2022 OCP Global Summit

The 2022 Open Compute Project (OCP) Global Summit in San Jose gathered experts to accelerate data‑center innovation, and Alibaba partnered with Broadcom to deliver two technical sessions.

The first talk, “Network Performance Anomaly Detection using In‑band Telemetry,” explained how Alibaba Cloud’s In‑band Flow Analyzer (IFA) inserts metadata into packets at each hop, enabling fine‑grained monitoring beyond traditional syslog, SNMP, or gRPC methods. The presentation described IFA’s workflow, the massive data generated at scale, and the challenges of resource consumption and lack of unified hardware standards.

The second session, “High Performance Congestion Control (HPCC++) for RoCEv2 Networks Leveraging SAI TAM,” introduced the HPCC++ algorithm, which uses IFA‑embedded probe packets to collect per‑hop congestion metrics (queue length, bandwidth usage, timestamps) and adjust sending rates for near‑zero queuing latency. Broadcom’s SAI Telemetry And Monitoring v2.0 (SAI TAMv2) architecture and Alibaba’s deployment experience were discussed, emphasizing the algorithm’s suitability for high‑performance storage networks where traffic exceeds 60 % of data‑center load.

Both talks highlighted the balance between monitoring fidelity and cost, the need for close collaboration with silicon vendors, and the growing interest from the OCP community in standardizing large‑scale IFA deployments.