Summary

  • DTEL-IX, the provider-independent exchange in Kyiv, puts four numbers on one page — 184 participants, more than 272 connected networks, 110,000 prefixes and a peak above 3.23 Tbit/s — and attaches no reference date, no measurement window and no counting rule to any of them.
  • The same page says that 7 international and 22 national resellers deliver remote participants' traffic to Kyiv. That sentence is the only published explanation for a gap of at least 88 networks between "participants" and "connected networks", and it means the traffic that makes the exchange's peak impressive arrives partly on other people's transport contracts.
  • The practical consequence: a port in Kyiv buys a metropolitan fabric with a documented peak and an undocumented geography. The part of the reach that is national or international is a separate commercial relationship with one of twenty-nine intermediaries, and nothing on the page tells you which one you are buying.

Two numbers, four units, no rule

On 2 November 2026 the operator's homepage carried, in Ukrainian, this sentence: "провайдеро-незалежна точка обміну трафіком з більш ніж 272 підключеними мережами; пікове навантаження перевищує 3.23 Тбіт/с; 7 міжнародних та 22 національних реселерів доставляють трафік віддалених учасників до Києва". A second excerpt of the same page, retrieved ninety-three seconds later, adds "184 учасників. 110 000 префіксів... Публічний піринг з роут-сервером". Both excerpts come from the operator's homepage, and both are excerpts rather than a full page capture.

Four numbers, four units, one page. Participants count organisations. Connected networks count something else, and the page does not say what. Prefixes count routes. Tbit/s is a rate. None of the four carries a date, a window or a definition.

The arithmetic is trivial and worth doing anyway. More than 272 connected networks against 184 participants is a difference of at least 88, or roughly 1.48 networks for every participant. Since 272 is a floor, 88 is a floor too. Divide the published prefix count by the participant count and you get about 598 prefixes per participant; divide it by the network count and you get about 404. Divide the peak by the participants and you get about 17.6 Gbit/s each — a number with no operational meaning, because an exchange is not an average and the page publishes no distribution, no per-port utilisation and no window.

The obvious explanation is the boring one: an exchange counts participants as organisations holding a port, and counts connected networks as everything reachable through those ports, including downstream customers who will never appear on a member list. If that is the rule, 272 is the honest number and 184 is the membership number, and nobody has done anything wrong. But the page never states the rule, so a reader holding both figures cannot collapse them, and every comparison with another exchange — or with DTEL-IX a year ago — inherits the ambiguity.

What the reseller sentence actually does

The clause about resellers is doing far more work than the headline numbers, and it deserves to be read slowly. "7 міжнародних та 22 національних реселерів доставляють трафік віддалених учасників до Києва" — seven international and twenty-two national resellers deliver remote participants' traffic to Kyiv. Twenty-nine intermediaries, in two categories, named by count and not by name.

Read one way, that is a distribution channel. A reseller buys a port on the Kyiv fabric and resells it to a network that cannot justify its own build to Ukraine; the remote network becomes a participant, the traffic terminates on the exchange's switches, and the exchange's reach is genuinely its own. Read the other way, it is a transport dependency.

A reseller carries the remote network's traffic from Warsaw, Frankfurt or Tbilisi on its own wavelengths and hands it over at a cross-connect in one of the two Kyiv data centres; the exchange's switches then see traffic that only exists in Kyiv because a third party decided to sell the path.

The page does not distinguish between those two models, and the difference is not academic. In the first model the 3.23 Tbit/s peak is a measure of the fabric's gravitational pull: networks come because other networks are there. In the second, the peak measures how much traffic twenty-nine companies chose to deliver, and the exchange's growth is a function of their commercial appetite rather than its own attractiveness. A buyer choosing a port cannot tell from the page which of those markets they are joining.

There is a third possibility, and it is the most likely one: both are true at once, in different proportions, and the operator has no reason to publish the split. That is exactly the kind of silence this column exists to point at. The reseller count is the only sentence on the page that explains where the ≥88 extra networks come from, and it explains them by reference to a set of companies the page declines to name.

3.23 Tbit/s and the window that is not there

A peak-traffic figure is the single most useful number an exchange can publish and the least useful without a window. "Пікове навантаження перевищує 3.23 Тбіт/с" could describe yesterday, last December, or the best hour in the exchange's history. Each of those readings implies a completely different story about growth, seasonality and headroom, and the page supports all three equally.

What can be said is narrower and still worth saying. The figure is above the capacity of any single port on any plausible port menu, which means it is aggregated across the fabric rather than quoted from one link. It is large enough that the exchange is a serious national aggregation point rather than a hobby project. And it is stated as an excess — "перевищує", exceeds — rather than a measurement, which is the honest way to publish a number you are not standing behind to three decimal places.

The route server is the part of the record that makes the rest coherent. "Публічний піринг з роут-сервером" — public peering with a route server — is the mechanism by which 184 organisations can produce a prefix count in the six figures without every one of them building bilateral sessions with every other. A route server lets a small network peer once and receive everything. It is also the piece of infrastructure that makes a reseller model technically easy: a remote network handed into the fabric by a reseller can take a route-server session and look, from inside Kyiv, exactly like a local member.

