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

[kraileth]

Since FreeBSD 1) is usually slightly more secure than Linux 2) has the highest performance in NginX and 3) has the most stable ZFS implementation, we can conclude that FreeBSD is unquestionably a higher-up operating system for the cloud compared to Linux.

I'm not going to comment on 2) as other people already pointed out that this used to be the case but Linux caught up later. But the other two points?

1) Since "security" is one of the worst set phrases in the IT, question here is what you actually mean. First things first: It's really hard to compare FreeBSD to "Linux". To do so we'd have to compare kernels only. But comparing FreeBSD as a whole we'd have to check against some distro - and those tend not to perform the same when it comes to managing e.g. vulnerabilities. It's fair to say that in general in both camps there are people who take things very seriously and do a great job.

But despite liking FreeBSD a lot (or maybe rather because of it because otherwise I wouldn't care), I will not hold back that there are areas where it leaves a lot to be desired. To say the least. If anybody cares about this, I've written two blog posts about it earlier this year when the OPNsense team announced to switch to vanilla FreeBSD for their base. The first is about where FreeBSD is doing well and the second one is about its bad parts (so that's what the BSD haters might want to read to gear up a little):

https://eerielinux.wordpress.com/202...-and-security/
https://eerielinux.wordpress.com/202...ense-for-2022/

3) This is not true anymore. Since FreeBSD 13 (or 12.2 + ZFS from ports) both Linux and FreeBSD are using the same implementation: OpenZFS. And that's really a great thing for both. Linux used to lag behind quite a bit then caught up and eventually even grew new features of their own. While FreeBSD always stayed close to the illumos upstream (e.g. they implemented TRIM support in the lower layers instead of directly in ZFS to not break compatibility), the Linux port didn't really seem to care for ZFS as a whole. But then they invited FreeBSD to join them - and FreeBSD decided to accept the invitation. So ZFSonLinux was forked and modified to become ZFSonFreeBSD for a short while and the changes were then merged into the new upstream. When it was ready (i.e. the kernel modules for both Linux and FreeBSD could be built from a common repository!), the project was released as OpenZFS 2.0, making both ZFSonLinux and FreeBSD's old implementation based on that of illumos-gate obsolete.

What is true however is that the integration of ZFS into the system is much better on FreeBSD. It's working well on Linux, too, but it always feels alien, you usually need to manage it yourself (or live with pretty old versions) and it's being actively fought by some of the kernel taliban (most prominently GKH who hates it for licensing reasons). But while ZFS in FreeBSD is very good, you might want to take a look at illumos for the cherry on top. The way that BEs are built to be a native part of the OS still surpasses FreeBSD (which is no wonder since Solaris is where it originated from).

But let's move on to poke the penguin troll a little more, shall we?

The link you provided is useless, because the one that made the benchmark had no clue about configuring Linux for this workload. Even BSD developer debunked your 'benchmark', but you ignored this. The same applies to nginx.

I'm not even going to comment on your judgement on uselessness (FreeBSD is popular with developer communities of both Postgres and Nginx - and it shows unless you keep your eyes shut tightly). But since you trust this "BSD developer" so much, I'd like to know how much time you spent on your background check. Do you know who this guy is? I don't. Maybe he is a FreeBSD developer, maybe not. If he is, some of his posts are in violation of the FreeBSD project's values and that's a pretty serious thing as FreeBSD had adopted a crude CoC exactly in that year and things weren't pretty. I've heard more than one random person on the internet claim to be a project member who clearly weren't. There are a couple of rather fishy comments. Then again, he's been involved in a campaign of throwing dirt at OPNsense, took position for Matt Macy and such - maybe he's a Netgate employee and does work on FreeBSD. Still: What's the result of your background check and why do you think that your source is that reliable?

Not a chance. Linux eats it for breakfast as shown in the link I provided. Furthermore, here are some very old benchmarks when Linux was young, but it didn't stop it from making BSD's to look like a toys:

http://bulk.fefe.de/scalability/

You don't know much about Unix and its history, you? The original BSD has been a contender in the "Unix wars" and managed to beat commercial Unix in many regards time and time again (before Linux was even a thought in Torvald's brain). It set the standards in many fields - to this day, BTW. It's not a toy and never was. The BSD derivatives - and especially FreeBSD - quickly built up a fine reputation being used in cases that even commercial offerings could not handle. Linux was wearing diapers at that time. A while later, there was a time when Linux and FreeBSD struggled for superior performance. For a moment one was ahead and a couple of weeks later the other. In addition to true technical advances, there was also a lot of "let's pretend" in the game. Von Leitner's comment that you quote here is from that time and age.

