A QSFP28 module labelled "100G" tells you almost nothing about whether it will work on your link. The suffix after the form factor - SR4, PSM4, DR, CWDM4, FR1, LR4, LR1 - is what determines the fiber type, the connector, the number of optical lanes, the modulation and whether forward error correction is required. Two modules can share a cage shape, a connector and a nominal reach and still be incapable of passing a single frame between them.
That is the single most expensive mistake in 100G procurement, and it is entirely avoidable. This guide works through the QSFP28 optical interfaces that are actually in production, explains which of them can talk to each other, and gives a selection and qualification sequence that network engineers and buyers can apply to a specific link rather than to a category.
QSFP28 Selection at a Glance
Use the table below to build a shortlist, not to place an order. Exact reach, power class, FEC behaviour and platform support still have to be confirmed against the module datasheet and the switch documentation for both endpoints.
| Connection requirement | Common option | Fiber or cable | Connector | Condition that decides it |
|---|---|---|---|---|
| Very short device-to-device link | Passive QSFP28 DAC | Twinax copper | Fixed QSFP28 ends | Both devices must support the specific cable and length |
| Short link with easier routing | QSFP28 AOC | Integrated optical cable | Fixed QSFP28 ends | A single end failure replaces the whole assembly |
| Up to 70 m on OM3, 100 m on OM4 | 100GBASE-SR4 | Parallel multimode | MPO-12 | Requires an MPO plant with correct polarity |
| Reuse of duplex multimode | SR1.2 or vendor BiDi | Duplex MMF | Duplex LC | Both ends must be the same interface family |
| Up to 500 m on parallel SMF | PSM4 | Parallel single-mode | MPO-12 | Consumes eight fibers; not interchangeable with DR |
| Up to 500 m on duplex SMF | 100GBASE-DR | Duplex single-mode | Duplex LC | Single-wavelength PAM4, FEC required |
| Up to 2 km on duplex SMF | CWDM4 | Duplex single-mode | Duplex LC | Four wavelengths; will not link to FR1 |
| Up to 2 km on duplex SMF | 100GBASE-FR1 | Duplex single-mode | Duplex LC | Single-wavelength PAM4, FEC required |
| Up to 10 km on duplex SMF | 100GBASE-LR4 | Duplex single-mode | Duplex LC | Four wavelengths; the IEEE PMD does not itself use FEC |
| Up to 10 km on duplex SMF | 100GBASE-LR1 | Duplex single-mode | Duplex LC | Single-wavelength PAM4, FEC required |
| Extended campus or metro | ER4-Lite, ER1, ZR4 | Duplex single-mode | Usually duplex LC | Needs a full budget, dispersion and platform review |
These reach and connector combinations line up with the vendor portfolios shipping today. Cisco's 100GBASE QSFP-100G modules datasheet (accessed July 2026) lists SR4 at 70 m over OM3 and 100 m over OM4, PSM4 and DR at 500 m, CWDM4 and FR1 at 2 km, and LR4 and LR1 at 10 km - and, importantly, documents that FEC behaviour differs by optical interface rather than by form factor.
What a QSFP28 Transceiver Actually Defines
QSFP28 stands for Quad Small Form-factor Pluggable 28. "Quad" refers to four high-speed electrical lanes on the host side, "Small Form-factor Pluggable" to the compact hot-swappable module format, and "28" to the approximate signalling class of each electrical lane rather than to 28 Gbps of user data. In a classic 100G implementation the host drives four roughly 25 Gbps electrical lanes into the module, and the module converts them into whatever optical or copper interface it implements.
That is the crucial point: QSFP28 defines the mechanical envelope, the host electrical interface and the management interface. It does not define one optical technology. The same cage can hold four parallel multimode lanes, four parallel single-mode lanes, four multiplexed wavelengths on one fiber pair, a single 100G wavelength, passive copper conductors or an active cable assembly. The management side is standardised separately: SFF-8636 covers the management interface for four-lane modules and cables, while SFF-8679 covers the general hardware and electrical specification across the QSFP family. Both are maintained as SNIA SFF specifications.
Four-Lane NRZ Versus Single-Wavelength PAM4
Traditional QSFP28 optics - SR4, PSM4, CWDM4, LR4 - carry the 100G signal as four optical lanes or four wavelengths, each using non-return-to-zero (NRZ) signalling, where one symbol carries one bit.
