Summary

  • The AMS-IX route-server page prints thirteen community values, and each of the seven that controls anything decides which peers receive a prefix — not whether traffic to that prefix is carried.
  • DE-CIX and SwissIX publish their own blackhole community, their own next-hop method and the prefix lengths they accept. The instruction exists in this trade and is missing from the AMS-IX pages I read.

The premise I brought, and what the pages did to it

I expected to find a remote-triggered blackhole documented on ams-ix.net: a community value, a next-hop convention, a prefix-length limit, and a sentence naming who may issue the instruction. That expectation is wrong. Over the eight first-party pages read on 21 November 2026 — route servers, config guide, allowed traffic, anti-DDoS, the supplementary page, resources, topology and get-started — there is no blackhole community, no drop next-hop, no drop-specific prefix length or count, no statement of who may authorise a drop, and no statement that a drop is logged.

The finding is a bounded negative over a named document set on a named date, not a claim about what the exchange supports.

Thirteen values, and not one of them drops anything

The AMS-IX route-server page is unusually explicit. Standard communities give four choices: 0:peer-as to withhold a prefix from one peer, 6777:peer-as to send it to one, 0:6777 to withhold it from every peer, 6777:6777 to send it to all. Large communities repeat the same three functions as 6777:0:peer-as, 6777:1:peer-as and 6777:0:0. Five tagging communities — 6777:65011, 6777:65012, 6777:65021, 6777:65022, 6777:65023 — and 6777:1900:0 for identification complete the set. Extended communities are refused outright. Thirteen values in total: seven that steer an announcement, five that label one, one that names it.

The limits around them are printed just as tightly. "We allow a maximum of 100 BGP communities per type to be attached in each prefix," the page says, "If a type of community exceeds the 100 occurrences, we remove them all." Next-hop is checked, not rewritten: "we do not allow prefixes to be sent where the BGP next-hop attribute is pointing to another router's IP address and not self." AS-PATH stops at 32 hops. Sanitisation blocks RFC 1918 space, bogons, Martians and the default route; Amsterdam additionally drops ROA status INVALID and anything not covered by an AS/AS-SET export. Per-AS prefix ceilings are dynamic, with an advisory maximum of 400K IPv4 and 110K IPv6. The config guide keeps the peering LAN prefix 80.249.208.0/21 inside AS6777 and forbids redistributing it. Route servers rs1 and rs2 run BIRD2.

That is a complete vocabulary for where a prefix may be announced, written by an operator that plainly enjoys being precise. Its own table of contents runs introduction, why use the route servers, max-prefix advisory, participation, use of the route servers, deployment guidelines, filtering, traffic engineering, prefix limits. No blackhole section. No remote-triggered black hole section. Announcement control is documented to the digit; delivery control is not documented at all.

Where a reader looking for DDoS help actually lands

The anti-DDoS page is the natural stop for a member under attack, and it sells an ecosystem rather than a mechanism: interconnect with providers that mitigate, early detection, scalable protection, smart filtering. No community, no next-hop, no prefix limit, no authorisation rule, no logging commitment. The supplementary page enumerates link aggregation, a quarantine VLAN for new ports, IPFIX and sFlow statistics, IPv6 numbering, ARP sponge and AS1200 peering — real features, none of them a drop. Resources lists an annual report, brochures, an SLA description, articles of association and Euro-IX material; no technical note on mitigation. Topology describes an MPLS/VPLS core and photonic cross-connects. Get started asks for an ASN and ownership documentation. Allowed traffic forbids exporting peering LAN space, directed broadcast and IGP traffic, and names BGP hijacking as abuse — a different act from a requested drop.

Across all eight pages the word "blackhole" appears exactly once, and it belongs to the member: inside a sample router fragment in the configuration guide, configure iproute add blackhole default. That is a customer discarding traffic on its own equipment. It is not something the fabric does on request.

The instruction exists in this trade

Two other exchanges publish precisely what I could not find in Amsterdam. DE-CIX's blackholing guide gives members two methods: mark the announcement with community 65535:666, which it calls the recommended way, or point the next-hop at a pre-defined blackhole address; it says up to /32 and /128 should be accepted, and that its route servers add NO_EXPORT automatically. SwissIX's guide documents the same two methods, publishes its own next-hop addresses, and tells members not to set NO_EXPORT or NO_ADVERTISE because the route servers add them. Those are DE-CIX's and SwissIX's policies, quoted as such and attributed to nobody else. Their value here is comparative: a member of either fabric can read the instruction, the accepted prefix length and the propagation rule before an attack, and a member of AMS-IX cannot read any of the three.

What the standard asks an operator to publish

RFC 7999, from October 2016, registers the well-known BLACKHOLE community as 0xFFFF029A — 65535:666 in decimal — and notes that operators use different triggers, including pre-defined blackhole next-hop addresses and specific communities. It recommends the most specific prefix available, /32 or /128. Then it imposes two conditions that matter more than the encoding: a blackholed announcement must be covered by an equal or shorter prefix the neighbouring network is authorised to advertise, and the receiving party must have agreed to honour BLACKHOLE on that session. Authorisation and consent are not decorations on the mechanism; the standard makes them part of it. An exchange implementing blackholing therefore has two sentences to publish, and these are the two sentences the reviewed AMS-IX pages do not contain.

The operator knows exactly how to issue one — for itself

An AMS-IX engineer's INEX members-meeting deck from 20 March 2025 shows the operator defending its own AS1200 by steering traffic toward NBIP/NaWas scrubbing with more-specific announcements, automated with FastNetmon and BIRD2 across the peering LAN and triggered by internal community 1200:511. That is a drop instruction, executed by the same people who run the route servers, on the operator's own routes. It is not a member-facing facility and 1200:511 is not a published AMS-IX blackhole community; treating it as one would be a fabrication. But it removes the last innocent explanation. The capability is not abstract here. It is on the same fabric, in the same hands, aimed inward.

The empty slot in someone else's database

The PeeringDB record for AS6777, last updated 26 October 2026, carries a field labelled "RTBH Community" — and no value visible in the retrieved record. Its notes send readers to the AMS-IX route-server page as the authoritative list of recognised communities, which is the page where no blackhole community appears. A labelled slot is evidence about how a database is designed. It is not a published value, and it settles nothing about the fabric.

What a member can know, and what they must ask

The exchange's own looking glass renders routes in a browser script and returned no readable route text in my checks, so nothing from it is quoted here. What remains is clean and checkable. A member of AMS-IX can read, before announcing anything, which peers will receive a prefix, how many communities may ride along, whose next-hop is acceptable, and what will be filtered. A member cannot read whether traffic to a prefix may be deliberately discarded, who inside the exchange may order it, whether the order must come from the prefix holder, or where the record of it would live. Announcement control is a published policy.

Delivery control is a private one, which is a statement about documentation and nothing more — and for anyone deciding to trust a fabric with revenue traffic, the second is the question that matters.