OSNews

I trust the graphs. I trust BSD to return to outstading performance levels. I trust FreeBSD developers and community.

Hey, I like BSD, what I was saying is not BSD specific, just a universal truth.

Even someone with the best intentions can unintentionally create a biased benchmark.

Sounds much like a profession of faith. I'm eager to see what FreeBSD will be like in action, but I'm not that religious about it.

I do not trust your premise either. At least in the document they have some references! You DONT!

The graphs you see were discussed heavily in the several mailing lists which made the NetBSD guys and the Linux guys improve things. You must have been missing a lot lately? Esp. regarding the Linux scheduler and FreeBSD's ULE 2.0.

http://jeffr-tech.livejournal.com/5705.html (follow the links in the comments)

http://mail-index.netbsd.org/tech-kern/2007/10/03/0003.html
(read the entire thread)

and
http://mail-index.netbsd.org/tech-kern/2007/10/05/0025.html
(in case you missed it)

Edited 2007-10-29 21:29

Excellent references.

Anybody knows what is the maximum number of threads the FreeBSD 7.0 kernel can suppport before stabilizing or losing performance ? Maybe we can dream HPC on FreeBSD systems....

The presentation showed an 8-core CPU. So you get performance up to 8 concurrent threads after that things will degrade "a little bit".

If you have more CPUs then you should get more benefit like if you have 128 cores:P but freebsd 7 is still not optimized for that though they are working on it so if you have the hardware you can always join the mailing list for testing and performance improving.

Jeez, you don't understand? I'm not disputing the numbers, or the graphs. xBSD may perform better than Linux, or Windows.

Just never trust the people who make something to tell you how much better it is than anything else.

Here are some examples:
Microsoft: http://download.microsoft.com/download/0/7/1/0715a190-70f5-4b0d-8ce...
LightHttpd: http://www.lighttpd.net/benchmark
Postgree: http://www.postgresql.org/docs/techdocs.83.html

>Just never trust the people who make something to tell you how much better it is than anything else.

No just educate yourself first, all of these things are explained at the current mailinglists. BSD is no hype-ware. The configuration of the benchmarks are done with help from some Linux developers too.

May be you don't understand what you are saying?

One time your telling me "Just never trust the people who make something to tell"
then you tell me "I'm not disputing the numbers"

"Not Trusting" is to doubt, suspicion, lack of confidence, etc.
"Dispute" is disagree, clash, question, impugn, etc.

Consequently, by distrusting you are disputing the facts provided somehow.

I believe you are contradicting yourself in your msgs.
The examples you listed don't mean anything they just say what the vendor is promoting.

So, in other news, Microsoft managed to push Windows Desktop Search to millions of corporate users without their consent.

If Microsoft were to say that they now had the hightest market share in the Desktop Search market, then I wouldn't dispute the fact, but I wouldn't trust the announcement. See?

I can't dispute that they have the most installs of WDS.
BUT
I don't trust the fact that a marketing person is implying that WDS is the most (willingly) used Desktop search software on the market.

You understand now? No? Try this: http://www.amazon.com/Logic-Dummies-Math-Science/dp/0471799416

lighttpd is faster then apps mentioned in their benchmarks (true)

postgresql is faster/better scale then mysql (true)

If you do not see the diffrence between MICROS~1 and FREE Open Source projects, then I would say poor you ...

postgresql is faster/better scale then mysql (true)

No it isn't True. I'm a postgres user and generally prefer it to mysql, but saying that it is unqualified faster than mysql is pure non-sense. Both have their strengths and weaknesses and which one is fastest depends on all kinds of factors like what kind of data you have and what kind of queries you are doing.

No trust the LKML, it was in discussion there.

Being critical of a source is not rejecting it blindly simply because of its nature, but rather assembling evidence to support this criticism that one rises - and the apply the same pragmatics to that set of sources as well in order to justify one's argumentation and to support a certain angle to the subject.

Just don't trust the graphs. Never trust data provided by the manufacturer, especially when it shows their product as the best.

May be I can give a little correction:

Just don't trust the graphs [if the result doesn't favor your OS]. Never trust data provided by the manufacturer, especially when it shows their product as the best[, but trust it if your OS is shown as the best].

