Speed of light in optical fibre
The speed of light in optical fibre is approximately 200,000 km/s — a useful engineering rule of thumb. The precise value depends on the refractive index (n) of the glass: for standard single-mode fibre (SMF-28), n ≈ 1.467, giving `v = c / n ≈ 204,400 km/s`. For multi-mode fibre, n is slightly higher (~1.48), giving ~203,000 km/s. The 200,000 km/s approximation is accurate enough for latency budget calculations and is the value used in all major distributed systems references. This physical constant sets the absolute lower bound on network latency regardless of software optimisation, hardware upgrades, or CDN placement. A London–New York round-trip (approximately 11,000 km of fibre) has a theoretical minimum RTT of **~70ms** using the precise value; actual measured RTT is ~80 ms due to routing overhead and equipment processing.
The 54ms figure assumes a straight-line geodesic distance; real submarine cables follow Great Circle routes along continental shelves, adding significant physical distance. 70ms is the accepted real-world lower bound before any switching or queuing is added.
Formula
RTT_minimum = 2 × distance / speed_of_light_in_fibre
London-NY RTT ≈ 2 × 11,000 km / 200,000 km/s ≈ 70msWhy it matters in practice
The speed of light explains why geographic distribution is not optional for latency-sensitive systems serving global users. No amount of code optimisation will reduce a transatlantic request below the fibre-limited floor of roughly **~70ms** RTT before switching and queuing — and real paths are higher (~80ms+). This is why CDN edge nodes, regional database replicas, and multi-region deployments exist — they bring the data physically closer to the user. For systems that serve users on multiple continents, any shared state (session data, real-time inventory) must be either replicated regionally or accepted as a latency cost.
Common mistakes
- •Planning sub-10 ms SLOs for services that make transatlantic calls — physics makes this impossible.
- •Not accounting for routing overhead when estimating inter-region latency — actual latency is typically 30–50% higher than the theoretical light-speed minimum.
- •Assuming CDN caching eliminates origin latency — dynamic, uncacheable requests still pay the full geographic RTT to the origin.