Dark Links, Intermittent CRCs: Auditing an OM3 Trunk Plant Against a 100G Requirement
A colocation customer leases two adjacent cages in a 2014-era hall. Rack-to-MDA trunks are 12-fiber OM3 MPO assemblies; the longest channel measures 85 m and traverses two patch panels plus one cross-connect, so every channel contains six mated MPO pairs. The customer needs 96 ports of 100G now and 200 ports within three years, all inside the same row. Quotes are in: 100GBASE-SR4 is the cheapest optic, SWDM4 about 1.6× that price and duplex over two fibers, 100GBASE-CWDM4 about 2.1× and single-mode. Spare single-mode strands already reach the row MDA from the entrance room, but there is no single-mode inside the row itself. Burn-in of three sample SR4 links shows intermittent CRC errors under sustained load; the SLA permits no errored seconds. Capital budget is fixed for this fiscal year, and the lease has a three-year renewal option the customer intends to exercise.
1. Does 100GBASE-SR4 satisfy the requirement over this existing plant? Show your reasoning and state what you would tell the customer.
Read the worked answer
SR4 fails on reach before loss is even argued. IEEE 802.3bm rates 100GBASE-SR4 for 70 m on OM3; the longest channel is 85 m, so it cannot be certified regardless of how clean the connectors are. Loss is also thin: 85 m of OM3 at roughly 3.5 dB/km is about 0.30 dB, six mated MPO pairs at a realistic 0.35–0.50 dB each add 2.1–3.0 dB, and the budget is on the order of 2.4 dB (confirm against the actual transceiver datasheet before quoting a number). That leaves nothing for dust, thermal drift, or re-mate after a maintenance window.
The intermittent CRCs are exactly the signature of a channel running at the edge: link training succeeds when cold, then degrades under load and temperature. Tell the customer this is a specification failure, not an installation defect, and that no amount of re-polishing or re-testing converts an 85 m OM3 channel into a compliant SR4 channel.
2. Recommend a media and optic strategy for 96 ports now and 200 in three years. Justify with fiber counts, capex, and risk.
Read the worked answer
Fiber count is the argument that usually wins. 100G parallel optics consume 8 fibers per link, so 96 ports need 768 fibers (64 twelve-fiber trunk equivalents); 200 ports would need 1,600 fibers. Duplex 100G (SWDM4 or CWDM4) uses 2 fibers per link: 192 fibers now and 400 at the three-year mark, or 16 and 34 trunk equivalents. The existing row cannot supply 64 trunks without new tray work.
Recommend installing single-mode inside the row and using CWDM4 (2 km, duplex). It pays for itself in trunk count, has distance and loss headroom the multimode options do not, and survives the 200-port step without re-cabling. If capex genuinely blocks the in-row single-mode pull this fiscal year, SWDM4 is the defensible stopgap: it is specified to 100 m on OM3, so an 85 m channel fits with margin, and it is duplex, so it halves fiber demand. But it commits the customer to a multimode row and a re-pull at the next refresh - state that in writing, with the lease term as the risk boundary.
3. Whatever multimode path is chosen, what must be verified before any link is accepted?
Read the worked answer
Do not accept 'it links, therefore it passes.' Require a per-channel loss calculation using measured, not catalog, values: length, actual mated-pair count, and connector attenuation from a sample of the installed plant. Then require 100% OLTS testing at the operating wavelength with a calibrated meter and reference cords, recording both directions per TIA-526-14 for multimode.
Set a policy margin - typically at least 1 dB of headroom against the transceiver budget, more if the channel crosses multiple cross-connects that will be re-mated during MACs. Require end-face inspection and cleaning on every MPO before the acceptance test, with photographic records, because a single contaminated ferrule in a 12-fiber array kills four links at once. Finally, require the certification report to show fiber mapping and polarity per channel, not just loss; an SR4 channel can be dark from polarity and still show clean loss numbers.