Lol, and how is FreeBSD 7 is going to change that? FreeBSD just offers a different kernel and a few low-level libraries. The rest is pretty much the same. X.org, Gnome/KDE, firefox, openoffice....

What's different between Ubuntu and FreeBSD? Well, quite a lot, like, for example, the whole concept and way of managing software. Available file systems are different. Installation is different. Hardware support is quite different. Community, philosophy and documentation is different. Etc.etc.

You can also use Firefox and OpenOffice on MS Windows or other operating systems but the underlying OS is still completely different. Also, for example, X.org for FreeBSD is not 100% the same thing as X.org for Linux.

Edited 2007-10-29 22:05

What do you mean?
That's like saying Firefox for FreeBSD is not
the same as Firefox for Linux.
It's surely the same source code.

"It's surely the same source code."

Yes, but depending on the importance of the software in question programs may have various dependencies etc. that may differ in various operating systems and that need to be taken into consideration too. Programs are often not only static separate pieces not interfering with each other but often share some libraries, have certain ways of communicating with the kernel etc. Those things may be different in different operating systems.

If the program is only some relatively small 3rd party extra app that does not have many dependencies and does not interfere much with the rest of the system, then it may be quite possible to just compile it from source code without much more thinking and headaches.

But porting important central pieces of software, like X.org to different operating systems requires testing, and often tweaking software is needed too. It took a rather long time for FreeBSD to have a stable modular X.org version, while many Linux distributions already had a working X.org. Software may need to be patched and tweaked to work well in a certain OS, or even when not, the rest of the system, like various other apps that depend on the new software (like X.org), may need to be tweaked and at least tested well in order to work well also within the new framework.

Edited 2007-10-30 11:56

That's silly. There are fundamental differences between FBSD and Ubuntu.

Different base system, different base system _management_, different software installation, different security features, different filesystem layout strategies, different approach to *nix.

That's off the top of my head...

You shoud first inform yourself about the BSD or especially the FreeBSD operating system. It's a whole system, not just kernel and some libraries.

There is a lot more different between candy-fied linux and FreeBSD. FreeBSD is a wholly different OS than GNU/Linucks. Oh, and the "OS" is not the desktop. The OS is what sits between the hardware and applications - and there is a lot that goes into it.

True, it does provide a different kernel and much of its own libraries, but its entire guts are different: the scheduler, network stack, memory management, file systems, etc.

It also provides its own "userland," default tools, and features. I attribute ignorance to statements like this, and I hope you one day get a clue.

Dont know if you are joking or not but I actually wanted to dabble in FreeBSD and LFS now having settled down into Linux , with Ubuntu as my main desktop distro.

Its so easy and works so well,its not boring really but its no challenge.I want to learn more and be challenged:)

That's exactly what I meant. I'm thinking of installing it on another harddrive to test something new.

Cool then.Good luck with it and cheers.

The long wait never bothered me because I usually start it just before going to sleep or leaving for work. When I wake up or get back (sometimes the same thing) long compiles like X or KDE are finished, and this on a P3 1.13GHz laptop without a nice fast disk.

Even compiling FreeBSD world and kernel isn't too bad. I just hope I can get 7 to compile, doing it on this laptop that had 6 on it didn't work (installed 7 on new / slice and kept the old /usr slice).
So now I'm copying my home and data directories to another server and wil go for a complete clean install.

That's fine, but you're probably not working on the system. Sometimes you just need some software to do your job NOW and unless you're lucky and the thing isn't in C++ and doesn't have megabytes of sources, you're going to have to waste time waiting..

That's wasting TIME AND ENERGY.

Don't get me wrong, I love their kernel and base system, but hate the ports. (and packages are usually too old, or not available).

Yeah because not having the absolute latest sub-sub-sub release of firefox would cause massive issues with "getting work done." For god sake just install the latest available package (hint. you can change PACKAGESITE to point to the bleeding edge crap on the pointyhat build cluster) and then compile the bleeding edge latest sub version, super duper, prolly gonna crash and cause more loss of getting work done version.

