Phoronix article – Dragonfly BSD (2018-05-21) – Performance analysis, tools and experiments

Phoronix posted an article comparing performance for newest release of Dragonfly BSD. The comparison made is between the old release (5.2) and the new release under development (5.3).

Overall, performance looks improved from new to older release. Unclear how this compares against other operating systems.

Below is a table of overall observations as well as my comments on the benchmarks.

Benchmark	Phoronix observations	My observations	Analysis
sqlite	New release is 1.3x faster	Single-threaded program. Database using a file with file locking.	Analysis
compilebench	New release is slightly faster	File system performance with only 1/7th of a core scheduled.	Analysis
rodinia	performance is the same	OpenMP, 100% CPU with varying amounts of backend stalls.	Analysis
polybench-c	performance is the same	Memory bound linear algebra operations.	Analysis
fftw	New release is faster	Single threaded, Short vectors are high IPC. Long vectors are limited by memory.	Analysis
go-benchmark	http workload is faster, others are the same	GO language; http is ~20% system time, others less.	Analysis
java-scimark	performance is the same	Java wrappers around backend bound compute. Single-threaded.	Analysis
java-gradle-perf	performance is slightly faster	Build system, many small processes. On_CPU of 31%	Analysis
build-php	performance is slightly faster	Build system, On_CPU of 82% with frontend stalls.	Analysis
pgbench	performance is slightly faster	Database with frontend stalls. On_CPU of 60%.	Analysis
pybench	performance is slightly better	Single-threaded python. High IPC	Analysis
phpbench	performance is slightly faster	Single-threaded php with high IPC	Analysis
osbench	individual components are mixed	Unit tests for OS operations, e.g. threads, processes, files.	Analysis

Performance analysis, tools and experiments