The newer single-wavelength interfaces - DR, FR1, LR1 - put the entire 100G signal on one optical wavelength using four-level pulse amplitude modulation (PAM4), where one symbol carries two bits. An internal digital signal processor performs the conversion between the host's electrical lanes and the single optical lane. IEEE standardised the 2 km and 10 km single-wavelength interfaces as 100GBASE-FR1 and 100GBASE-LR1 in IEEE Std 802.3cu-2021, after the 100G Lambda MSA published the 100G-FR and 100G-LR specifications that preceded them.
PAM4 packs more bits into the same baud rate, but with four amplitude levels instead of two the receiver has far less margin between levels. That is why single-wavelength 100G optics rely on forward error correction as a normal part of operation, while four-wavelength NRZ interfaces such as LR4 do not require it at the PMD level. The distinction propagates into interoperability, host configuration, power, latency and the migration path to 400G breakout - which is why it appears again in almost every section below.
QSFP28 Compared With SFP28, QSFP+, QSFP56 and QSFP-DD
| Form factor | Typical lane structure | Common Ethernet rate | Where it is used |
|---|---|---|---|
| SFP28 | 1 × 25G | 25G | Server, storage and access links |
| QSFP+ | 4 × 10G | 40G | Legacy aggregation and uplinks |
| QSFP28 | 4 × 25G | 100G | Data center, enterprise core, telecom |
| QSFP56 | 4 × 50G PAM4 | 200G | Higher-speed aggregation |
| QSFP-DD | 8 electrical lanes | 400G and above | High-density spine, AI and HPC fabrics |
Physical fit proves nothing. A QSFP28 module cannot form a 100G link in a QSFP+ port, because the older host cannot drive 25G per lane. The reverse case - a QSFP+ module running at 40G in a QSFP28 port - is often supported, but it is a platform and software decision, not a property of the cage. The same caution applies to the higher-density formats: some QSFP-DD and QSFP56 ports will operate a QSFP28 module, and some will only do so on particular port groups or software releases. Confirm it against the specific platform documentation rather than the connector geometry.
The Main QSFP28 Optical Interfaces
SR4 - Parallel Multimode
100GBASE-SR4 splits the signal across four transmit and four receive fibers on OM3 or OM4 multimode fiber, terminated with an MPO-12 connector, reaching 70 m and 100 m respectively. It is the default for switch-to-switch links inside a data hall where a parallel multimode plant already exists.
The trap is choosing SR4 simply because the link is short. If the building is wired with duplex LC multimode patch panels, SR4 forces either a recabling project or a fan-out arrangement using MPO/MTP patch cords and cassettes. Both add insertion loss and both add cost that rarely appears in the module comparison spreadsheet.
SR1.2 and Vendor BiDi - Reusing Duplex Multimode
Bidirectional and single-lane multimode modules carry 100G over a duplex LC multimode pair, which lets an existing 10G or 40G duplex plant carry 100G without new trunk cabling. They are the right answer when replacing LC patch panels with MPO infrastructure would be disruptive and both switches support the same module family.
"100G over duplex multimode" is not one standard, though. Wavelength plans differ between implementations, so a module from one vendor's BiDi family may not pair with another's. If this is the direction you are heading, it is worth understanding how BiDi transceivers split and pair wavelengths before committing to a mixed-vendor design.
PSM4 - Parallel Single-Mode
PSM4 carries four parallel 25G optical lanes over single-mode fiber with an MPO-12 connector, reaching roughly 500 m. It suits sites that already have parallel single-mode trunks, but it burns eight fibers per link where a duplex interface uses two. Over a large fabric that difference dominates the cabling budget.
PSM4 and DR are not alternatives to each other in the interoperability sense. They overlap in reach and nothing else: PSM4 is four NRZ lanes on eight fibers, DR is one PAM4 wavelength on two.
DR - Single Wavelength to 500 m
100GBASE-DR puts the whole 100G signal on one wavelength over a duplex single-mode pair for links up to 500 m. It fits leaf-spine connectivity in facilities with duplex LC structured cabling, and it is the natural 100G endpoint for architectures that will later break out a 400GBASE-DR4 port into four 100G links. It requires host FEC and platform support, so it is a poor choice for older switches that predate single-wavelength optics.
