A recent patch submission from Bosko Milekic features on-demand memory allocation for mbufs and flexible sysctl tuning options – no more fiddling with NMBCLUSTERS and recompiling. The code does not appear to be committed yet and could probably stand some testing by the willing user population.
FreeBSD: Dynamic Mbuf Allocation via UMA
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
If UMA mentioned is the Universal Memory Allocator is makes IMHO a lot of sense.
Carsten
What it means network buffers are dynamically allocated. ^_^
For more info check out and search for “kern.ipc.nmbclusters”…
http://www.freebsd.org/cgi/man.cgi?query=tuning&apropos=0&sektion=0…
wow im glad im not the only one that thought so haha. least it sounds like a good thing.
Alright, no more need to recompile the kernel when network traffic increases.
you don’t really need to recompile, as @Velox mentions a loader knob you twist.
It seems people really like to recompile their FreeBSD kernel. Both my boxes run GENERIC, one is a desktop, and the other a server. I just load and tune what I need to, and it makes doing updates easier.
It seems people really like to recompile their FreeBSD kernel
Obviously I can’t speak for everybody, but from the people I know, it tends to be recent converts and refugees from the Linux world that still think that everything requires a kernel recompile.
What’s wrong with recompiling your kernel, you can take out some of the options you don’t need add some that you do need. Sure beats having a boat load of stuff you probably would never need or use.
Never said it requires a recompile ^_^
Never needed to do that befor. All it required was tweaking the a few of the sysctl’s.
what is it about the freebsd (bsd in general?) kernel that makes it less needing of recompiles compared to linux?
Disclaimer: the only kernel i ever compiled was gentoo
Hi
Somehow freebsd people need to bash linux people. see how linux migrants are called “refugees”. it somehow conveys the images of a angelic freebsd.
FreeBSD is a operating system that can stand on its own without bashing anybody else.
regards
Irek
Isn’t this something Linux and Windows have done since like 1992?
Somehow freebsd people need to bash linux people. see how linux migrants are called “refugees”. it somehow conveys the images of a angelic freebsd.
Don’t be dim. I wasn’t bashing Linux people by calling some of the switchers refugees, I was bashing Linux for being crappy enough to have caused the need for so many switchers!
BTW, I am well aware that no software is perfect, and I am growing rather tired of having to type this particular disclaimer every time I say anything even remotely true about how bad Linux can be, and how other things are better.
OSNews needs user profiles methinks.
Hi
“Linux world that still think that everything requires a kernel recompile.”
thats a big lie. what do you mean everything requires recompilation. what the key difference in the freebsd kernel that makes it not require a recompilation at all while linux needs a recompilation for *everything*
dim?
regards
Jess
I am dim?
Seeing as how you failed to grasp what I was bashing, then yes, I’d say you were being dim.
Hi
“I am dim?
Seeing as how you failed to grasp what I was bashing, then yes, I’d say you were being dim. ”
you said i was dim. i questioned that. i understand your points well but question your logic. you havent answered the questions. this is a discussion about what you commented on linux. answer that please. you can resort to name calling somewhere else
regards
Jess
Hi
“hat is true. I’ve not answered any of your questions regarding Linux.
Furthermore, I did not call you dim, but said you were being dim. There is a distinction. Had I thought that being dim was in your nature, I’d not have told you to not be dim, as it would have been as pointless as telling a fish to try and breath the air.”
fine. now can you please answer the question. you have been blatantly telling that everything requires a recompilation in linux and calling linux a crappy thing. well what feature does freebsd kernel have that makes it not require a recompilation at all.
i have been using both freebsd and linux for quite sometime and never required a recompilation of the kernel for linux. both are modular kernels and functionally similar.
the end user experience with a desktop system is in favor of linux while i like the ports system i dont think it has anything superior over a good yum repository or debian package lists. i dont understand the need to bash other operating systems because you like freebsd. just point out where freebsd is good at and drop the anti linux or anti windows thing. my os is better than your os is a pretty bad argument
regards
Jess
you have been blatantly telling that everything requires a recompilation in linux
Incorrect again my friend! I’ll quote my original post here:
Obviously I can’t speak for everybody, but from the people I know, it tends to be recent converts and refugees from the Linux world that still think that everything requires a kernel recompile.
I said that it’s converts that think things require the recompile, not that they actually do.
i dont understand the need to bash other operating systems because you like freebsd
I bash everything that bothers me, even in systems I prefer. It gets people thinking, and it can be kind of fun.
As to some of the things about Linux that make me call it crappy, here goes:
* The documentation sucks. A lot. It’s lacking, outdated, and if you’ve ever tried to submit updates or cleanups, you’d know that the maintaners aren’t always easy to convince of the need for it.
* No kernel debugger, no crash dumps by default. Very bad for ensuring the quality of the software.
