Benchmarks Of FreeBSD 8.1 RC2

[monraaf]

I find freebsd defaults to take less tweaking to get the secuirty I want, I love the ports system and the documentation. If there is one thing I laugh at the most when people talk about linux is when they say it has good documentation. The Freebsd handbook http://www.freebsd.org/doc/en/books/handbook/ is the reason I took up freebsd over a linux distro in the first place.

FreeBSD does have good documentation, I'll give you that. And that was also one of the reasons I picked it up in the first place. The problem is that the documentation just can't make up for the poor hardware support, to some extent software support and the overall lack of developers working on FreeBSD.

The FreeBSD project is quite aware that it can't compete with Linux on these things. So what are they doing? They are marketing FreeBSD as an OS that offers "true freedom", where users are not bound by the restrictions of the GPL. Of course they don't tell their users that without all the GPL code in ports and to some extent in the base system FreeBSD would be pretty useless. Hey but one day FreeBSD will be truly free!

In the meantime users -- who just want things get done without installing Linux compat layer and half of Fedora libs, without waiting for someone to port software to FreeBSD or look for 'inspiration' at GPL'd Linux kernel code before slapping a BSD license on it -- are switching to Linux, and FreeBSD is becoming a refuge for GPL haters and other extremists.

The only regret I have switching from FreeBSD to Linux is that I didn't do that much earlier.

[soupbowl]

freebsd fanbois always claim that freebsd is faster.
and there are always benchmarks disproving them
and then they always claim some fault, like 'this is the wrong version' or 'that was last version. X+1 is so much faster'.
they also always get a knot in their panties because of softupdates.

This might be true, but linux guys are the exact same.
http://www.phoronix.com/forums/showthread.php?t=23555:
If you read through there you will find exactly what you are saying freebsd guys do, but its the linux fanboys doing it.

Mr.kraftman can be seen http://www.phoronix.com/forums/showpost.php?p=125882&postcount=44 crying that one reason the benchmarks are not fair is because the nvidia drivers are not the same versions(lol).

Others are crying that slackware,gentoo etc were not used over ubuntu, even more people crying "what GCC flags were used!" "omg the file system used is not the best".

As you can see freebsd and linux fanboys, as do all fanboys are the same and just as annoying. As for the commenter above, you are correct about freebsds hardware support, it is not even close to linux. I do have modern hardware and I have a perfectly working desktop without any trouble. I have only had a problem with some hardware on some newer labtops.

But again, linux hardware supports is impressive.

[kraftman]

This might be true, but linux guys are the exact same.
http://www.phoronix.com/forums/showthread.php?t=23555:
If you read through there you will find exactly what you are saying freebsd guys do, but its the linux fanboys doing it.

Not exactly, because even freebsd devs are doing lame benchmarks and write fanboy like articles while Linux devs don't.

Mr.kraftman can be seen http://www.phoronix.com/forums/showpost.php?p=125882&postcount=44 crying that one reason the benchmarks are not fair is because the nvidia drivers are not the same versions(lol).

Don't you consider driver version matters? Like the GCC version matters? If not then why the bsd fanboys are crying here? Does Windows provide a binary driver for nvidia cards for default? If not, then why it has been installed?

Others are crying that slackware,gentoo etc were not used over ubuntu, even more people crying "what GCC flags were used!" "omg the file system used is not the best".

Damn, did you turned soft updates on? They speed up our stone age file system a lot!

As you can see freebsd and linux fanboys, as do all fanboys are the same and just as annoying.

The difference is, some bsd devs are one of the worst fanboys.

[kraftman]

Also, as you can notice the main concern when comes to benchmarking Linux vs Windows in 3D was the misleading title and FUDish intro. While the Phoronix tests defaults and doesn't draw strange conclusions then I've got no problem with such benchmarks, but it seems bsd fanboys have got.

[trasz]

Wow. Phoronix guys yet again proved that newer versions of GCC produce better code that GCC 4.2.1 (cannot use newer one due to unacceptable license restrictions in GPLv3, move to Clang is planned), and that "threaded I/O benchmark" that does not actually do any disk I/O (seriously, go and look yourself how does this "benchmark" work) shows that doing nothing is faster on Linux.

