Zone parsers, like other parsers, are mostly sequential. To improve zone loading performance NLnet Labs researched ways to tune the process for much higher throughput. We analyze why current parsers are relatively slow and present ideas to leverage data parallelism (Single Instruction Multiple Data, or SIMD) and adjacent technologies to achieve a performance of (eventually, hopefully) GB/s.
Although the focus is on parsing, vectorization has a much wider area of application.
We hope to familiarize (DNS) developers with vectorization concepts and show performance can be improved by taking a slightly different programming approach.
(Please note that this talk replaces one entitled "External DNS Operator in Kubernetes" that was due to have been given by Servesha who sadly could not make it)