Skip to content

libknot: optimize knot_dname_lf for many short labels

Libor Peltan requested to merge libknot_lf_optimi into master

The goal is to improve answering performance when the query name has many short labels (e.g. reverse IPv6 queries).

Test gcc 10, 8 cores, 15M QPS, example.com.:

  • 1 x 7 label qname: RPS change 75% -> 75%
  • 7 x 19 label qname: RPS change 52% -> 55%
  • 12 x 4 label qname: RPS change 42% -> 49%
  • 25 x 2 label qname: RPS change 25% -> 27%
  • 50 x 1 label qname: RPS change 10% -> 19%
Edited by Daniel Salzman

Merge request reports