CWDM4 - Four Wavelengths to 2 km
CWDM4 multiplexes four 25G optical lanes onto four coarse wavelengths carried on a single duplex single-mode pair, reaching 2 km. The CWDM4 MSA technical specification defines the optical parameters; note that the MSA deliberately leaves the high-speed electrical characteristics to the host-side standards. CWDM4 remains a practical, widely available choice for data center interconnect and campus links that do not need the full 10 km of LR4, and it works over the same OS2 single-mode fiber as its single-wavelength competitors.
FR1 - One Wavelength to 2 km
100GBASE-FR1 covers the same nominal 2 km over the same duplex single-mode fiber with the same LC connectors as CWDM4 - using a completely different optical interface. One PAM4 wavelength with FEC, instead of four NRZ wavelengths without.
Because everything visible from the outside matches, this pair produces more field failures than any other in the QSFP28 range. A CWDM4 module needs a CWDM4-compatible endpoint; an FR1 module needs an FR1-compatible endpoint. Decide between them on the interface installed at the far end, the FEC the hosts can enforce, the switch software, the power and thermal envelope of the port, and whether a 400G breakout migration is planned - in that order.
LR4 and LR1 - Two Different Routes to 10 km
100GBASE-LR4 carries four wavelengths over duplex single-mode fiber for 10 km and is the established interface for campus backbones, inter-building links and enterprise cores. The IEEE LR4 PMD does not itself use FEC, which is why LR4 modules interoperate cleanly with a large installed base going back years.
100GBASE-LR1 reaches the same nominal 10 km with one PAM4 wavelength and FEC. Choose LR1 when both endpoints support single-wavelength optics, the required FEC is available and configurable, and the network roadmap favours that direction. Choose LR4 when the link has to interoperate with existing four-wavelength LR4 equipment. The two are not directly connectable, regardless of the matching connectors.
ER4-Lite, ER1 and ZR4 - Extended Reach
Extended-reach QSFP28 optics require real link engineering rather than a distance lookup. Depending on the implementation they may require host FEC, a minimum link attenuation as well as a maximum, dispersion assessment, receiver-overload protection, specific switch firmware and platform-specific qualification. Cisco, for example, documents an ER4-Lite implementation reaching 40 km with host FEC and 30 km without, and a ZR4 implementation that requires host FEC.
The minimum-attenuation requirement catches people out. On a short lab link, an extended-reach module can overload its own receiver - the link fails not from too little light but from too much, and needs a fixed attenuator to work at all.
Which QSFP28 Types Can Talk to Each Other?
The general rule is simple: both endpoints must implement the same optical interface. The exceptions are worth knowing precisely, because they are where vendor documentation adds value beyond the standards.
| Endpoint A | Endpoint B | Interoperable? | Reason |
|---|---|---|---|
| SR4 | SR4 | Yes, when specifications match | Same parallel multimode interface |
| SR4 | SR1.2 or BiDi | No | Different fiber count and optical signalling |
| PSM4 | PSM4 | Yes, when MSA and polarity match | Same parallel single-mode interface |
| PSM4 | DR | No | Parallel NRZ versus single-wavelength PAM4 |
| CWDM4 | CWDM4 | Yes, when MSA specifications match | Same four-wavelength interface |
| CWDM4 | FR1 | No | Four-wavelength NRZ versus single-wavelength PAM4 |
| LR4 | LR4 | Yes, when IEEE specifications match | Same four-wavelength LR4 interface |
| LR4 | LR1 | No | Four-wavelength NRZ versus single-wavelength PAM4 |
| DR | FR1 | Vendor and distance dependent | Same single-wavelength PAM4 family, different reach class |
| DR or FR1 | 400G DR4 breakout lane | Often yes, when documented | Single-wavelength lanes of a 400G parallel module |
| Vendor BiDi A | Vendor BiDi B | Only when explicitly supported | Wavelength pairing may differ between families |
The DR-to-FR1 row deserves the extra words. Writing a flat "no" there is inaccurate: both are single-wavelength PAM4 interfaces in the same family, and vendors do publish supported combinations. Cisco's single-lambda 100G solution overview describes FR1 modules used in applications facing 100GBASE-DR and interoperating with DR4-compliant 400G modules in breakout. Rather than a yes or a no, judge each case against six things: whether each end is an IEEE or MSA standard interface, which end is DR versus FR1 versus LR1, the maximum permitted distance for the weaker of the two, whether transmit power and receive sensitivity actually overlap, whether the vendor lists the pair in its interoperability matrix, and whether a 400G breakout module is involved. Cisco's optics compatibility and interoperability matrix is the reference for its own products; other switch vendors publish equivalents.