It would be great if Phoronix guys got a clue and created suite of real, application benchmarks, for example benchmarking DNS, PostgreSQL or HTTPD performance, instead of synthetic ones. Would be even better, if they knew what they were benchmarking (most of the tests here depend on compiler, not the kernel). Obviously, none of these will happen.

[trasz]

Although we've had a no real religious debate here as a result of the article, I'd like to suggest something. **NOTE** that this will require work from a FreeBSD person.

When 8.1 is released, we do the following...

The FreeBSD person creates a blog page with a tuning guide to make FreeBSD shine in the benchmarks. The rules are as follows...

1) The tuning guide is explicit enough for a reasonably technical person (but not FreeBSD guru) to follow and create a customized FreeBSD for the purposes stated.
2) The person hosting the page provides Michael support in preparing the system for benchmarking following the directions.
3) Any new tests that the FreeBSD person wants to have executed are worked out/contributed/made available prior to the benchmarking run.

[..]

Matthew

Geez, it's incredible how much you don't get it. You are simply not measuring _operating system_ performance - you are measuring either GCC performance (most of the tests) or nothing at all ("threaded I/O tester" that doesn't do I/O). For start, you would need to get some real tests, for example PostgreSQL benchmarks. Until that, what you do might look interesting to some newbies, but nothing more than that.

[mtippett]

... You are simply not measuring _operating system_ performance - you are measuring either GCC performance (most of the tests) or nothing at all ("threaded I/O tester" that doesn't do I/O). For start, you would need to get some real tests, for example PostgreSQL benchmarks. Until that, what you do might look interesting to some newbies, but nothing more than that.

Let's go back to the post. My suggestion is that if you feel strongly about it, join the fray and get your hands dirty. You are making reference to a set of tests and benchmarks that should be run. Feel free to go the next step and explicitly identifying some. I am sure there will be minimal concerns in rolling those in.

Realistically, gcc does have some impact, but there are lots of other aspects that are relevant beyond gcc. If xBSD will never pick up a gcc version, then comparing newer versions of gcc on other *nix's is 100% fair. xBSD is making a conscious decision to exclude that from it's initial ecosystem. The different platforms have their own ecosystems which is what is measured.

If you feel that you can capture conveniently the work needed to remove the compiler from the picture, feel free to document it here or put it on a blog or a webpage somewhere. I'm just looking for people to go the next step and constructively guide where they think things are wrong vs just complain that people are clueless.

[kraftman]

Geez, it's incredible how much you don't get it. You are simply not measuring _operating system_ performance - you are measuring either GCC performance (most of the tests) or nothing at all ("threaded I/O tester" that doesn't do I/O). For start, you would need to get some real tests, for example PostgreSQL benchmarks. Until that, what you do might look interesting to some newbies, but nothing more than that.

Yeah, fanboy of fanboys have said. Tell this freebsd guys, because it seems they don't know how to make a proper benchmarks. Instead, they were making propaganda ones.

[kraftman]

Wow. Phoronix guys yet again proved that newer versions of GCC produce better code that GCC 4.2.1 (cannot use newer one due to unacceptable license restrictions in GPLv3, move to Clang is planned), and that "threaded I/O benchmark" that does not actually do any disk I/O (seriously, go and look yourself how does this "benchmark" work) shows that doing nothing is faster on Linux.

It shows out of the box Ubuntu kicks your lovely freebsd ass. :>

It would be great if Phoronix guys got a clue and created suite of real, application benchmarks, for example benchmarking DNS, PostgreSQL or HTTPD performance, instead of synthetic ones. Would be even better, if they knew what they were benchmarking (most of the tests here depend on compiler, not the kernel). Obviously, none of these will happen.

You saw a DNS benchmark many times and guess what was a winner? Quite old Linux kernel. The same about scalability benchmarks etc. Tell Kris to not be such jerk and to make some real tests.