Benchmarks: FreeBSD 13 vs. NetBSD 9.2 vs. OpenBSD 7 vs. DragonFlyBSD 6 vs. Linux

[Danielsan]

Nice, these comparisons help the BSD Folks to focus on areas which is needed more optimization.

We should also says that, for instance, FreeBSD doesn't really optimize ZFS tasks, or tasks that rely upon the filesystem, hence by default the results might be ugly.
Linux distributions have more the habits to tuning default, and I am not joking, to look cool exactly here on Phoronix, hence I take the results without being very impressed.

Anyway it is fine that distros baked by gazillion of dollars have great performance and it is nice that BSDs backed by hundred of dollars have nice performance too!

[Volta]

FreeBSD's Bhyve Overview: Why it's better than other hypervisors. At least for our use-case. https://www.youtube.com/watch?v=uV61mVYsFM8

The VMs start and stop much faster in Bhyve. KVM and Xen are much slower in this area. Bhyve is also seriously superior in the memory footprint.

So, you changed rhetoric, because your original point got burned? KVM is superior in performance so your 'start up' benchmark (valid or not) is POS like the rest of your comment. KVM is superior in memory footprint, but I bet you have no idea why it's like that.

Is that right?

Yes, furthermore BSD developer explains the results:

As a FreeBSD developer, I will actually tell you that Linux could be selected to graph massive wins by cherry picking hardware with very high core count and several NUMA domains. But even then, by selecting kernel features (which could be innocuously hidden in a version number/vendor patch set) you can cherry pick large swings. That said, FreeBSD+ZFS+PGSQL (https://www.slideshare.net/SeanChittenden/postgresql-zfs-bes...) is a joy to administer, and is unlikely to be the weak link in a production setup if you stick to a two socket system. There is a lot of work going on in HEAD that is relevant to this workload in the memory management subsystem, including NUMA support. And some TCP accept locking changes that'd be relevant for TCP connection turnover.

There are currently +- 500 million NginX servers active. And even more Apache server. Have you noticed how much faster FreeBSD is in NginX compared to CentOS?

And most of NginX is running on Linux just like Apache, so what's your point? No, I haven't, because Linux performance is superior to FreeBSD:

https://matteocroce.medium.com/linux...g-cbadcdb15ddd

The first thing that struck me, is that FreeBSD packet rate was substantially the same with one or 8 CPU. I investigated a bit, and I’ve found it to be a known issue: bridging under FreeBSD is known to be slow because the if_bridge driver is pratically monothread due to excessive locking, as written in the FreeBSD network optimization guide.

Both OS performs well, being able to forward more than 1 million of pps per core, which lets you achieve the 10 gbit line rate with 1500 byte frames.
FreeBSD scales relatively well with core numbers (except in bridge mode which is kinda monothread), but Linux does a near perfect job using all the power of a multicore system. The same applies to firewalling, where we can see that a large firewall ruleset can disrupt the performance of both kernels, unless using tricks like fastpath and HW offloading.

***************

And do you know how many data centers exist on the planet? I think it is definitely more important that FreeBSD is 2x faster than Linux in PostgreSQL (the best SQL database).

Even netflix storage runs on Linux and Linux PostgreSQL performance is superior to FreeBSD, so everything is as it should. Linux for serious business while BSD for few fanboys. Linux is more than 2x faster than FreeBSD in PostgreSQL:

https://www.phoronix.com/scan.php?pa...ebsd-bsd&num=5

When looking at the PostgreSQL database server across the varying core/thread configurations, CentOS Stream was the front-runner at 16 cores through 128 threads followed by Ubuntu 20.04 while FreeBSD 12.1 atop ZFS saw much less scaling.

That's why I say that many of Michael's benchmarks have little or no value.

That's why your post have no value.

Kebbabert, is that you? Your slowlaris got wiped out, so you found some new toy?

[Volta]

Fedora has no LTS releases and is thus irrelevant for serious production servers. Fedora is a playground for Red Hat Linux in order to test new technologies. I'm writing this as a Fedora desktop user on my private PC.

It's not a playground, but very solid desktop distribution.

[neoe]

Sorry for the seemingly ignorant question..

Is there a reason for not including Fedora?

Is Ubuntu just more recognized as a server OS?

(I would think more people could come up with Fedora as a server than Clear..)

Thanks in advance.

Intel sponsored. For CPU and Clear Linux and these benchmarks.

[F.Ultra]

Kebbabert, is that you? Your slowlaris got wiped out, so you found some new toy?

So far no links to PR brochures as "technical guides" and random out of context Torvalds quotes as "evidence", so not so sure.

[F.Ultra]

I wonder what FreeBSD does to make it create threads so much faster than anything else here, are their spectre mitigations applied in a different manner perhaps? Just speculating since those mitigations took a hard hit for context switches.

Also strange how it changed behaviour for zstd compression when the level was decreased.

[KesZerda]

Inasmuch as the benchmarks are showing performance as of the time of testing, you could argue that the benchmarks give a realistic picture of the real world performance you can expect with the respective zstd packages available ootb?

I do agree that an apples-to-apples zstd 1.5.x comparison would've been interesting though.

FreeBSD 13 has both 1.4 and 1.5 available, though. While the base install has 1.4.8 for compatibility purposes, 1.5 is available through packages and ports, and most packages are built against it instead, so for day-to-day performance, 1.5 would be a more realistic measure as to what people will actually be using.

[Steffo]

It's not a playground, but very solid desktop distribution.

When I say "playground", I don't mean this in a bad way. Fedora has established many technologies, but because it has not LTS releases, it is not really suited for production.

[kylew77]

I have time to sleep now and had a few thoughts. Maybe the FS benchmarks on OpenBSD were lower because it doesn't enable soft updates by default? Also, as a positive the OS ships its own system version of perl that is modified for security, that may be why the perl benches didn't run. Also, the timed LLVM benchmarks I think OpenBSD finishes last because of its lack of SMT.

[jacob]

There are currently +- 500 million NginX servers active. And even more Apache server. Have you noticed how much faster FreeBSD is in NginX compared to CentOS?

No I haven't. Any actual benchmark data to support it?

And do you know how many data centers exist on the planet? I think it is definitely more important that FreeBSD is 2x faster than Linux in PostgreSQL (the best SQL database).

It absolutely isn't and of the many data centers on the planet, approximately zero run on FreeBSD. I "wonder" why.