IDN TLD Root Server Performance / Tolerance | Draft
A draft IANA process for inserting and managing IDN TLD A-labels in the root is currently under public review. That process contains an Emergency Revocation Procedure which will be invoked if certain fault conditions take place. This paper describes a draft tolerance measure that invokes the emergency procedure.
This paper identifies the tolerance measure that invokes emergency removal of A-labels from the DNS root zone. This ensures that if any critical stability issues are introduced by the introduction of A-labels these are identified and appropriate action following the IANA process for removing A- labels from the DNS root zone is conducted.
The root-server operators will be asked volunteer participation by monitoring traffic during the evaluations periods. The traffic monitoring is anticipated to discover critical stability issues and if such are above the below identified tolerance measure then the corresponding A-labels are removed immediately. Root-server operators are asked to sign up for participation prior to the launch of the evaluations. While the level of participation can be decided individually by the respective root-server operators, the anticipated monitoring includes:
- The number of queries for records related to each of the .test zones (those beginning with xn--).
- Measured as average queries per second and queries for the previous 24 hours.
- The average load relating to the .test zones measured as a percentage of overall traffic.
- Relating to both (queries) q/sec and (bandwidth) bytes/sec
The tolerance for introduction of errors or damage to the provisioning of the root zone caused by introduction of A-labels (inserted as NS-records) is zero.
Any string that is inserted into the root zone file, addition, or deletion of a record for an existing zone should have no negative effect on the availability of the zone as a whole or the resolution of other delegations within that zone.
Acceptable levels may be indicated by the ability to resolve the pre-existing child zones.
As one cannot realistically monitor resolution from the entire Internet it may be fruitful to set up a series of testing stations. These stations could run an automated process to query each of the delegations within the zone from multiple locations. These processes could run on a timed schedule and/or directly after a change to the zone.
Locations would be chosen by ICANN technical staff based on the availability of infrastructure but would include no less than three geographically and topologically diverse locations.
Variance in query times would be measured between each query series with an unanswered query being considered as 100% variance or critical failure.
A various in response time will be calculated as (((new query time (X) divided by previous query time (Y) )minus 1) multiplied by 100 = percentage change (N)
((X/Y) -1) x 100 = N
It may be more suitable to generate an average variance across both the multiple locations and time. Where by X is the average over the various locations and (Y) is that averaged over time.
A reasonable scale of the tolerance measure is currently being decided in consultation with the DNS Root Server System Advisory Committee (RSSAC).