Seven Dark Trunks After a SAN Cutover: Polarity Method Versus Loss
A payments processor is cutting a SAN island into a new row during a 6-hour maintenance window. The MDA-to-HDA backbone is 18 × 12-fiber OM4 MPO trunks documented as Method B: Type B (key-up/key-down) trunks with Type A harnesses at each end. New 40GBASE-SR4 switches use 8-fiber MPO ports (pairs 1–4) and connect through 12-to-8 harnesses. After cutover, 7 of 18 trunks are completely dark and 4 more link but drop frames under sustained load; the remaining 7 are clean. Field staff substituted two Type A 12-fiber patch cords for suspect harnesses, and one dark trunk came up, so they propose converting the whole row to Type A. Documentation is missing for three trunks, and the manufacturer's 25-year warranty requires a documented, consistently applied polarity method. The next maintenance window is in 12 days; the SAN replication lag SLA is already being breached.
1. Diagnose the two failure populations: 7 dark trunks versus 4 trunks that link but drop frames.
Read the worked answer
Treat them as two different faults, because polarity failures are binary. A polarity error either puts transmit into transmit or reverses lanes, and the link never trains - that is your 7 dark trunks. The most probable cause is a Method B plant containing Method A assemblies: in a B plant, a Type B trunk plus Type A cords at both ends yields correct duplex polarity, so a Type A trunk or a flipped harness in that path inverts the pairs at the far end. The fact that one dark trunk came up after a cord swap is consistent with correcting an inversion, not with fixing a loss problem.
Frame drops under load are a loss or signal-integrity signature, not a polarity one: contaminated end faces, a marginal 12-to-8 harness with a damaged fiber, or mismatched polish. Do not let the crew 'fix' those with crossover cords; that converts a measurable problem into an unmeasurable one. Isolate by testing each trunk's 8 fibers with an OLTS and inspecting every end face before touching polarity.
2. Evaluate the proposal to convert the entire row to Type A polarity.
Read the worked answer
Reject it as stated. Converting a documented Method B plant to Method A means changing every trunk or every cord, at both ends, across all 18 paths - that is a full re-cable of a live SAN backbone inside a 6-hour window, with the highest-risk links being the ones currently working. It also voids the basis of the 25-year warranty, which requires one method applied and documented consistently; a mixed plant with three undocumented trunks is precisely what a warranty claim reviewer will reject.
The valid reason to choose Method A is a plant built for it. Here the installed base, the documentation, and the harness standard all point to B. The correct action is to restore B: verify polarity trunk by trunk, replace the non-conforming assemblies with Type B trunks and Type A harnesses, and correct the record. If a specific path cannot be brought into B before the window closes, put a labeled, documented exception on it and schedule the permanent fix - do not silently change the method for the row.
3. Lay out the recovery plan for the 12-day window, including what you will and will not promise.
Read the worked answer
Before the window: pull the as-built records, then physically verify polarity and mapping on all 18 trunks with a polarity tester or OLTS, one fiber at a time, against a written mapping table (fiber 1 to fiber 12 for B). Identify the 7 dark paths and confirm each is an inversion rather than a break. Pre-stage the correct Type B trunks and Type A harnesses, plus spare 12-to-8 harnesses with known mapping, and pre-inspect and clean every end face.
In the window: correct polarity first on the dark paths, then re-test; then address the 4 marginal paths with inspection, cleaning, and re-mate before replacing anything. Certify every restored path with bidirectional OLTS and record the mapping.
What you will not promise: that all 18 are clean within one window if spares are wrong or undocumented trunks turn out to be Type A assemblies. Give the business a prioritized list - replication-critical paths first - and a documented fallback for anything deferred.