After a township marks ordinary backup flows with the RoCE codepoint, operators see PFC storms across the pod. What should they fix first?
Select an answer to reveal the explanation.
Short Explanation
If the wrong trucks get the “fragile freight” sticker, the whole loading dock starts slamming the brakes—classic PFC storm from mis-marking. Fix classification first so only real RoCE rides the lossless class.
Full Explanation
PFC storms after a township marks ordinary backup flows with the RoCE codepoint mean classification—not routing—is broken. Fix traffic classification and marking first so only true RDMA flows enter the lossless RoCE class. Priority Flow Control acts on the traffic class; mis-marked backups will trigger pause behavior, congest unrelated queues, and create the storm operators are seeing across the pod.
Increasing BGP MED on all spines does not rate-limit pause frames through routing policy and confuses underlay path selection with CoS. Disabling ECMP so PFC storms follow a single deterministic path fails to correct misclassification and unnecessarily removes fabric multipath. Lowering underlay MTU below 1500 does not stop PFC frame generation caused by wrong markings and can introduce unrelated overlay blackholes.
Exam caveat: when lossless storms appear right after a marking change, the first conceptual remediation is classification/marking integrity, not underlay BGP or MTU knobs.
Operational check: capture or sample DSCP/CoS of backup versus genuine RoCE flows at the leaf; correct classifiers and rewrites; monitor PFC counters under the same backup job and confirm pause activity collapses for non-RDMA classes.