Note (added after the meeting): the tasks were not discussed as little work has been done on Kea during the past two weeks owing to preparation for IETF-83 and the conference itself. Accordingly the sprint has been lengthened to four weeks.

Kea Meeting




We have a performance requirement for 300 leases/second, rising to 1,000 leases/second at peak times. Currently we believe that we are restricted to between 5 to 50 leases/second (due to disk access). (Note: setting up a performance test suite is one of the tasks for the near future.) We hope to get a significant increase in performance with the delayed-ack feature in the next release. If we get a factor of 30 increase, that could be enough. If nothing else, we need to make sure that Kea is no worse than the current stuff.

DDNS in 4.2 is putting a bigger load on the system. Previously we do one write/lease, now we do up to three writes/lease. A separate module for DDNS is likely to improve things:

  • It will make use of multiple cores.
  • There may be no need for the entry to be in the DNS immediately the lease is allocated. DNS is inconsistent - a change made at a primary may take time to propagate to a secondary. In the meantime, queries to the secondary will not return the information added. (We need to establish the limits to the inconsistency - by how much and for how long can the DNS be out of step with the leases issued?)

For the future, we have prospect of more funding. We need to figure out what we need to do. One engineer full time?

Last modified 6 years ago Last modified on Apr 3, 2012, 2:50:54 PM