A question I often get asked is what to do if the distances between a 400G switch to the 100G servers is greater than 7 meters. A quick answer is to use 400G DR4 / DR4+ / DR4++ optical transceivers as shown here:
These optical transceivers apply 100G over single-wavelength (100G Lambda) technology specified in IEEE 802.3 standard and 100G Lambda Multi-Source Agreement (MSA). IEEE 802.3cd standard defines 400GBASE-DR4 and 100GBASE-DR for 500m reach.
100G Lambda MSA defines 100Gbps per optical channel for 2km and 10km reaches to support 400G and 100G applications. In these specifications, Pulse Amplitude Modulation 4-level (PAM4) signaling and encoding, and Forward Error Correction (FEC) are applied.
What is special in these technologies? What problem do these technologies try to solve?
Most 100G optical transceivers – e.g., 100GBASE-SR4, LR4, CWDM4, PSM4 and ER4 – use 25Gbps non-Return to Zero (NRZ) signaling and encoding for each optical lane and have total 4 optical lanes. If applying this 4x 25G NRZ optical interface structure to 400G, it would require 16 optical lanes, which is impractical from economic and technical perspectives.
To implement 4x 100G format, 100G per optical lane (or 100G lambda in another term) is needed, and the more efficient signal modulation PAM4 is applied. In brief, the 400G DR4/DR4+/DR4++ optics operate at 4 parallel optical lanes with 100 Gbps PAM4 signaling per optical lane.
The 400G DR4/DR4+/DR4++ optics can be very helpful in Data Center networking to achieve high port density. The 400G QSFP-DD DR4/DR4+/DR4++ transceivers can be deployed in the 400G spine switches and the 100G QSFP28 DR1/FR1/LR1 optics can be deployed in the 100G leaf switches as illustrated in this diagram . . .
The fiber length of the fabric between spine and leaf can be flexible, either 500m, 2km or 10km, depending on the technology chosen. The set-up you use is important in speed, efficiency and distance.