... The idea of having to build KDE*.* just because KDE 3.5.8 is released is problematic.
Do I have to get KDE 3.5.8? No. Do I want to get 3.5.8? yes. Do I want to compile it using ports? No.

Binary packages are there for a reason - doing apt-get install or yum install XX is faster and for most users easier then using ports.

P.S. The Firefox example was a (very) bad example.
Give the fact the most post-release updates are security-oriented I doubt that using an old(er) version can be considered prudent or even sane.

- Gilboa
P.S. Before I get the usual flame, IMHO pkg_add cannot rival apt and yum. Life sucks.

Edited 2007-10-30 17:26

The story of BSD systems on laptop hardware is not exactly a love story. And while I am also totally commited to get FreeBSD to install on my IBM Thinkpads I must also acknowledge that this is not really the platform that it was designed for.
And anyway, my Debian systems (running and humming away on this hardware) get as close to the FreeBSD *nix approach as Linux can get.
Best of luck anyway. I still have trouble instaling it what seems to be related to an old BIOS bug that FreeBSD has no exception handling for.

I've had the luck with this Dell C810 that I was able to run FreeBSD as of 4.8 on it. That was the time I started using FreeBSD so I don't know if older version would run as well.

The only thing that bothered me about FreeBSD is that you have to set so many things manually; auto-mount drives, GDM, etc. I want to be able to build from a base install with no fuss.

But I like FreeBSD. Fun to geek on.

Check out DesktopBSD and PC BSD.

Not to mention that sysbench is one of those benchmarks that gives very different numbers depending on very subtle changes in semantics (and mysql versions). And that sysbench is just one benchmark...

At least they (Linux kernel developers) could find some heavy bug, while using these benchmarks ;-)

It depends what you use ports for. I would say ports is terrible for Gnome, Xorg, OpenOffice, and other bloated monsters (generally desktop use.) However, it is great for daemons and server tasks as it takes minutes to recompile the entire world, if necessary, and you get to compile the exact options you need.

... it takes minutes to recompile the entire world, ...

I like FreeBSD and NetBSD. And I like that recompile the world concept. But I have only succeeded at it once, That is, I only ended up with a stable system on one of my several attempts at it. And I usually run BSDs on cast-off low-power machines which take forever to compile anything, so I'm very sympathetic to the desire for apt or similar, or at least better archives for pkg_add.

I have recompiled world close to a bagillion times. I don't think I've ever had it crap out on me once. The only time I've had trouble is following current instead of a stable branch... even then, current is usually pretty reliable unless you're unlucky.

apt is no silver bullet. My ubuntu system has had all sorts of little hiccups when upgrading to new releases... things like xmms sound suddenly quit working (and never worked again), or other programs just got screwed up. And even worse -- no error in the upgrade process was reported... the problems virtually impossible to solve.

That being said -- what FreeBSD really needs is a SANE update system for ports/packages. It's not so much that ports sucks, or pkg_add sucks... it's that there's no really good high level piece of software that can update stuff safely. I know some of this stuff exists -- but hey -- it tanked my system too many times so I quit using it.

First read FreeBSD Handbook [http://freebsd.org/handbook] then spread FUD/shit. FreeBSD offers BOTH building from source thru Ports and adding binary packages using pkg_add and handles dependencies the same way apt does, or even better, because it is able to detect and fix loop dependencies.

pkg_add is much less efficient at everything than apt[-get]. Sorry to put you out of your illusion but I used both and it's like comparing rotten tomatoes to apples.

apt-get works within seconds, and since the WHOLE system is packaged with it, you don't get crappy situations with ports/package mixups (yes, userfault, but sometimes required, eg: when packages are too old and you just NEED that latest lib).

Sorry to put you out of your illusion but I used both and it's like comparing rotten tomatoes to apples."

I also used both (and still use), I just do not get why you jerk off so much about apt, its not bad, but its also not such that awesome.

And how about latest versions of software avialable in apt?

Do they come from XX or XXI century?

"since the WHOLE system is packaged with it, you don't get crappy situations with ports/package"

That is consequence of (not well thought) mixing of ports and packages, Debian limits you ONLY to use bianries, in FreeBSD you can use binaries, sources, or even both, but only if you KNOW what are you doing.

