Summary

  • FICIX describes itself as Finland's largest internet exchange, a registered non-profit association serving 67 customers across five sites, with EVPN/VXLAN switching and 10G, 100G and 400G port options; this is the exchange's own account, reported here rather than independently measured.
  • The capacity side of any utilisation figure exists only as port ceilings that participating networks file themselves, while the load side has no published per-port figure and the exchange's statistics page could not be read in this review — so a member-level utilisation ratio cannot be constructed from the public record, and any fabric-level ratio measures how full the shared switch is, not how much headroom any one network has.

What the exchange says about itself

The first document is the exchange's own front page, retrieved at 21:38:24 UTC. In the passage the research provider retained, FICIX describes itself as "the biggest Internet exchange point (IXP) in Finland", a registered association and "a non-profit organisation currently serving 67 customers", which operates five sites and whose switches "use EVPN/VXLAN technology and support 10/100/400G port speeds" (FICIX front page).

Four things are worth separating in that sentence, because they are routinely collapsed into one another. A customer count is a commercial relationship. A site count is premises. A port-speed class is an interface configuration. None of the three is a measurement of traffic. The exchange's self-description as the largest in the country is exactly that — a self-description, reported here and not checked against any third-party measurement in this review.

The same retrieved material enumerates the five sites as FICIX1 across Espoo and a Telia Helsinki data centre, FICIX2 across Helsinki and an Equinix Helsinki facility, and FICIX3 in Oulu. That enumeration comes from the provider's extracted claim candidates rather than from a topology document, so it is recorded here as the provider's reading of the page, not as a site list verified against an operator diagram.

The word the page uses matters: customers, not members. Whether those 67 customers are the same population as the networks that file ports at the fabric is not established by anything retrieved. A Finnish exchange serving 67 customers could be 67 organisations with one port each, or a smaller number of organisations holding many ports, or a larger number of networks among which some are reachable through a reseller relationship. The record does not say, and this article does not guess.

The capacity side is a declaration, not a measurement

The second document is not about FICIX at all, and that is why it is useful. The public peering database's own documentation, retrieved at 21:44:23 UTC, describes a data model in which a network's connection at an exchange is recorded through an object carrying a speed field that represents port capacity — and that carries no traffic volume, real-time or historical, per port (PeeringDB documentation).

Read plainly, that means the capacity figure a planner can obtain for a fabric like this one is a number a network typed in about its own interface. It is a configured ceiling. A 100G port that carries two gigabits of offered load is still a 100G entry in any sum. Nothing in the record distinguishes a port that is running hot from a port that was provisioned and never filled, and the exchange does not audit the figure on the network's behalf.

This is not a criticism of the database, which is doing what it says it does. It is a statement about what the number is. Port speeds are declarations of intent and configuration, filed by the party that benefits from being listed, and they answer the question "what could this connection carry" rather than "what is it carrying".

The load side is absent at the resolution that matters

The third document is the one the commission actually turned on: the exchange's statistics page, at a distinct address, retrieved at 21:42:42 UTC. In this review its contents could not be obtained. The page exists; nothing beyond its existence was retained. No chart, no value, no unit, no averaging definition (FICIX statistics page).

That produces a narrow and specific boundary, and it is important to state it precisely rather than in the direction that flatters a negative finding. This review cannot say that FICIX publishes no traffic series, because the page that would contain it was not read. It also cannot say what such a series measures — whether the plotted value is a peak, a percentile or a mean, over what interval, across which of the five sites, in decimal or binary units — because the definitional notes, if any, were not obtained either.

What can be said is what the retrieved material contains. A search pass over the exchange's public pages produced an explicit statement that no per-member or per-port traffic figures were detailed in the material provided (FICIX front page). That is a statement about the material retrieved, not about the operator's publishing practice.

The practical consequence is that the load side exists, if at all, only as one aggregate number for a shared fabric — and even that could not be read here.

Why the division fails even when both numbers exist

Suppose the statistics page had opened, and suppose it displayed a headline figure for the fabric across its five sites. It is worth being exact about what dividing that figure by the sum of filed port speeds would produce, because the answer is routinely misread.

The aggregate is counted once at the switch: a bit crossing the fabric once contributes once to the total. Capacity is counted once per port: every network's ceiling is added in full. The quotient therefore describes how efficiently the fabric's own participants collectively fill the capacity they have collectively declared. It is a property of the fabric. It is not a property of any member (this is the author's construct, offered as an analytical argument rather than as a sourced fact about FICIX).

Two consequences follow. The first is that a membership dominated by a few very large ports makes that denominator insensitive to the smaller networks that also depend on the fabric, so a comfortable-looking aggregate and a congested individual port can coexist without contradiction. The second is that the ratio cannot see queueing inside a port. A network whose own interface is filling toward its ceiling contributes a tiny fraction to a large aggregate and disappears into it entirely.

This is the same category error in a different costume as the opening paragraph's: treating premises, commercial relationships and interface configuration as interchangeable measures of an exchange's capacity. They describe different objects, on different clocks, published by different parties.

What would settle it

Four documents would each move a specific layer from unverified to located.

A per-member or per-port traffic report published by the operator would convert the ratio from a fabric-efficiency proxy into an actual member-level measurement, and would be the single most consequential document here (FICIX front page). The operator's statistics page read as text, including whatever definitional notes accompany the chart, would settle the definition layer — and would also establish whether a series exists at all (FICIX statistics page). A published site or fabric topology document would settle the question the five-site statement raises, since one fabric across several sites and several fabrics behind one customer count are different objects for anyone reasoning about resilience. And a schema page or API record showing a traffic-volume field per exchange port would defeat this article's central claim outright (PeeringDB documentation).

None of the four was obtained. The dated result, scoped to three documents and a six-minute retrieval window on 18 October 2026, is negative: member-level utilisation at this exchange is not publicly measurable from the records located in this review.

The background for the entry is held in the BTW directory record for FICIX.

A bounded consequence

A network sizing a port purchase against an exchange's headline traffic number, or a reader judging an exchange's headroom from a ratio of its own construction, is measuring the fabric's filling rather than anyone's queue. The shortfall, when it arrives, arrives inside a single port whose aggregate neighbour looked comfortable — and the measurement that would have warned of it is not in the public record. That is not a scandal. It is a gap between two publication layers that happen not to meet, and it costs nothing to name.