I don’t think there’s a good reason. BSD has comparatively poor hardware support, and software availability, while generally ok, is worse than Linux. BSD is simpler to modify because it’s all one stack and it has a license that benefits selfish users. That’s about it. Linux mopped the floor with BSD and it’s probably in large part due to a license that prohibits selfish users. Everybody needs an operating system and the GPL benefits everyone who uses Linux, including those who compete with each other. I wonder how many BSD modifications and drivers never see the light of day because developers aren’t required to share.

Peace of mind and much less maintenance work.

In April this year we had to move a FreeBSD server to another colo: Uptime 3172 days, just rebooted, did not bother with an update and ever since it has been serving some high traffic sites. All relevant security patches are applied but in all this time there was not a single vulnerability we had to fix that required a reboot.

FreeBSD doesn't suffer from the bazaar complex that linux does, in that for each given system there is generally only one way to do things, and the documentation is both succint and comprehensive. One of the most frustrating things about Linux to me is that every flavor requires its own understanding and it doesn't seem like anyone can agree on standards (gnome vs kde, systemd vs init.d, rpm vs deb vs snap etc etc)

It does have a Linux compatibility layer that sort of works, but not enough to run everything I tried to run on it.

Can't speak to FreeBSD, but I've run OpenBSD on firewalls for like 30 years, mostly because it is actually good at this, and the syntax, while changing, did so more slowly, and was more sensical to use than all of the linux ipchains/iptables/nftables menagerie (which, I will admit has improved - nftables isn't half bad).

It's also good from an ecosystem perspective to have a few parallel implementations of the same thing, as it avoids a bugs/flaws that could affect all of a monoculture.

Also, I'd rather have something made by people who are more paranoid than me, and are driving forward unix implementations - for example, the work done on 64 bit time_t.

It just works better for my specific use case. If you have a use case where FreeBSD works better, go for it.

Some BSDs are known for having more performant network stacks, and more cohesive software libraries. Each BSD seems to stake out a niche: OpenBSD is security, NetBSD runs everywhere, and (IIRC) FreeBSD started free when others cost money.

Were known for high performance TCP/IP stack around 10-15 years ago. Linux caught up (and overtook) FreeBSD long ago. The rest were simply never there.

Allan Jude is involved in a couple of podcasts and his advice is always the same: you pick FreeBSD because that's what you know. That's pretty much it.

Do you have any performance metrics for this assertion? My understanding was that Netflix and others have contributed very performant code.

I haven't seen an apples to apples network performance test in a while. Maybe some shops pick their OS based on network performance, but most pick their OS for other reasons and then bend it until the network performance meets their needs. Most OSes have at least good enough network performance; I wouldn't run a public tcp server on MacOS because they don't have syncookies, and I wouldn't run a large multicore server on OpenBSD because afaik, they're missing cpu pinning and I don't know if their scheduler is biased towards keeping processes on the same core --- everything else should be fine.

Netflix has contributed some general network performance increases, but the real big increases are only there if you are serving files from disk, with TLS, and you have a NIC that can accelerate the bulk crypto.

I like FreeBSD and choose it when I can, but I've never had occasion to benchmark it. I prefer the stability of user experience and the expectation that old documentation still applies that FreeBSD has and Linux doesn't. When I've dug into kernel source, I feel like FreeBSD source is better organized and easier to understand, but that might be familiarity bias.

Ultimately it becomes just as silly as any microbenchmarks. Most people don’t care that they can do 100G line rate. But the superior QoS implementations available for Linux make a real performance difference for many people. For example, as a typical gateway, Openwrt will perform better than pfsense simply because of cake. I am not Netflix, I do not have their problems and very few do. Nevermind their competitors doing fine with Linux.

I’ve messed with it as as a desktop OS and always found it not worth the setup when fedora/ubuntu just work on most systems I’ve used them on. FreeBSD is an outstanding server though, I run several services on cheap VPS out in the cloud and it’s rock solid and security issues see to come up far less often which is very important to me for anything outward facing to the wild and wooly internet