You can also use only binary packages like in Debian you know?

Yes, but I'm comparing Ubuntu to FreeBSD on the desktop in regards to packages here

Sorry, I should have made it clearer. With ubuntu you get at most 6 (+ initial age of import) months old packages, which is quite ok unless you REALLY need bleeding edge, but that's usually ok to go from source manually (eg: you're most probably not going to need more than few things from source).

In FreeBSD the packages are (by default) quite old. Take latest release for example. Now, I know that you can change the repository to an updated one, but those are usually just broken. (I guess this has to do with userbase too tho). Pkg_add is also simply slow..

FreeBSD might be a killer server but it's simply not a good desktop IMHO (as I said before, I think that the FreeBSD kernel and base system is superb).

I'm probably biased tho, because I used it for desktop in the "turbolence" period when things got changed around. For example I had these experiences:

1. renaming of "libgtk12.so" to "libgtk-1.2.so" (breaks binaries and some compilation of non-ports)
2. SDL renaming (sdlxxx1.1 to just sdl)
3. X11R6 -> /usr/local (btw. I mentioned the idiotic idea of putting special dir for X in your channel some years ago and got kicked for it.. I guess I should be the one laughing now)
4. python update (one day it was 2.5, next day it was 2.4, guess why...)

I think the ports/packages system might "work" if there was much more users and willing testers with quick-build testfarms for at least x86 systems so you couldn't break your system by updating eg: python (or just "update everything")

I'd really love to be on freeBSD, I think it is based on better design and generally better ideas. Sadly tho it has also a lot of incosistencies on the upper layers. I personally like it more than linux tho.

Beep, wrong, you can use binaries and source packages in Debian too, just as easy as in Gentoo or FreeBSD (probably easier & better).

Yes the Debian bloatware-management. It fails very often, so don't bother with it. FreeBSD is a source-based operating system like Gentoo. It's safe to use pkg_add shortly after the release or if you do know other sources (and there are a lot of it) with the latest binary packages. So yes it's indeed the users fault - nothing more, nothing less.

"pkg_add is much less efficient at everything than apt[-get]. Sorry to put you out of your illusion but I used both and it's like comparing rotten tomatoes to apples."

More like comparing pipes to taps. If you want compare pkg_add to something, that should be dpkg.

if you read the handbook or you can also try to search on google, you will observ that FreeBSD has also binary updates, security binary updates, ... like debian or any different distro.

4.3? Are you kidding? I still have my 2.x CD's! Now let's wax poetic about days of yore... ;-)

I still have my IBM CPM/86 floppy disks. Get off my lawn.

Okay, I did not need that mental image this early in the morning. -.-

:-P

I am not sure what you are saying. The NetBSD numbers in the graphs in this PDF were the absolute best case scenarios following from the benchmarking you cited, i.e. using the experimental scheduler with experimental libpthread. The "stable" NetBSD code has much lower performance, so this is being somewhat generous in only graphing the experimental code.

Ditto with Linux; the 2.6.22 data is the best that Linux have been able to produce with glibc (and they regressed significantly with 2.6.23). It is still "not good enough" on this test.

Almost the same can be said about the FreeBSD ones, none of the best results compared are production ready yet.

What I was trying to point out is that all them are making strong progress and that this dispute is very good to make the developers even more aware of the weakness and virtues of their systems. This can be better emphasized by "http://kerneltrap.org/Linux/Measuring_Process_Scheduler_Performance... on linux side.

Why I would like to see the results updated? For sure not because I want Linux or FreeBSD to win over each other but because:
- that would make the scenario a bit more fair to each camp, showing the real progress of each team;
- this would put a bit of pressure on the developers to tune their algorithms and try harder to improve the weak points.

Doesn't it looks good for all?

It looks a brighter future than ever.

The FreeBSD numbers are for the version that is currently in beta. This means it is basically ready now but undergoing minor polishing in the lead up to the release process. You can choose to interpret this as "not production ready" if you like

I still dont understand what you mean by wanting to "see the results updated". What updates are you asking for, when it is already the most current data being graphed?

Nice LaTeX presentation!