Everything else in the table reduces to one sentence: a matching connector is not evidence of a matching interface. Two modules can both terminate in duplex LC on OS2 fiber and transmit signals that have nothing in common.
The Seven-Step Selection Process
Step 1 - Pin Down the Host Port
Record the switch or NIC model, the line card, the port number, the operating system and firmware versions, the supported port speeds and the supported breakout modes. "Cisco QSFP28" or "Arista 100G" on a purchase request is not a specification - support varies between products from the same vendor, between line cards in the same chassis and between software releases on the same line card.
This step also covers the mechanical and electrical layers of compatibility: whether the cage accepts the module, and whether the host can supply the module's power class and drive the required lane rate. A module that fits can still fail electrically.
Step 2 - Identify the Interface at the Far End
Find out what is actually installed at the remote endpoint: SR4, PSM4, DR, CWDM4, FR1, LR4, LR1 or a vendor BiDi family. Unless that endpoint is also being replaced, the new module has to match it. Where the far end belongs to another team or another organisation, get the part number in writing rather than a reach figure.
Step 3 - Confirm Fiber, Connector and Polarity
Check the fiber grade (OM3, OM4, OM5, OS2), the connector type, the polish where relevant, the MPO gender and polarity, the number of spare fibers, and every patch panel, cassette and intermediate connection in the path. Parallel optics fail on polarity far more often than on loss, and MPO polarity conventions are easy to get wrong across a mixed plant - the terminology alone causes confusion, which is why it helps to be clear on what distinguishes MTP from MPO and how the polarity methods differ before ordering trunks.
A module can be perfectly compatible with the switch and still be unusable because the building has the wrong connector.
Step 4 - Work Out the Real Reach and Optical Budget
Use the routed length, including patch cords, trays, risers, building entrances, patch panels, splices and planned service loops. Then, for anything long or heavily patched, do the budget properly. The commonly quoted starting point is:
Available budget = minimum transmitter power − minimum receiver sensitivity
That check only tells you whether enough light arrives. A complete assessment also has to cover the maximum transmitter power against the receiver's overload threshold, both the minimum and maximum expected link loss, connector and splice loss across the whole path, an engineering margin for future repairs and re-splicing, chromatic dispersion and any optical penalties on longer spans, whether an attenuator is required on short extended-reach links, and whether the datasheet figures are per lane or for the aggregate link. The loss contributions are rarely where people expect them - insertion loss accumulates mostly at connections rather than in the fiber itself on typical data center distances.
Use the datasheet for the exact part number. Generic values from an article, including this one, are not a design input.
Step 5 - Resolve FEC Properly
Forward error correction is the step most often reduced to a single switch setting, and that is where links break. There are several distinct questions:
- Does the optical PMD itself require FEC? DR, FR1 and LR1 do; the IEEE LR4 PMD does not.
- Does the host electrical interface require FEC independently of the optics?
- Which FEC - RS-FEC as used for 25G/100G lanes, or a KR4-class scheme on the electrical side?
- Is the FEC performed inside the module, or expected from the host?
- What does the platform enable by default, and does that default change between software releases?
- Do breakout lanes carry their own FEC setting, separate from the parent port?
Confirm the answer for both endpoints. A blanket "enable RS-FEC on all 100G links" policy will break LR4 links as reliably as leaving FEC off will break FR1 links.
Step 6 - Check Coding, Software, Power and Monitoring
Verify the vendor coding or supported identifier, the switch compatibility list entry, the minimum software version, digital optical monitoring (DOM, also written DDM for digital diagnostic monitoring) support, the module temperature grade, the maximum module power the port can supply, breakout support on that exact port, and any required port configuration. The SFF-8636 management interface standardises the module inventory and diagnostic data, but whether a host accepts and enables a given module remains a platform decision.
This is also the point to settle warranty questions, because they are a software matter as much as a commercial one - see the procurement section below.
