Summary

  • HKIX publishes populated traffic graphs and a participant table. Our earlier article wrongly let unsuccessful page reads stand in for an assessment of what the exchange disclosed.
  • Aggregate traffic, route-server prefixes and listed connections answer different questions. None, alone, promises a buyer a particular route or amount of usable bandwidth.

Paying for a connection and getting the connection you expected are not quite the same thing. A network could choose a busy exchange, order access, and still need to settle how it will exchange traffic with the network carrying the content its customers want. The port may be real; the assumed arrangement with the other network may not be. That is a purchasing risk, not a failure we have observed at HKIX.

For the person signing off the expense, the distinction matters before the order, not after it. A large traffic figure can make a location look like an obvious choice while leaving the most consequential question unanswered: does the proposed connection do the job it is being bought for?

HKIX offers a good place to see the distinction. Its participant table contains network names, autonomous-system numbers and connection information. Its traffic page contains actual graphs, numeric legends and update times. The numbers exist. The harder question is what to do with them.

First, a correction is necessary. The earlier version of this article recorded five unsuccessful document reads on 17 October, then made too much of the apparent absence of figures. Our fresh inspection found the working statistics and participant pages. A failed read was a limit on our reporting, not evidence that HKIX withheld its measurements. It certainly did not establish a motive for doing so.

A large number with a small label

In the captured daily graph, the maximum incoming reading is 2.632 Tb/s. The label above it matters: five-minute average. The graph's printed update time is 20 October 2026, 07:21:17, without a stated timezone; we captured it at 23:24 UTC on 19 October. This is a dated operator reading, not a claim about traffic at the instant someone opens this article.

It is not an offer of 2.632 Tb/s to an individual customer, either. Traffic carried across an exchange and capacity available to a particular connection are different things. Longer-view graphs use other averaging intervals and also display separate five-minute maxima. A comparison that strips away those labels can turn unlike measurements into an impressive but misleading league table.

The participant list has a similar trap. A flag in a connection-speed column tells the reader something useful about a listed network. It is not a count of all its ports, a statement of spare capacity, or an agreement to peer with a new arrival. Nor do four menu headings make four populations that can safely be added together: one of the headings concerns sites.

The route server is not the whole exchange

The separate prefix statistics concern routes announced by HKIX's AS4635 route servers. Prefixes describe address ranges in routing information, not traffic volume. Those graphs do not claim to enumerate every bilateral route between participating networks.

The operator's FAQ describes both route-server arrangements and bilateral peering. A route server helps networks exchange routing information; it is not the box through which all their payload traffic must travel. For a buyer, that means a published prefix total cannot replace checking the arrangement with a particular counterpart.

APNIC's AS4635 record identifies HKIX-RS1 and links it to The Hong Kong Internet Exchange Limited. That supplies a specific registry connection which our earlier account had not verified. It does not establish the operator's entire resource footprint, its present membership tier or every route a customer can use. We have not obtained a current corporate-register extract, and the old article's membership and legal-form descriptions should not be treated as freshly verified facts.

Why meet locally at all?

HKIX dates its establishment to April 1995. Its own account describes a layer-two exchange operated by Hong Kong Internet eXchange Limited, with links to The Chinese University of Hong Kong's institutional structure. One purpose is straightforward: traffic that can be exchanged locally need not make an unnecessary overseas trip. HKIX also supports exchange involving international networks.

That purpose gives the statistics their interest. They describe activity at a place intended to bring networks together. They do not establish that every customer's path is shorter or cheaper. The result still depends on who is present, the arrangements between networks and the connections beyond the exchange.

For someone choosing a connection, the sensible reading starts with the destination and works back. Is the relevant network there? What arrangement is available? What does the proposed service actually include? Those questions connect the statistics to the expense being considered. The curve can show why a place deserves attention; it cannot commit another network to carrying your traffic on the terms you assumed.

Correcting the reference trail

The original homepage remains relevant. The old traffic URL and participant URL returned 404 in this check; use the working pages linked above. The original PeeringDB entry is NYIIX New York, not HKIX; the HKIX entry is here. We retain the old Wikipedia comparison reference for transparency, not as a newly verified ranking. None of these corrections establishes exactly why each 17 October read failed.