Own premises, and what "own" means

The claim that precedes the numbers is the one most likely to be over-read: "єдина точка обміну трафіком в країні, що володіє власними приміщеннями в двох найбільших дата-центрах України: Ньютелко Україна і БІ МОБАІЛ". The only traffic exchange point in the country that owns its own premises in two of Ukraine's largest data centres, Newtelco Ukraine and BE MOBILE.

Two claims sit inside that sentence, and neither can be checked from the page. The first is uniqueness — "the only" — which is a comparative statement about every other exchange in Ukraine and is asserted without a list. The second is ownership. In Ukrainian commercial usage "власні приміщення" can mean freehold premises or dedicated space held under an agreement, and inside a third-party data centre the second meaning is far more common than the first. A cage with a lock and a cross-connect panel is a real operational asset and a much smaller legal one than a building.

The more consequential silence is about fabric. Two named data centres is a footprint claim; one switching fabric spanning them is a redundancy claim, and the page makes the first without making the second. If the two sites sit behind a single pair of switches, a failure takes both. If they are separate fabrics joined by leased capacity, the second site protects against a building problem and not against a transport problem. A buyer sizing a dual-site cross-connect is making a resilience decision on the strength of a sentence that does not address resilience at all.

The registry record points at Tbilisi and Frankfurt

BTW's directory record for the company Only for Strategic Circle is built from the RIPE NCC public member directory rather than from the operator's marketing. It records Digital Telecom Internet Exchange LLC as a private company registered in Ukraine, a RIPE NCC member, with service-area context for Ukraine, Georgia and Germany. It also records, honestly, what it does not have: no mapped ASN, no prefixes, no relationships, no leadership contacts, and an explicit note that ASN and IP resources have not been established from RDAP, WHOIS, IRR, BGP or RIR delegation data.

The Georgian and German entries are the interesting part precisely because they are unexplained. A company that owns premises in two Kyiv data centres and describes itself as the country's exchange, yet appears in a registry directory with service-area context in Georgia and Germany, is either an operator with sales, support or point-of-presence arrangements abroad, or an artefact of how a member record was parsed into service areas. This review could not determine which, and the directory record does not pretend to.

Three attempts, two failure classes, zero documents

What this review could not get matters enough to state precisely, because the precise version is less flattering to the article and more useful to the reader. Three attempts to reach an independent routing or registry record returned no document at all, in two distinct failure classes. The first was capacity unavailable: the research route had nothing to give after a bounded wait. The second was an authorisation refusal at the research endpoint while a document address was supplied. The third was the same refusal, followed by a second failure reporting that the request required document context.

Three attempts, two classes, no evidence. That is a retrieval failure and nothing else. The absence of a routing record in this article is not evidence that DTEL-IX lacks an autonomous system, route servers or announced prefixes; the operator's own page mentions a route server, so the opposite is more likely. Anyone reading a later, thinner version of this piece should know that the gap was produced by a broken route, not by a discovered emptiness.

The other provenance limit is harder to fix. The two successful retrievals are the same URL, so the record here is one publisher counted twice, not two independent reports. Everything about Arista Networks hardware, operation since 2012, a Kyiv Telecentre site, an L7 switching node and a 10, 40, 100 and 400 Gbit/s port menu appears only in the research layer's own prose summary of that page, not in the retained Ukrainian text. Those items are therefore not used here as established fact. They may well be true. They are not evidence this article can carry.

Two outcomes, named in advance

If a public interconnection or routing record shows DTEL-IX operating its own autonomous system with route servers and remote participants' sessions terminating on the same Kyiv fabric, then the gap between 184 participants and more than 272 connected networks is bookkeeping inside one fabric, the exchange's reach is its own, and the reseller sentence describes a sales channel.

If instead the record shows remote participants reaching Kyiv only through the twenty-nine resellers' transport, then the peak describes a metropolitan aggregation point whose wider reach is rented, and the reseller count is a structural dependency rather than a distribution choice.

There is a smaller falsifier underneath. If the two named data centres turn out to sit behind one published switching fabric, "own premises in two of the largest data centres" becomes a redundancy claim. If they are separate fabrics joined by leased capacity, it stays a footprint claim. And if the operator ever publishes a counting rule showing that connected networks include resold downstream networks while participants counts only direct members, the 88-network gap closes and the inference in this article is confirmed as arithmetic rather than conjecture.

What it costs to guess wrong

For the three audiences this piece is written for, the cost of the missing sentence is different in each case and concrete in all of them. A network operator considering a port in Kyiv is deciding whether to buy national reach or a metro fabric plus a reseller's transport contract, and the two are priced differently. A foreign network weighing remote peering is deciding whether its path into Ukraine depends on one of twenty-nine companies whose names it does not know.

An investor or a regulator reading the 3.23 Tbit/s figure is deciding whether Ukrainian interconnection is growing on its own or being sold into existence by intermediaries who could stop.

None of those questions is answered by the four numbers. All of them are answerable, by one paragraph on a homepage. That is the gap, and it is 88 networks wide.