Cross-Connects And Exchange Network Topology

A cross-connect is a dedicated physical cable — usually fiber — run directly between a trading firm's server rack and the exchange's matching engine within the same colocation data center, bypassing the shared internet or a general network switch entirely. Firms pay the exchange a recurring fee for each cross-connect, and its physical cable length (which can differ by a few meters between racks in different parts of the data center hall) translates directly into a small but real latency difference, since signals travel through fiber at a fixed fraction of the speed of light.

Exchange topology refers to how the matching engine's network is architected: whether every colocated rack is truly equidistant from the matching engine, or whether some physical positions have a structural cable-length advantage over others. Exchanges that care about latency fairness design topologies where every cross-connect is cut to the same length regardless of physical rack position, so no client gains an edge purely from being assigned a favorable rack; exchanges that don't equalize cable lengths create a race for the physically closest racks, which is itself a form of unlevel latency competition unrelated to any trading skill.

The latency gap involved is tiny in absolute terms — a few nanoseconds per meter of extra fiber — but tiny is exactly what matters in a market where the fastest response to a price update wins the race to trade on it, so firms have historically paid premiums for specific rack positions, and litigation and regulatory scrutiny have both targeted exchanges accused of not disclosing or equalizing these differences.