Linux has had two main advantages: 1) It was a relatively fresh system in contrast to BSD that was weighted down by years and years of old ballast. 2) The USL vs. BSDi case lead to more and more companies going the "better safe than sorry" way and adapting Linux even though it was vastly inferior at that time.

When it came to the network stack, contrary to your belief, Linux was never a match for FreeBSD. It's not terribly hard to come up with scenarios that make one look better than the other. But you might want to take a look at why Netflix (and thus a company that at a point used to be responsible for 1/3 of the US internet traffic - all from FreeBSD boxes!) is using their OS of choice (though heavily tuned of course). Or why Facebook did quite a bit of work on the Linux side of things to make it perform comparably (they wanted to get rid of the FreeBSD machines they inherited when they acquired WhatsApp among other things).

FreeBSD doesn't come close to Linux when comes to security and networking performance.

Depending on what you mean by "security", there's actually some truth to it. Regarding networking performance - that's not generally the case.

Linux own the cloud and where's FreeBSD? Nowhere.

Err... Nope. You cannot have done a lot in the cloud if you never ran across FreeBSD. Of course it's available on AWS and the like. Microsoft even maintain their own downstream distribution with a few customizations for Azure.

FreeBSD doesn't scale above two sockets, so it's a toy OS.

When did you last try?

I don't care about ZFS.

Of course not! Ext4 is totally fine if you're not doing serious work.

Oh, legacy OSS crap again.

Good lord! OSS is neither legacy nor is it crap. What is crap is Linux' half-assed attempt at getting sound done that unfortunately also had the name "OSS". It's totally not the same thing. But I guess it's a bit much to ask a penguin to know that there's more out there than just Linux (and Linux is at times much closer to worst-in-class than the opposite)...

FreeBSD has no single advantage on desktop.

No idea where you draw your illusions from or why you seem to seriously believe that you could judge a system that you don't know much about (except for some misinformation obviously)? In fact it has enough advantages that I made the switch to FreeBSD after about a decade of only using Linux on all of my devices. I won't claim that there were absolutely no downsides to it, but the benefits clearly outweighed them.

Linux is better in every single case (including community).

Now that was a good one! While there are some very bright heads in the Linux community, the amount of trolls, ignoramuses and sheer arrogance is extremely high in comparison. Even though I had witnessed it time and time again, I was a bit surprised of how much hatred fanatical penguins spew at me the renegade, when I started exploring BSD, too. I can guarantee you that the same thing would not happen if I decided to leave *BSD e.g. for illumos.

Just keep in mind that we're there, too, for when you're ready to leave Linux kindergarten. BSD is for grown-ups.

Linux is Linux while FreeBSD is Linux wannabe: 'stealing' graphic stack, desktop environments, file systems (this one from slowlaris), it's even emulating Linux, because it can't exist without Linux. You have no clue and I don't care about trollish opinions.

The troll is complaining about "trolling" when confronted with the fact that not everybody is narrow-minded enough to buy a single "one size fits all" solution? Nice! Setting things straight for anybody who might read this: Linux began as a simple toy project (not "big and professional like GNU"). When it comes to filesystems, they quickly "stole" (to re-use your wording) that of Minix. After more of a decade of crying how unfair it was that companies only ever supported Windows and OSX and demanding more attention "for the others", they quickly forgot about the latter once they started receiving it and have now even turned into the new bully that's out hazing its former comrades and bragging about their status. Now isn't that proof of a lovely character?

FreeBSD is not trying to be like Linux. Why do the penguins assume that there was any desire for inferior design? Because they think that anyone else would be glad to succumb to "success"? No, sorry. Not everybody agrees to have their values for sale. And when it comes to desktops that's a great field to argue over as well. Why? Because it's the one top area where Linux fails to gain any relevance to this day. There's multiple reasons for this and most of them are not Linux' fault at all. Still it's entertaining to see a troll clutch for such a straw - and even ignoring that among the many popular *nix DEs out there there's no deliberately Linux-only DE!

[Volta]

According to the definition of 'performance' it is correct to say that start-up times and stop times are part of performance [...]

Ok, when your bhyve crashes thousand times a day its startup performance indeed matters.

Do you have any benchmarks showing at what point KVM has better memory efficiency than bhyve? And what is your explanation for the fact that VMWare is the most popular virtualization software, although it has lower performance than bhyve?

