DragonflyBSD 3.0 Performance Benchmarks

Near the end of February marked the release of DragonflyBSD 3.0 with multi-core speed boosts and other improvements, but how does this correlate to performance improvements in our usual open-source benchmarks?

DragonflyBSD 3.0 by default finally switches away from its uni-processor kernel to now the SMP multi-processor kernel. The multi-core DragonflyBSD 3.0 kernel has improved thanks to work on VM system concurrency and locking along with other areas of its kernel. Last year there were benchmarks coming out of the DragonflyBSD camp of huge performance boosts on a 48-core workstationm, while recently I've been doing some tests on hardware with fewer cores.

An extensive set of multi-system DragonflyBSD 3.0 benchmarks is still being conducted, but here's a preview of some early results using a dual AMD Opteron 2384 quad-core workstation.

DragonflyBSD 3.0.1 64-bit was compared to DragonflyBSD 2.10.1 with the SMP kernels. Both DragonflyBSD releases were using a GCC 4.4 series compiler.

With the compiler tests using multiple jobs (16 jobs; twice the number of available CPU cores) on both Dragonfly releases, the new 3.0.1 release was much faster than the previous 2.10.1 release.

While the compilation tests were faster, other heavily multi-threaded tests like C-Ray and 7-Zip compression weren't faster with DragonflyBSD 3.0.1.

Other tests had little in the way of changes with the new DragonflyBSD.

One area that had apparently regressed, at least for this dual-socket AMD workstation with a SATA HDD, was the disk performance. Both DragonflyBSD installations were using the default HAMMER file-system. Under DragonflyBSD 3.0.1, the HAMMER file-system performance was slower. At least though there's the HAMMER2 file-system moving forward.

With the more extensive DragonflyBSD 3.0 benchmarks will also be some numbers against the recently-released FreeBSD 9.0.

Those wanting to try out DragonflyBSD 3.0 can fetch it from DragonflyBSD.org. More DragonflyBSD benchmarks are available from OpenBenchmarking.org.