Refer to the exhibit.

Priority Flow Control must be configured on the direct Core-R1-to-Server-12 link because that is the link on which the question requires lossless behavior. PFC, defined by IEEE 802.1Qbb, provides per-priority pause rather than pausing every Ethernet class. Cisco Nexus switches send a PFC pause frame to the directly connected link partner when a no-drop queue reaches its configured threshold. The link partner temporarily stops transmitting the affected priority while other traffic classes continue, preventing buffer exhaustion and frame loss for the protected AI or storage class. Cisco explicitly notes that PFC pause frames are local to a specific link. Consequently, enabling PFC only between the edge and core switches does not protect the final link to Server-12, and configuring it only on the compute-cluster-facing links addresses a different traffic segment. ECN marks packets so capable endpoints can reduce their sending rate, but marking alone does not guarantee that frames will not be dropped before the senders respond. A complete routed RoCE design often combines ECN for end-to-end congestion control with PFC as a safety mechanism, but the answer must protect the identified adjacency. Both Core-R1 and Server-12 must therefore use matching no-drop priority and PFC settings on their connected interfaces. Option C places the lossless control exactly where congestion threatens the Server-12 traffic path and is the only answer that directly satisfies the stated requirement. Cisco Nexus Priority Flow Control guide

================