Read about mentioned KSM first. By asking for this you're proving again you have no clue what you're talking about. We're not talking about VMWare if you didn't notice.

His intent was to compare stock OS configurations. You can make a lot of changes to the configurations of both Linux and FreeBSD. Each distro makes its own choices. And Linux has clearly made the wrong choices here.

I didn't know Linux has to be configured for stupid people, but ok. Blame 'genius' who made a benchmark. NginX made 'wrong' choices when comes to defaults as well.

I know that PosgreSQL was mainly developed on FreeBSD in the beginning, and back then it was found to be more stable on FreeBSD because the Linux filesystems were very buggy.

Linux file systems were years ahead of POS that UFS was. An only good file system FreeBSD has is 'stolen' ZFS from Solaris.

Why is there such a huge amount of money invested in Linux if it can't even stand out in terms of performance, and failed to develop a stable file system for databases in all these years? Wouldn't all this money have been better invested in FreeBSD?

It's because Linux eats FreeBSD for a breakfast in term of performance and nobody gives a shit about FreeBSD except BSD license 'abusers'. ZFS isn't FreeBSD file system, so try again.

The comparison maker says he optimized both CentOS and FreeBSD well for NginX.

I don't care what he says, because he's clueless like you. Prove me.

The world's most popular gaming console (PlayStation) and the world's most popular gaming handheld (Nintendo Switch) both run on FreeBSD.

Keep dreaming. There are some BSD parts used, but that's all. They don't even come close to Linux market share - Android.

The truth is FreeBSD is a toy and doesn't scale above two sockets, have terrible file system (except ZFS which was 'stolen' from Solaris, because FreeBSD can't do file system on their own, it lacks in every area in comparison to Linux.

FreeBSD was less secure than Windows:

https://aboutthebsds.wordpress.com/2...d-is-insecure/

As you can see, all the memory addresses have not changed which shows that FreeBSD lacks Address Space Layout Randomization or ASLR, an important security feature that all other operating systems including Windows have.

FreeBSD jails as a security nightmare and copycat of Linux original jails:

https://aboutthebsds.wordpress.com/2...curity-danger/

The first implementation of Jails was actually done by Daniel Lezcano on the his own copy of GNU/Linux in 1996 (contrary to what the FreeBSD project claims). Lezcano wanted something that was more flexible and yet more secure then a standard GNU chroot and so he set about modifying GNU chroot to achieve that task. What he ended with was a feature that was more secure then chroot but much less fexible and powerful.

To this day [2013], FreeBSD jails still have this backdoor in them and in fact, was used by attackers during the April 9, 2010 security intursion into Apache.org. Apache has just recently initiated a programming to convert all of their FreeBSD servers to linux.

In 2000, Michael Calce (Mafiaboy) was able to jam up Yahoo’s servers which were at that time using the newly released FreeBSD 4.0 (first version to implement jails). Mafiaboy used only 10 to 20 hijacked computer to cause Yahoo’s FreeBSD powered servers out jam up, shutdown and even burn out due to the load already placed on the system by the jails it ran.

So if you are thinking of using features similiar to jails (OS level virtualisation), please don’t use FreeBSD jails as they do more harm then good. Using LXC, Linux-VServer or OpenVZ as they are far more secure and are many times faster then FreeBSD jails.

FreeBSD insecure by default:

https://www.youtube.com/watch?v=60vDy7Uzpp4

10 years of FreeBSD stupidity says Theo de Raadt:

"Basically, it is 10 years of FreeBSD stupidity. They don't know a thing about security. They even ignore relevant research in all fields, not just from us, but from everyone."

Theo de Raadt follows Linux security specialists:

It follows best practices in papers by (senior Linux developer) Ted Ts'o and many others in the crypto industry."

FreeBSD does security intentionally wrong unlike Linux and OpenBSD:

He said this code was added in 2003. "Like I say, it was using the same method of passing entropy into the bottom of the smoothing subsystem. We don't consider it better than mouse movements. Linux is also using an un-biasable method like us."

"Now we all discover that FreeBSD has been doing it wrong. It's not as if they operate in a closed source world, and couldn't have looked at what others did. They must have chosen a few years ago to do this wrong, intentionally.

*** ignoring rest of bsd delusional fanboism ***

[Volta]

Probably yes, but I don't think it's Kebabbert, don't know what happened to him since he has been gone quite a while now. AFAIK he was just super mad at Linux because he worked at OMX (a derivatives exchange here in Sweden) that when merged with Nasdaq 2008 threw out all Solaris machines and replaced them with Linux machines.

