Introduction

Data Center Interconnect is the optical transport that links two or more data centers so they can replicate storage, migrate workloads, and fail over without losing state. It spans three distance regimes that each pull a different technology: intra-city and campus runs under a few kilometres on short-reach gray optics, metro links of 10 to 80 km on either passive DWDM or 400ZR-class coherent pluggables, and regional or long-haul routes beyond 500 km on full coherent line systems. This guide works through the capacity and power-budget math that decides what fits, the components and impairments behind those limits, the architecture choices an operator actually makes, and two calculators to run a budget and weigh the options.

1. Fundamentals and Core Concepts

DCI carries high-capacity, low-latency optical transport between sites using wavelength division multiplexing, coherent optics, and pluggable modules from 100G to 800G and beyond. The business case is concrete: real-time replication between a primary and a backup site for disaster recovery, workload distribution to balance compute, geographic placement to cut user latency, and capacity growth by adding interconnected facilities rather than rebuilding one. Each driver translates to a number on the link budget — replication wants bounded latency, distribution wants aggregate capacity, and growth wants spare wavelengths.

The infrastructure stake is large: hyperscale clouds run DCI meshes linking hundreds of facilities, 5G cores and edge nodes depend on it for backhaul, and hybrid multi-cloud architectures rely on it for interconnection. A useful mental model is a road network — wider roads carry more traffic, shorter routes cut delay, and the topology decides how well the whole system uses its capacity.

2. Mathematical Framework

Each step up the constellation buys spectral efficiency and costs OSNR, which is why short links run dense formats and long links drop back to QPSK. Across an amplified chain OSNR adds reciprocally, 1/OSNR_total = Σ 1/OSNR_i.

At 17 ps/(nm·km) the tolerance scales as 1/B²: roughly 1200 ps/nm at 10G, 75 ps/nm at 40G, 12 ps/nm at 100G direct-detect. Coherent systems sidestep the limit entirely — DSP compensates dispersion digitally, which is the single biggest reason coherent replaced dispersion-managed line systems.

3. Types and Components

Transceivers

The module converts electrical to optical and back. QSFP-DD hosts 400G across SR8/DR4/FR4/ZR variants; OSFP adds thermal headroom for 800G; CFP2-DCO carries digital coherent for long-haul. The full pluggable family spans direct-detect short reach through coherent long-haul.

DWDM elements and coherent systems

A DWDM layer adds transponders or muxponders, passive or active mux/demux, EDFAs for span loss, and ROADMs for wavelength routing and add/drop — on a 50 GHz grid that is 80+ channels in the C-band, or flex-grid for super-channels. Coherent transmission multiplies it: polarization multiplexing doubles capacity, advanced modulation trades reach for spectral efficiency, and DSP compensates dispersion, PMD, and some nonlinearity. Coherent reaches 1000+ km against under 120 km for direct detection, at 2–8 bits/s/Hz versus 1–2, for 15–25 W versus 3–14 W.

4. Effects and Impacts

Four physical mechanisms degrade a DCI link. Fiber attenuation (0.2–0.25 dB/km at 1550 nm) sets the unamplified reach — 80 km at 10G, dropping to a few kilometres at 400G direct-detect, but 120 km on a coherent ZR. Chromatic dispersion (~17 ps/(nm·km)) broadens pulses and scales with the square of bit rate. PMD (0.01–0.5 ps/√km) randomly splits the polarizations and matters above 40G. And above a few dBm per channel the Kerr nonlinearities appear.

5. Techniques and Solutions

Four implementation models, in rising order of reach and cost. Gray optics use standard client transceivers, cheapest and simplest but one wavelength per fiber and under ~10 km — good for intra-DC and campus. DWDM with external transponders gives the most wavelengths (40–80+) and full protocol flexibility at the highest capital cost and complexity — the carrier choice. Pluggable colored DWDM puts the wavelength in the module, eliminating the transponder shelf for moderate channel counts. Coherent pluggables (ZR/ZR+/DCO) carry the longest reach at the highest per-port cost and power, and underpin the IP-over-DWDM model where the wavelength lives in the router line card.

6. Design Methodology

Work in four phases. Define requirements — capacity now and on a 3–5 year forecast, distance and fiber quality, availability target, and the cost and power budget. Select technology against distance and capacity. Build the power budget. Then verify OSNR for amplified routes.

For amplified routes verify OSNR with the engineering form — OSNR_span = 58 + P_ch − L_span − NF, then subtract 10·log(N) for N spans; the +58 carries the 0.1 nm reference, so never subtract a bandwidth term again. A 400G 16QAM channel wants about 22 dB at the receiver. An unamplified 80 km metro span has no EDFA ASE, so OSNR there is set by transmitter and receiver noise and typically exceeds 30 dB.

7. Interactive Calculators

Two tools: a link-budget calculator that breaks the loss down and tests it against receiver sensitivity, and a heuristic technology-fit scorer that weighs five DCI options against a distance, capacity, and cost-versus-performance priority. The second is a planning aid — it ranks options, it does not replace the budget the first one computes.

8. Applications and Case Studies

Main Points