ITHI Metric M5, Recursive Resolver Integrity

-

The M5 metrics measure the "integrity" with which resolvers process queries. The metric is computed by sending specially crafted queries at DNS resolvers from a large number of clients, including queries to domains under the experimenters' control. This allows us to analyze how recursive resolvers manage key part of the service, and how well responses sent to stub resolvers track data updates by the authoritative servers. There are two ways of measuring these metrics: either as a fraction of the user population; or, as a fraction of the total number of resolvers. We provide both evaluations based on estimation of the population size behind each of the resolvers that we track. We also provide an estimate of the fraction of the global population behind the 10000 largest resolvers.

There are seven components in the M5 metric:

• The metric M5.1 measures the % of visible recursive resolvers that refresh their local cache in accordance with the time specified in the SOA TTL, and the effect on users. For each population, we compute the fraction of users or resolvers that re-fetch erlier than the stated TTL, and the fraction that re-fetch according to the TTL. These two numbers do not add to 100%, because there is also a fraction of resolvers for which our measurements are inconclusive.
• The metric M5.2 measure the % of visible recursive resolvers that refresh their local cache automatically (without a triggering user query), as well as the estimated query load % due to automated cache refresh and the % of users using resolvers that auto-refresh their cache.
• The Metric M5.3 measures the % of visible recursive resolvers that set the DO (DNSSEC OK) bit in their queries, as well as the fraction of users using resolvers that set the DO bit in queries.
• The Metric M5.4 measures the % of visible recursive resolvers that perform DNSSEC validation of responses, as well as the fraction of users using resolvers that perform DNSSEC validation of responses.
• The metric M5.5 estimates the fraction of the clients that is served by the 10,000 largest visible recursive resolvers.
• The metrics M5.6 and 5.7 measure the concentration of the resolver services by tracking the number of resolvers required to serve 50% or 90% of the number of clients.

M5.1 to M5.5 only consider the individual resolvers identified by an individual IPv4 or IPv6 address, but for M5.6 we may group several addresses together if they belong to the same "resolver service". We also observe that clients may send their first queries to one resolver, but if it does not respond they may try other resolvers. Some clients send these first requests in parallel use multiple resolvers. For that reason, be use four different methods for computing concentration, leading to four different submetrics. The details of grouping and computation are explained in the "about M5" page.

The metric M5.7 presents a simplified view of the M5.6 measurements. When a client request triggers multiple query arrivals at the test server, we only retain the first one, the "initial" query.

The following table provides the value observed for the metrics in the current month, as well as the average value over the 3 previous months, and the "historical" minimum and maximum observed since the beginning of the measurements.