The best news I've heard in a long time. Thanks!

[Volta]

Many of the older benchmarks prove that 10 years ago Linux was mostly slower than FreeBSD in almost all domains. What I read recently is that Linux has generally not become faster than 10 years ago, in a phoronix article I read that. What do you think this means? You say it's sad that FreeBSD has to emulate Linux for certain things. But is it not sad that FreeBSD is often faster even though it emulates Linux, eg for gaming in wine this used to be the case.

Many of the older benchmarks prove Linux is much faster than FreeBSD in all domains. I recommend you to stop reading fairy tale books and start doing something more serious. Linux improved a lot while FreeBSD is still years behind in everything including performance.

[ermo]

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.

Fedora is also the recommended Desktop Linux Workstation OS inside Facebook. (source)

[Classical]

Startup times and performance are two different segments and bhyve is not competitive in term of performance. Even if KVM uses more memory for single VM it will use much less memory for many VM's. It's because Linux' KSM.

According to the definition of 'performance' it is correct to say that start-up times and stop times are part of performance:
https://dictionary.cambridge.org/dic...sh/performance (how well a computer, machine, etc. works)
Do you have any benchmarks showing at what point KVM has better memory efficiency than bhyve?
And what is your explanation for the fact that VMWare is the most popular virtualization software, although it has lower performance than bhyve?

The link you provided is useless, because the one that made the benchmark had no clue about configuring Linux for this workload. Even BSD developer debunked your 'benchmark', but you ignored this.

His intent was to compare stock OS configurations. You can make a lot of changes to the configurations of both Linux and FreeBSD. Each distro makes its own choices. And Linux has clearly made the wrong choices here.
I know that PosgreSQL was mainly developed on FreeBSD in the beginning, and back then it was found to be more stable on FreeBSD because the Linux filesystems were very buggy. And what has changed in this area? Btrfs is still buggy on Raid, and XFS is still buggy on CentOS as well, which I recently noticed. Apparently important parts of SQLite are faster on FreeBSD: https://openbenchmarking.org/embed.p...ha=c253c2f&p=2 Why is there such a huge amount of money invested in Linux if it can't even stand out in terms of performance, and failed to develop a stable file system for databases in all these years? Wouldn't all this money have been better invested in FreeBSD?

Your data is very deprecated and debunked by the benchmark I showed before. Furthermore, CentOS isn't the latest and greatest Linux distribution. When comes to NginX (which was not so optimized for Linux in the past) its performance can be boosted even up to nine times. It's thanks to thread pools. And when comes to default NginX (thus your meaningless benchmark) configuration.

The comparison maker says he optimized both CentOS and FreeBSD well for NginX. So your explanation that multi-threaded was not enabled does not apply. How come multi-threading isn't automatically enabled for NginX when Linux is such a wonderful system as you claim? After Ubuntu, CentOS is the most widely used in the cloud. After all the huge sums invested in Linux, how can CentOS not compete with FreeBSD which is much less subsidized ?

Linux own the cloud and where's FreeBSD? Nowhere.

But Linux doesn't perform well in the best web server, furthermore lacks stable file systems, and Docker may be the biggest security blunder man made. Isn't the 'success' of Linux in the cloud mainly driven by the incompetence of most cloud engineers, and therefore not really something to be proud of? In a few months or years, NanoVMs will also be faster and more secure than Linux in the cloud. So will the Linux cloud engineers still use Linux? (although FreeBSD and NanoVMs are faster, more stable and more secure) That would be very sad.

How come Linux, although it gets a lot more subsidies than FreeBSD, has less security than FreeBSD? https://www.techradar.com/news/usb-s...ially-on-linux

Oh, legacy OSS crap again. Wake me up when FreeBSD catches up in graphic stack and DE support. FreeBSD has no single advantage on desktop.

The world's most popular gaming console (PlayStation) and the world's most popular gaming handheld (Nintendo Switch) both run on FreeBSD. There's no doubt that FreeBSD's graphics capabilities are extremely good. If Linux can't even play audio the way it should, then it might not be a great fit for gamers and music lovers, which probably ninety percent of people are.

BTW there are tools to make pythonic code do the heavy lifting without leaving performance on the table.
Maybe the next time take a look at
Cython
Numba
Nuitka

Although Numba increased the performance of the Python version of the estimate_pi function by two orders of magnitude (and about a factor of 5 over the NumPy vectorized version), the Julia version was still faster, outperforming the Python+Numba version by about a factor of 3 for this application. Numba is useless for speeding up operations even with a numeric library like Pandas because it doesn’t know anything about how Pandas works. Numba is essentially a JIT compiler for a tiny subset of numeric code. Julia’s JIT works on everything, if you do it correctly.