* The new system filesystem (sysfs). BSD used to have something similar (NetBSD still has it IIRC) but it was dropped from most as better (non-fs) approaches were implemented (like sysctl etc.) What is wrong with using ‘/proc/sys’?
* The ‘NFSD filesystem.’
* The fact that it is not a complete system, nor is it developed as part of one. (which is likely a contributing factor to the appearance of monstrosities like sysfs and nfsdfs.) This has lead to hundreds if not thousands of incompatible distributions; a situation that could have been largely avoided by developing it as a complete minimalistic system.
* The fact that version control was not used until quite recently, and that a non-free system is being used to develop the so called flagship OSS product (this is more a quirk in my thinking than something that makes Linux crappy).
* The lack of a reasonable method of preserving backwards compatibility. Linux is abysimal here. FreeBSD does backwards compatibility very well in comparison.
* The fact that large changes are made in the so called stable branch (like the VM wackyness in the early 2.4 series. Why was that not done before 2.4.0 was released???).
* The fact that it has grown very large in size and added not quite so much new functionality as one would expect for that dramatic increase. Sure FreeBSD (for example) doesn’t support every filesystem that Linux does, but it’s largely equivelent in functionality, and a fraction of the size. Heck, for a few tens of megabytes more, you have a complete system with a BSD. Mind boggling.
There are more reasons, but I’ll let you ponder the ones I’ve given you for the time being. Also, many of the other issues I have with Linux is also common to the BSDs, including the ones I prefer, so they are not Linux specific.
* Recompiling the Linux kernel is ridiculously difficult and time consuming, and the recent changes in 2.6.x only do a piss=poor job of hiding that fact.
Hi
”
* The documentation sucks. A lot. It’s lacking, outdated, and if you’ve ever tried to submit updates or cleanups, you’d know that the maintaners aren’t always easy to convince of the need for it.
freebsd is a complete operating system unlike linux and its more easier to write documentation for the core systems than multiple distrubitions. hence tldp.org takes a different approach. the individual distrubition documentation has been quite good. take for example redhat or gentoo documentation. the rest of the programs have different levels of documentation which affects freebsd too.
* No kernel debugger, no crash dumps by default. Very bad for ensuring the quality of the software.
wrong. there are multiple versions of kernel debuggers. its just that nothing is there in linus tree. it hasnt affected the way linus developers his code.
* The new system filesystem (sysfs). BSD used to have something similar (NetBSD still has it IIRC) but it was dropped from most as better (non-fs) approaches were implemented (like sysctl etc.) What is wrong with using ‘/proc/sys’?
sysfs is a virtual filesytem for system device hierachy. whats wrong with it?
* The ‘NFSD filesystem.’
whats wrong?
* The fact that it is not a complete system, nor is it developed as part of one. (which is likely a contributing factor to the appearance of monstrosities like sysfs and nfsdfs.) This has lead to hundreds if not thousands of incompatible distributions; a situation that could have been largely avoided by developing it as a complete minimalistic system.
its a kernel. distributions take care of integration. the one size fits all approach wouldnt have enabled different approaches like gentoo or mandrake to exists with the core software. linux is about enabling dynamically different approaches to compete. you dont use linux. you use a linux distrubition. how is this affecting the end user experience?. this is the approach thats has enabled multiple commerical distributions to exist and compete. linus has never been interested in userland. thats his decisions. if you want a minimalistic system use a minimalistic distrubition
* The fact that version control was not used until quite recently, and that a non-free system is being used to develop the so called flagship OSS product (this is more a quirk in my thinking than something that makes Linux crappy).
version control was not used?. cvs was used for quite long time and is still used by many kernel developers.linus prefers to use bitkeeper. if you dont want to use bitkeeper there is a cvs and subversion gateway available. this is no way makes linux a crappy operating system. its just linus’s personal choice for his version control software.
* The lack of a reasonable method of preserving backwards compatibility. Linux is abysimal here. FreeBSD does backwards compatibility very well in comparison.
freebsd uses a evolutionary approach which has both advantages and disadvantages. linux has always put out effort in not breaking userland unnecessarily. internal api’s do change radically which has enabled removal of compatibility cruft and enables it be developed more faster.
”
* The fact that large changes are made in the so called stable branch (like the VM wackyness in the early 2.4 series. Why was that not done before 2.4.0 was released???).
”
a stable series means that the particular series would attain stability after a few versions. after it was marked as stable the field testing exposed problems with the vm which was the reason linus decided to swap it. distributions continued to evolve the older vm at times. this was controversial but i never experienced any problem at all.its been 3 years or so. so thats not the current problem.
