Implement ECN Support for NVMe/RoCE Flow Control on FS7600

1. Current Limitation

IBM FS7600 currently supports PFC (Priority Flow Control) but does not support ECN (Explicit Congestion Notification) for NVMe/RoCE traffic.

2. Customer Impact and Technical Concerns

In an RoCE network, ECN and PFC serve complementary roles in congestion management. ECN provides congestion notification to trigger the sending end to reduce its transmission rate, while PFC provides hop-by-hop flow control by temporarily pausing traffic when necessary to prevent packet loss.

Without ECN support, the FS7600 cannot participate in ECN-based congestion notification. The network may have to rely more heavily on PFC to prevent congestion-related packet loss, potentially causing traffic pauses and impacting performance, especially in high-throughput, multi-host environments such as TSMC.

The customer is concerned that supporting PFC without ECN does not provide the expected end-to-end congestion management behavior for their NVMe/RoCE deployment.

3. Feature Request

We request that IBM Development prioritize and accelerate ECN support on FS7600, alongside PFC, to provide a more complete congestion management solution for NVMe/RoCE environments.

The implementation should support the appropriate ECN marking and congestion notification mechanisms required for RoCE, ensure interoperability with host NICs and network switches, and align with the applicable RoCE congestion management specifications and best practices.

Please also review and correct the relevant technical documentation to ensure that IBM's stated NVMe/RoCE flow control behavior accurately reflects the supported capabilities and limitations.

4. Business Priority

This is a critical requirement for TSMC's NVMe/RoCE deployment. The customer expects IBM to provide a technically complete and interoperable flow control solution, rather than relying on PFC alone.