LMFAO!!! You should get in contact with VALVe, since they still have time to correct their mistake and switch SteamOS over to a FreeBSD kernel!

It might have been beneficial if Steam had chosen FreeBSD over Linux, both in terms of stability and performance.

FreeBSD: A Faster Platform For Linux Gaming Than Linux? https://www.phoronix.com/scan.php?pa...inux_games_bsd

Both Nintendo Switch and PlayStation are based on FreeBSD, and they are both probably the two most reliable and stable gaming platforms to ever exist.

I have some reservations regarding the benchmarks on this page. If you look at the JetStream and Octane benchmarks, you will see that Google and Apple have adjusted the benchmarks in such a way that their software comes out better than the competition. How do we know that this is not the same story with the Phoronix test suite? And how do we know that the results on Phoronix are reliable at all?

I think this is a telling anecdote: https://unix.stackexchange.com/quest...-than-on-linux

So many downvotes stimulated me to install FreeBSD (TrueOS) on my 6-core desktop computer to test it myself. (NOTE: I do not recommend trying to install TrueOS alongside other operating systems, because this installation wiped one of my hard drives, even though I tried to install it on a USB drive... Not a user-friendly experience.)

As a result, after running some tests from the Phoronix test suite on both Ubuntu and FreeBSD, I couldn't see the “slow applications on FreeBSD” effect. Quite the contrary, some applications ran significantly (10–25%) faster on FreeBSD.

[Linuxxx]

Amdgpu support? Can't complain. Even Dragonfly supports my card (Vega 64). FreeBSD amdgpu is somewhere past Linux kernel v5.xx equivalent and Vega's been supported since 4.16. Playable, mind ya, tested with some 3D racing game in ports and Flightgear flight sim.

Playable, because BSD is able to run an unnamed 3D racing game (probably TORCS) & Flightgear?

Here, let me open 'ya' [] eyes, since 'ya' have been obviously living under a (BSD) rock:

According to this quote, the Steam Deck with its 800p screen should be able to at least come close to the pixel performance of the Xbox Series X in 4K mode, which is the fastest console currently availabe:

Twitter user Iashman has shared an extensive comparison of the Steam Deck's specifications to rival consoles such as the PlayStation 5 and Nintendo Switch. The summary comes from Peter "Durante" Thoman, a popular developer and model, who posted the breakdown on Steam. Durante reveals that the Steam Deck's graphics processing unit is as fast as a PlayStation 4, but the newer architecture and smaller screen give it the same raw GPU performance as an Xbox Series X. The central processing unit is half that of an Xbox Series X or PlayStation 5, but the console's 16 gigabytes of random access memory is equal to that of a PS5 and four times that of the Nintendo Switch.

https://screenrant.com/steam-deck-performance-xbox-ps5-switch/

Now, take a look at what is already possible with this kind of hardware power for real-time 3D graphics in 4K:

And just to be clear, I'm not saying that the Steam Deck will be able to run games with this much visual fidelity @ 800p, but just to give an example what kind of stress upcoming AAA+ games will put on the Steam Deck running SteamOS!

So, ask yourself this:
Would a toy OS like FreeBSD be up to the task like the above?

If you answer yes, then please try to run properly complex & advanced software like RPCS3 on FreeBSD and report back how it went!

[moilami]

This is when GPL killed BSD license.

[aht0]

Amdgpu support? Can't complain. Even Dragonfly supports my card (Vega 64). FreeBSD amdgpu is somewhere past Linux kernel v5.xx equivalent and Vega's been supported since 4.16. Playable, mind ya, tested with some 3D racing game in ports and Flightgear flight sim.

Yo, Volta &Co.. don't you feel like ultimate losers? "Oh there's a benchmark showing my favourite operating system in maybe bad light. Must attack them now.."

90% of people would suggest straightjacket and padded room.. getting angry and aggressive over abstract software, Jesus.

[Linuxxx]

But is it not sad that FreeBSD is often faster even though it emulates Linux, eg for gaming in wine this used to be the case.

LMFAO!!!

You should get in contact with VALVe, since they still have time to correct their mistake and switch SteamOS over to a FreeBSD kernel!

Also, how is the AMDGPU support on the various BSDs these days?
Or are you guys still hailing nVidia's blob, because that's the only viable option you actually have?