When the first 100G standard for Ethernet networks, the transceiver industry launched an alphabet soup of form factors. The CFP emerged first, "C" for 100, and FP for "Form factor, Pluggable." Like the early versions of 10G transceivers, the CFP was huge. When compared to the most popular 40G form factor, the QSFP, front-panel density decreased by a factor of three. Most CFP implementations doubled the power consumption per bit. And, if those two disadvantages were not enough, the price per bit increased by a factor of ten.

The next version of form factors, the CFP2, CFP4, and the CPAK, improved upon the CFP. But when compared to the popular 10G SFP+ and 40G QSFP+, none of these new members of the CFP family improved density, power consumption, or cost.

Enter the 100G QSFP28. The QSFP28 is the exact same footprint as the 40G QSFP+. The "Q" is for "Quad"; just as the 40G QSFP+ is implemented using four 10-Gbps lanes, the 100G QSFP28 is implemented with four 25-Gbps lanes. In all QSFP versions, both the electrical lanes and the optical lanes operate at the same speed, eliminating the costly gearbox found in CFP, CFP2, and the CPAK. The QSFP28 module has an upgraded electrical interface to support signaling up to 28 Gbps signals, yet keeps all of the physical dimensions of its predecessor.

The 100G QSFP28 makes it as easy to deploy 100G networks as 10G networks. When compared to any of the other alternatives, 100G QSFP28 increases density and decreases power and price per bit. It is fast becoming the universal data center form factor. Here are some of the reasons.

The QSFP28 increases front-panel density by 250% over QSFP+. The form factor is the same and the maximum number of ports is the same, but the lane speeds are increased from 10 Gbps to 25 Gbps. The increase in panel density is even more dramatic when compared to some of the other 100-Gbps form factors: 450% versus the CFP2 and 360% versus the CPAK.

Like the QSFP+, the same form factor supports both cables and transceivers. In the first generation of 100G switches and routers, the smaller CXP form factor was used for cabling and the CFP or CFP2 was used for transceivers. This forced the equipment designer to make huge sacrifices. A switch with CXP ports could not be used in any data center with singlemode fiber; a router with CFP2 or CPAK ports was limited in bandwidth by the 8-10 ports that could fit on the front panel.

The QSFP28 resolves this problem. A 1 rack-unit (RU) switch can accommodate up to 36 QSFP ports on the front face plate. Many varieties of either transceivers or cables can plug into these ports. The cables can be either direct-attach copper cables, commonly referred to as DACs, or active optical cables (AOCs). DACs offer the lowest cost but are limited in reach to perhaps 3 m. They are typically used within the racks of the data center, or as chassis-to-chassis interconnect in large switch and routers. AOCs are much lighter and offer longer reaches up to 100 m and more. Customers like AOCs because they are much cheaper than optical transceivers.

QSFP28 transceivers can be based on either VCSELs (useful for shorter distances on multimode fiber) or silicon photonics (for longer distances on singlemode fiber). The advent of silicon photonics enables QSFP28 transceivers to support any data center reach up to 2 km or more. Silicon photonics provides a high degree of integration; the CMOS chips are small enough to fit within a QSFP package. Silicon photonics is low-power; even WDM designs can fit within the 3.5 W maximum of QSFP.

With all of the technology choices available in the same form factor, the coming generations of high-bandwidth switches, routers, and adapters will all feature QSFP28 ports, ensuring data centers can scale to 100G networks with the simplicity as 10G networks.