A Red Hat intern investigated whether OVN and OVS hashmap performance could be improved by replacing the existing intrusive chained hmap with Swiss table designs. The work covered three approaches: wrapping a custom C Swiss table library (dshmap) under the existing hmap API, integrating dshmap directly into hmap with dual chained/Swiss modes, and targeted replacement of specific high-traffic maps (ovsdb-idl and lflow-mgr). All global replacement attempts regressed performance by up to 2.4x. Targeted Swiss table usage showed modest improvements (up to 11% faster for ovsdb-idl) but not enough to justify adding a new hash table library to OVS. The conclusion is that OVN/OVS workloads are mutation-heavy rather than lookup-heavy, which favors the existing intrusive chained design over Swiss tables.
