Summary

  • RIPE NCC says its experimental RIS Parquet update files now generate hourly, after a once-a-day publication rhythm. Its 2025 design set a 14-day window for the derived files.
  • The raw MRT archive remains the deeper record. A team that wants to rerun the convenient SQL analysis months later has to keep the relevant Parquet partitions, schema and query while they are available.

At 02:09 UTC on 28 October, RIPE NCC's public directory showed a Parquet file for the preceding hour of routing updates. At 03:06, the next one appeared. Yesterday's directory held all 24 hourly partitions. A researcher asking what changed in BGP this morning can now reach for a file named for the hour instead of waiting for a daily bundle.

That is a useful piece of engineering. It also starts a timer.

RIPE NCC collects routing observations through its Routing Information Service, or RIS. The underlying MRT archive contains raw update segments at five-minute intervals and full routing-table views every eight hours, according to the registry's documentation. MRT is durable and rich, but extracting a narrow question from many collector files takes parsing and compute. At RIPE 91 in October 2025, RIPE NCC engineer Ties de Kock showed a different doorway: indexed Parquet files that ordinary SQL and data-science tools could query.

His demonstration described analyses that took more than ten minutes against a traditional workflow running in seconds on the prototype. That was a presentation result, not a promise that every question would run at that speed.

The prototype separated a daily routing-table view from update files partitioned by hour. Those file names described the data's time slices. At launch, the presentation also said automated publication was still to come. The distinction matters because a dataset can contain 24 hour-labelled pieces and still arrive as one late batch. At RIPE 93 on 27 October, RIPE NCC's Robert Kisteleki reported the operational change: the update files are now generated every hour rather than once per day. He said enough researchers were using the experimental files to continue the effort. The slides gave no user count.

The live RIPE directory supplies a narrower check of the delivery claim: the 27 October files carry successive publication times during the following hour, and the first three files for 28 October appeared at 01:06, 02:09 and 03:06 UTC.

Now look at the other clock. De Kock's original design gave the Parquet RIB and update files a 14-day short-term archive. On 28 October, the October update directory began at 14 October, while the September directory had no files. That snapshot fits the published design. It is enough to tell a research desk that the convenient copy needs prompt handling; it is not a service guarantee about every file or an audit of RIPE NCC's deletion process.

The consequence lands on an ordinary Tuesday. An operator investigates a brief route leak, or a researcher tests whether a prefix changed upstreams during a regional disruption. The hourly Parquet file gets an answer into view quickly. A paper, customer report or incident review may be challenged months later. By then, the same short-lived file may be absent from the prototype directory. The raw MRT material is still there, but replaying the analysis means reconstructing the input and query path. Same underlying observations, different labor bill.

The sensible capture is more than a screenshot of a chart. Keep the exact hour partitions, their hashes, the collector set, the prototype schema version, the query and the moment the files were fetched. RIPE's directory path includes v0, a useful reminder that this is an evolving representation. In 2025 the project promised at least 18 months of production, a go/no-go after 12 months and three months of coexistence for an old schema if the format changed. Kisteleki's 2026 update says the experiment continues. It does not turn an experimental file into a permanent record by applause at a conference.

RIPE NCC has improved the first half of the research bargain: an hour of route changes can be queried while the question is still warm. The second half belongs to the people who rely on the result. The institution should state the file's age, schema and retention plainly; the analyst should preserve the exact inputs behind a consequential claim. A fast answer without its replayable evidence becomes a story about the tool. A fast answer with its files attached becomes research.

Sources: RIPE NCC's RIPE 93 Tools Update and official slides; RIPE 91 presentation and original slides; RIS 27 October and 28 October file indexes; MRT documentation.