Step 7 - Qualify a Sample Before the Volume Order
Test the exact module SKU on the same switch model and software release you will run in production. Confirm detection and inventory data, port speed and FEC state. Read transmit power, receive power, module temperature and supply voltage, and save those numbers as the healthy baseline you will compare against later. Run sustained bidirectional traffic and watch CRC errors, FEC corrected and uncorrected codeword counts, and link-flap counters. Test a warm restart and a cold restart, and test breakout mode if you need it. Then record the accepted configuration as a whole - module, host, line card, software version, port mode, FEC mode.
A module that works in one lab switch is not qualified for every line card or every release. The record of what was tested is what makes the next purchase fast.
Three Failures That Keep Recurring
The duplex plant that bought parallel optics. A team specifies SR4 for a 60 m switch-to-switch link because the distance is short and the price is good. The floor is wired with duplex LC multimode to patch panels. The modules arrive, nothing can be connected, and the project stalls while MPO-to-LC breakout assemblies are sourced and the loss budget is recalculated. The correct answer was a duplex multimode interface, or a decision to invest in an MPO trunk with eyes open.
Two ends, 2 km apart, both duplex LC, no link. One site installed CWDM4, the other later standardised on FR1. Both are 2 km, both use OS2 and duplex LC, both modules power up and report normal transmit power. Neither receiver can decode the other's signal, because one is sending four NRZ wavelengths and the other one PAM4 wavelength. Nothing in the physical inspection reveals this; only the part numbers do.
The upgrade that rejected working optics. A pair of third-party modules runs for eighteen months without an error. A network operating system upgrade tightens module validation, and after the reload the ports report an unsupported transceiver. The modules did not change; the acceptance logic did. This is why qualification records should list the software release, and why compatibility should be retested before major upgrades rather than after them.
Beyond the Datasheet: The Commercial Side of QSFP28 Buying
Two modules can be technically equivalent and commercially very different. These are the dimensions that separate them.
| Dimension | What to ask | Why it changes the decision |
|---|---|---|
| Power consumption | Typical and maximum draw, and the module power class | Determines whether a fully populated line card stays within its port budget |
| Temperature grade | Commercial, extended or industrial range | Edge cabinets and telecom enclosures routinely exceed commercial limits |
| Cooling and airflow | Faceplate density, airflow direction, adjacent high-power modules | Thermal throttling and flapping often trace back to airflow, not optics |
| Lifecycle and supply | End-of-sale status, lead time, expected availability window | A discontinued interface makes spares expensive for years |
| Standard versus proprietary | Is the interface IEEE, MSA or vendor-specific? | Proprietary interfaces lock both ends of every future link |
| OEM versus third party | Support policy, coding method, upgrade behaviour | Cost saving is real; so is the software-acceptance risk shown above |
| Warranty, RMA and batches | Turnaround time, batch traceability, failure-rate data | A large deployment needs consistent batches, not just a good sample |
| Cabling cost for the same reach | Fibers consumed, trunk changes, panel work | PSM4 versus CWDM4 at 500 m is a cabling decision more than an optics one |
| Maintenance model | Replaceable module versus fixed assembly | An AOC failure replaces the whole cable; a module failure replaces one end |
Separate the two questions deliberately. "Will it work?" is answered by steps one to seven. "Should we buy it?" is answered by this table. A module can pass the first test and still be the wrong purchase.
DAC, AOC or Pluggable Optics?
| Option | Main advantage | Main limitation | Best fit |
|---|---|---|---|
| Passive DAC | Lowest cost, power and latency | Short, bulky, platform-specific | Same-rack links |
| Active DAC | Longer than passive copper | Still heavy and stiff to route | Adjacent racks |
| AOC | Light, flexible, no connector cleaning | Fixed ends; one failure replaces the assembly | Short row-level links |
| Pluggable optics | Replaceable, works with structured cabling | More components and connector maintenance | Patch panels and longer links |
| Breakout DAC or AOC | Direct 100G to 4 × 25G | Needs a supported port mode and lane mapping | 25G servers on 100G switches |
Choose copper when the devices are close, the approved length exists and port power matters. Choose an active optical cable when copper is too heavy to route and a fixed assembly is acceptable. Choose separate modules when the link crosses structured cabling and patch panels, when individual components must be replaceable, or when the optical type may change later.
Breakout deserves its own check. Whether one QSFP28 port can split into four 25G connections depends on the switch, the port, the software release and often the port group - enabling breakout on one port can change the behaviour of its neighbours. On the passive side, the same applies to the cabling: an MPO breakout assembly has to match the lane mapping the platform expects, not just the fiber count.
