Summary

  • WIDE’s archive made backbone traces more reusable by joining privacy processing and repeatable publication with measurement, rather than treating release as a separate clerical step.
  • Akira Kato co-authored foundational repository and traffic-profiler papers with colleagues; they document a team design, not his sole control of today’s MAWI archive.
  • A trace is evidence about a particular link, interval, capture setup and anonymization method. It is not a complete view of a network or a record with unrestricted access.

A trace describes a window, not the whole network

The current MAWI FAQ says its packet capture can miss traffic. A router’s port-mirroring feature copies packets to a server and network card, but that chain does not guarantee every packet reaches the file. The FAQ also says timestamps are usually accurate to less than a millisecond, yet can be worse during congestion. The archive chooses equipment and operating practices it can sustain over the highest possible precision.

Those details are not footnotes to the research. They define the observation. A trace taken at a transit link can show what crossed that link during a stated period, subject to the capture process. It cannot show every path elsewhere. WIDE’s own FAQ notes that universities connected to both WIDE and SINET may send traffic through either network, while asymmetric routing can put only one direction of a flow at the monitored point.

This is the right place to begin a history of Akira Kato’s repository work. The data is valuable because a researcher can inspect and reuse it. The same openness requires a clear account of what was observed, what was transformed, and what remains outside the record.

Privacy became part of the publication path

In October 1999, the WIDE Project and its MAWI Working Group described a plan for making traffic traces easier to share. The guideline recognized two distinct privacy problems: protocol payload could contain user information, and IP addresses could help identify a user or organization. It proposed removing sensitive payload and scrambling addresses before public distribution. Public use was limited to research, and privacy trespass was prohibited.

The guideline also states the hard trade-off. Prefix-preserving anonymization keeps address structure that researchers need to study routing, but a well-known server can offer a clue for reverse engineering. More consistency across traces can improve analysis while making linkage easier. Anonymization therefore reduces risk; it cannot certify that identification is impossible.

The repository paper that Kato co-authored with Kenjiro Cho and Koushirou Mitsuya in 2000 treated privacy and automation as coupled design problems. Raw traces were handled by a limited group. A scripted process transferred data from sampling points, applied tcpdpriv, generated summaries and a web page, updated the archive index and prepared the trace for distribution. That sequence turned a privacy policy into a repeatable release procedure.

The authors’ concern was also practical. A long-running archive cannot depend on people remembering to perform every step by hand. Automating collection, summary and visualization made routine publication more sustainable. It did not remove the need to inspect a faulty input or a changed capture environment; it made those conditions part of an operational system that could be documented and improved.

A useful summary spends some resolution

Kato also co-authored Aguri, an adaptive traffic profiler, with Kenjiro Cho and Ryo Kaizaki. Aguri groups traffic along address prefixes and produces summaries that can help operators see patterns without retaining every detail in every view. The paper describes a bounded-memory design in which near-real-time aggregation trades resolution for a fixed resource budget.

That trade is productive if the reader understands it. A summary can make a trend visible, but it cannot stand in for packet-level evidence when an investigation depends on details the aggregation removed. Aguri is not a privacy guarantee, and an aggregate does not reveal every event that produced it. It is an analytical layer whose useful abstraction must remain distinct from the underlying trace.

This same discipline applies to the archive itself. A chart, a compressed trace and a packet capture are different representations of an event. Each preserves some distinctions and discards others. Research is stronger when those choices travel with the result rather than disappearing behind a clean graph.

The archive records its own changing vantage

The MAWI Working Group maintains the current archive. Its present index records why comparisons across time need care. It identifies a missing interval in late 2022 after hardware failed during a point-of-presence migration, duplicated packets in part of 2015 following a VLAN misconfiguration, and a change to samplepoint-F when BBIX traffic was added in 2023 after transit traffic fell.

Those notes make the archive more credible, not less. A measurement series can be long-lived without being stationary. A new interconnection changes which traffic reaches the sensor. A hardware replacement can change loss or timing behavior. A gap is missing evidence; duplicated packets are not a sudden doubling of demand. The researchers who compare years need to know where the observation process moved.

Anonymization has a similar time boundary. The current FAQ says address mappings are consistent within a daily trace, or within the corresponding Day in the Life of the Internet dataset, rather than across every day and collection. It also says the method changed from tcpdpriv to Crypto-PAn in August 2015. A cross-period analysis that assumes one stable identifier would mistake a privacy design choice for an enduring identity.

The current FAQ further says that WIDE backbone traces mix ordinary traffic with research experiments. That matters for anomaly analysis: a visible probe or test may reflect an experiment, not a sudden change in user behavior. The archive identifies some such traffic and marks known limitations, but a reader still has to establish what belongs to the question being asked.

Public access has a separate raw-data boundary

Open sanitized traces make it easier for independent researchers to check an analysis or ask a new question. The 1999 guideline argues that a shared dataset can support work without requiring each researcher to negotiate access to raw traffic. But it does not make raw data open by default. Specific requests for addresses were to be handled case by case, with care for user privacy. The current FAQ likewise limits unanonymized traces to research and asks applicants to justify the request; students need an adviser’s endorsement.

The access boundary is part of the evidence contract. A public trace has undergone a transformation designed to lower privacy risk and preserve some analytic structure. A special raw-data request has a different purpose, risk and decision process. Treating both as “open” would blur the difference between a reusable research artifact and information that still needs a human decision.

Keio University’s profile connects Kato’s work to computer networks, Internet infrastructure, measurement, WIDE Project and M-Root operations. The 1999 guideline and the papers tell a more precise story than a biography alone can: his role sat within a collaborative effort to make measurement usable over time without pretending privacy, completeness or precision had been solved once and for all.

What the work makes possible—and what it cannot

The repository’s lasting contribution is a repeatable public measurement surface with explicit processing and access boundaries. It allows researchers to examine traffic patterns, compare methods and build on earlier observations. It also provides operators with a way to discuss behavior using data that other people can inspect.

But no archive label can turn one link into the whole Internet. No anonymization method removes every inference risk. No timestamp or capture device is exact by declaration. No traffic summary retains distinctions it intentionally aggregates away. These are not reasons to abandon shared measurement; they are reasons to make its provenance and uncertainty travel alongside it.

The lesson fits a broader principle from Heng Lu’s notes, used here only as an analytical lens: rely on the operating mechanism and its evidence, not on a title or a diagram. For a traffic trace, that means asking where the sensor sat, how data moved from it, which fields changed, what the tool discarded, and who approved access to the remainder. Coordination becomes more trustworthy when those choices remain visible to the next researcher.

Sources