The primary objective of this benchmarking is to compare various deployments of authoritative name-server implementations in different environments. All the scripts, data, and configurations are open-source and stored at dns-benchmarking project.
Disclaimer: As many things can affect the measurement (hardware, operating system, configuration, human mistakes, etc.), the results provided below are just informative.
- December 10, 2019: Epic DNS Benchmarking
Response Rate Benchmark
The test is based on two physical servers directly connected via 10GbE. One server replays proper PCAP file at various speeds to the second server, which is running the testing name-server software. Outgoing responses are monitored on the network stack to eliminated various network issues.
Target hardware is Intel Xeon E5-2630V3 (HT enabled), Intel X710 10GbE (FW 6.80), 64 GiB RAM.
Notes & Remarks
- Explicit CFLAGS="-O2 -g -DNDEBUG"
- Enabled SO_REUSEPORT (SO_REUSEPORT_LB on FreeBSD) and minimal responses if supported
- The number of threads/processes is the same as the number of CPU cores (16)
- The number of network card IO queues is 16 on Linux and 8 on FreeBSD