Troubleshooting Common QSFP28 Problems
| Symptom | Likely causes to check | Corrective action |
|---|---|---|
| Module not detected | Coding, firmware, power class, damaged module | Check the platform matrix, system logs and module inventory |
| Detected but no link | Optical interface mismatch, wrong fiber, transmit disabled, FEC | Compare the part numbers at both ends before touching anything else |
| Low receive power | Dirty connector, excess loss, tight bend, poor splice | Clean, inspect and measure the optical path |
| High CRC error count | Thin optical margin, contamination, FEC mismatch, cable damage | Review DOM values against the saved baseline and check error counters |
| Intermittent flapping | Thermal conditions, marginal receive level, software issue, FEC | Correlate temperature, power and event logs over time |
| Only some breakout lanes work | Unsupported breakout mode, lane mapping, cable fault, port group | Verify the port map and test lanes individually |
| High module temperature | Airflow, high-power module, blocked intake, power class | Correct airflow and confirm host power support |
| Right connector, no link | Different optical standards on a matching interface | Check CWDM4 against FR1, LR4 against LR1, SR4 against BiDi |
| Short extended-reach link fails | Receiver overload from too much input power | Add the attenuation the datasheet specifies as a minimum |
Work by controlled isolation: change one component, record the result, then change the next. Swapping the module, the patch cord and the port at the same time will often restore the link and will always destroy the evidence of what caused the fault.
Is QSFP28 Still the Right Choice?
Yes, where the required link rate is 100G and compatible 100G infrastructure already exists - brownfield data center upgrades, enterprise core links, 25G server breakout, telecom and edge sites, and cost-controlled expansion on installed OM4 or OS2 plant.
Look at 200G, 400G or higher-density platforms instead when the design is greenfield, when spine-port bandwidth is the binding constraint, when the switch roadmap already centres on QSFP-DD or OSFP, or when a large share of the network would need another upgrade within a couple of years anyway. Some higher-density ports will run QSFP28 modules, but confirm it per platform and per port rather than assuming it from the cage.
FAQ
Q: Is QSFP28 the same thing as 100G?
A: No. QSFP28 is a form factor and host interface that commonly carries 100G. It says nothing about the optical interface - SR4, PSM4, DR, CWDM4, FR1, LR4 and LR1 all run at 100G in a QSFP28 shell using different fibers, connectors and signalling.
Q: Can a QSFP28 module work in a QSFP+ port, or the reverse?
A: A QSFP28 module cannot form a 100G link in a QSFP+ port, because that host cannot drive 25G per lane. Many QSFP28 ports will run a QSFP+ module at 40G, but this depends on the switch, the port and the software configuration.
Q: What is the practical difference between CWDM4 and FR1?
A: Both cover roughly 2 km over duplex single-mode fiber with LC connectors. CWDM4 uses four multiplexed wavelengths without a PMD-level FEC requirement; FR1 uses one PAM4 wavelength and requires FEC. They will not link to each other, so the deciding factor is what is installed at the far end and whether your hosts can enforce the right FEC mode.
Q: Does every QSFP28 link need FEC?
A: No, and applying it universally causes failures. The requirement follows the optical interface and the host electrical interface, not the form factor. See step five above for the six questions that actually settle it.
Q: Can third-party QSFP28 modules run in branded switches?
A: Often, when the coding, hardware, software release, optical interface, FEC and port configuration are all accepted by the platform. The technical risk is manageable; the change-over-time risk is the real one, since a software upgrade can alter acceptance behaviour. Qualify the exact combination and retest before major upgrades.
Q: What should I send a supplier when requesting a quote?
A: Switch or NIC model, line card, software version, port speed, routed link distance, fiber type, connector and polarity, the optical interface at the remote end, breakout requirement, FEC mode and operating temperature range. A quote based on "100G, 2 km" is a guess.
Q: Which single check prevents the most failures?
A: Getting the exact part number of the module at the far end, in writing, before ordering. Reach and connector type are not enough to identify an optical interface.
Summary
The safest QSFP28 purchase is not the one with the longest reach or the lowest unit price. It is the one that has been matched against the whole link - both hosts and their software, both optical interfaces, the installed fiber and connectors, the FEC configuration, the port power and the operating environment - and then tested in that exact combination before the volume order is placed.
