Ask a room of network engineers how to choose an SFP+ module and most will start with the data rate — which is exactly backwards. The rate is printed on the port you already own; it is the one fact you cannot get wrong by accident. Module selection actually fails on five other questions — host, standard, wavelength, reach, budget — each of which has sunk real orders. This guide walks them in the order a professional quotation asks them. (A framing note: in this manufacturer's hubs, the module word is heavily used; this guide concerns the module as a component choice, and it applies whether the module lands in a switch, a converter or an OLT.)
Question one: what host, exactly?
SFP+ is a form factor and an electrical interface, not a guarantee of universal acceptance. Hosts differ in what they validate, what power class they supply, and what they were designed alongside — the module families in this catalogue include lines documented by chipset lineage (the HiSilicon-based PON family) precisely because host alignment is a real property. The first line of a module inquiry is the host's make and model. Any seller who quotes modules without asking it is guessing with your money.
Question two: which standard lives on the link?
'10G' is a rate, not a standard. The same SFP+ cage can be asked to carry 10G Ethernet, a PON line (XGS-PON's rates sit at the same 9.95 Gbps neighbourhood in entirely different clothing), a video signal in the SDI world, or Fibre Channel in a storage plant. These modules are not interchangeable despite sharing a cage and a rate class. State the service, not the speed: the standard determines the module's protocol behaviour, and no amount of rate-matching rescues the wrong one.
Question three: what is the wavelength plan?
Every link is two modules agreeing on light. Dual-fiber links use one wavelength both ways on separate strands (1310 nm is the single-mode workhorse; 850 nm the multimode one). Single-fiber BiDi links pair two wavelengths — one per direction — so the two ends are deliberately different modules, ordered as a matched pair (the family documents BiDi variants, for instance the 25G SFP28 LR BiDi pair). DWDM plans assign channels from a grid. The question to answer is not 'which wavelength' in isolation but 'what is this link's place in the plant's wavelength map' — including what the far end already has, if it exists.
Question four: what reach, over what fiber, with what margin?
Reach classes are budgets, not distances. A '10 km' module rating assumes a loss budget spent across fiber attenuation, connectors, splices and patch panels, with margin for ageing and repairs. The specification the seller needs is the span's real loss picture: fiber type (and grade, for multimode), path length, and how many connections live in it. On PON modules the same discipline wears class names — the budget classes (C+, D, the XGS-PON E/N classes) are reached by exactly this arithmetic against split ratio. Whatever the family, 'the building is about 300 metres' is the beginning of the answer, not the answer.
Question five: what environment and what fleet?
Two closing questions decide between the modules that survive the first four. Environment: temperature range and enclosure reality separate commercial from industrial-grade selections in plants that live in cabinets and outdoors. Fleet: modules are spares-shelf items, and every additional variant stocked is a variant that can be grabbed wrongly at 3 a.m. If the plant already standardises on a module for this duty, 'different but marginally better' needs a strong argument. State the existing fleet in the inquiry and let consistency compete fairly against novelty.
- Host make and model, first line of the inquiry.
- The service standard the link carries (Ethernet / PON class / video / storage).
- Wavelength plan: dual-fiber or BiDi pair, channel if DWDM, far-end reality.
- The span's loss budget: fiber type and grade, length, connection count, margin.
- Environment class and the existing fleet standard.
Five questions, and the rate never needed asking. That is not a trick — it is the shape of a component whose visible specification (10G, SFP+) is the least informative thing about it. Answer the five in your inquiry, require the answers to be confirmed against the exact module in writing before payment, and the small metal cage will do what it has quietly done for two decades: disappear into a working link.
The two failure stories everyone recognises
Story one: the modules arrive, the link does not train, and after a day of cable-swapping somebody reads the label closely — the pair are BiDi partners ordered as two identical units, both 'A' ends, each transmitting on the wavelength the other cannot hear. The fix is a purchase order; the lesson is that BiDi ends are a pair, not a quantity. Story two: a link trains beautifully at installation and degrades across its first summer, because the modules were reach-classed on the floor-plan distance while the actual routed span — up the riser, across the tray, down the far riser, through three patch panels — was half again as long and twice as lossy. The fix is a higher-budget class; the lesson is that reach is loss, and loss lives in the route, not the map. Both stories are why the five questions are asked in writing and answered against the exact module before payment: they are the cheapest engineering review in the entire project.
A worked selection, end to end
A campus aggregation link makes the five questions tangible. Host: a named aggregation switch model, stated. Service: 10G Ethernet. Wavelength plan: dual-fiber single-mode, 1310 nm, because the plant standardises on it and the far end (an existing switch) already carries that optic class. Reach: the routed span measures 6.8 km with four patch points — inside the loss budget of the 10 km LR class with margin, beyond any SR discussion. Environment and fleet: indoor racks both ends, and the fleet standard for this duty is that same LR class, so the shelf gains no new variant. Five answers, one module, zero drama — and, just as usefully, the same five answers with 'PON' or 'BiDi' substituted in questions two and three select the access-network modules with equal finality. The method is the product; the module is its output.
A note on price, since it steers so many selections astray: module price differences that look significant on a quote line are usually the smallest money in the link's life. The expensive module events are the truck roll to a dark link, the mis-ordered pair discovered at installation, and the non-standard variant that nobody stocks. Optimising the purchase line while discounting those is how a 10% saving becomes a 300% incident. Buy the module the five questions select; negotiate afterwards if you must.