You need to design a network that supports transmitting RDMA traffic over multiple IP hops. Which protocols must the switches support? (Select all…

Designing a network for multi-hop (routed) RDMA over Converged Ethernet (RoCEv2) requires a fabric that is both lossless and capable of managing congestion end-to-end.

Priority-based Flow Control (PFC) is essential to prevent packet loss from buffer overruns on a hop-by-hop basis, creating the necessary lossless behavior for RoCE.

Enhanced Transmission Selection (ETS) allocates a guaranteed minimum bandwidth to the RoCE traffic class, ensuring it is not starved by other traffic types.

IP Explicit Congestion Notification (ECN) is the standard Layer 3 mechanism used in RoCEv2 deployments to signal network congestion from switches to endpoints without dropping packets, allowing the sender to adjust its rate and prevent congestion collapse across multiple hops.

D. QCN: Quantized Congestion Notification (QCN) is a Layer 2 congestion management mechanism and is not used for routed, IP-based RoCEv2 traffic, which relies on IP ECN.

1. HPE Aruba. (2023). ArubaOS-CX 10.12 Fundamentals Guide. Chapter: "Lossless networking for RoCE". The guide states

"Lossless networking for RoCE is achieved by configuring the following features on all switches in the RoCE network: Priority-based flow control (PFC)

Explicit congestion notification (ECN)

[and] Enhanced transmission selection (ETS)..." This directly confirms the necessity of all three protocols.

2. NVIDIA. (2020). RoCE End-to-End Congestion Control for Spectrum Switches and ConnectX-5 Adapters - Recommended Configuration. Section 1.1

"Introduction". This official vendor document details the solution for RoCE congestion control

stating

"The solution is based on ECN marking by the switch... PFC is used for short burst absorption." This highlights the combined role of ECN for end-to-end control and PFC for lossless behavior.

3. HPE. (2021). HPE FlexFabric and HPE M-series Networking for RoCE Solutions. Page 6

"Data Center Bridging (DCB)". This document explains the core components for RoCE

noting

"PFC provides a lossless service... ETS provides a method for allocating link bandwidth... ECN is an extension to the IP protocol that enables end-to-end notification of network congestion without dropping packets."