”
* The fact that it has grown very large in size and added not quite so much new functionality as one would expect for that dramatic increase. Sure FreeBSD (for example) doesn’t support every filesystem that Linux does, but it’s largely equivelent in functionality, and a fraction of the size. Heck, for a few tens of megabytes more, you have a complete system with a BSD. Mind boggling. ”
linux and freebsd have very different approaches here. linux supports far more filesystems and stuff that freebsd doesnt or need to. there is also the tiny tree for these purposes. many of the modules are optional and the binary isnt much different at all. so your point is moot.
i havent answered you comprehensively but you need to understand that the development methodology is different between freebsd and linux and both of these sides have advantages and disadvantages.
i could honestly say that none of the alleged problems that you mention with linux is a big thing and has not affected me at all. sure linus could have waited for linux 2.5 to change the vm or something like that but these are not the reasons to call linux a crappy operating system. i could say such problems with the present status of freebsd 5 or the problems with ports systems not taking a approach similar to openbsd
regards
Jess
“Isn’t this something Linux and Windows have done since like 1992?”
From what I understand, this patch just removes the hard-limit for mbuf cluster pools and allows dynamic network resource allocation via UMA, the general kernel memory allocator in BSD (as opposed to a separate mbuf specific allocator that used the hard-limits). The question shouldn’t be whether or not this was already done in Linux or Windows, the question should be *how* it’s done.
Granted, Linux implemented the slab allocator around ’96, so chances are they may have made a similar modification much ealier than BSD. I’m primarily a BSD user, so I don’t know how sk_buf’s (that would be the equivalent, right?) are allocated in linux and how that method compares with this patch.
Maybe some folks here can actually *read* the referenced post and make some knowledgeable comments on the differences of the implementations… beats arguing, no? Better yet, maybe some BSD users here could test it instead of feeding the trolls… that beats arguing too.
I am well aware of the differences in development models, and the reasons for them. It doesn’t change the fact that I’m right.
As to the kernel debugger, it needs to be something that is used everywhere, especially in Linus’ tree, as his is what everybody syncs up with.
As to Linux just nebing a kernel and it’s up to the distributers to do the integration, this fact has produced countless systems that are not especially distinct from each other, while at the same time being ridiculously incompatible with each other. Seeing as Kernel.org mirros GNU code anyway, there’s not good reason for Linus to not develop Linux as a complete OS, thus removing the vast majority of the compatibiliy problems while still allowing folks to customize their distributions.
Version control was not used in Linus’ branch until he jumped on the BitKeeper bandwagon, and the fact that other developers used it it was about as relevant to Linux’s evolution as the fact that FreeBSD uses it, because again, everyone goes back to Linus’ tree.
As for the backwards compatibility argument you made, it’s not so much that FreeBSD takes a more evolutionary approach to software design, but the fact that they have very well maintained compatibility libraries that allow one to use older binary only software under modern versions. Linux has NOTHING like this.
As to these issues not being a big thing, well, I’d never trust my business to Linux until my issues with it have been properly adressed. I do not trust in Linus’ ability to properly assess and assure the quality of Linux, no matter how good a coder he himself may be. There are too many things that are going unnoticed because of his stubborn refusal to use basic QA/QC techniques that have been standard practice in the larger programming world since before Linux was first written; things that apply moreso now than ever, even in OSS.
And correct me if I’m wrong, but the VM issues were rectified around 2.4.10, which is a tad outside of the ‘first couple of revisions.’
and as to my problem with NSFDFS, it exists. Another damned virtulal filesystem! Linus seems mad with the love of those… I can’t stand the everincreasing overabundance of them in Linux.
Hi
Linus says that the debugger allows him to walk through the code which isnt the way he debugs his code so he isnt using one. Andrew for example using the debugger. there is no single freebsd like thing. linux developers have their own trees and work in the fashion they like. it doesnt dictate the development process as a whole
“And correct me if I’m wrong, but the VM issues were rectified around 2.4.10, which is a tad outside of the ‘first couple of revisions.’ ”
let me explain that. fedora is about to ship a 2.6 based distro while the exact version is going to be something like 2.6.5. so its only a few revisions if you consider what is actually being shipped out by distributions. redhat was even more cautious that the fedora policy by only backporting features. they didnt swap the vm like linus did.
you got it all wrong when you say that everyone syncs with linus tree. linus is active in the development tree not the stable tree. he isnt managing the stable tree at all for quite sometime. in 2.2 it was alan cox. in 2.4 it was marcles. in 2.6 its andrew. in 2.8 its going to be someone else. so you dont need to trust linus or those guys managing the code. you need to trust the distributions like redhat or debian and they have been doing a pretty good job
regards
Jess
Hi
“and as to my problem with NSFDFS, it exists. Another damned virtulal filesystem! Linus seems mad with the love of those… I can’t stand the everincreasing overabundance of them in Linux.
”
linus didnt manage or code nfs. why do you attack him personally?
whats your problem with them. they are not visible to the end user at all. its they way the kernel does its tasks internally. i dont understand your issue with them.
regards
Jess
In Linux, you have to specifically choose whether a driver is compiled into the kernel, compiled as a loadable module, or not compiled at all. This means, if you don’t mark a driver as “module” and later add that hardware, you have to recompile the kernel.
In FreeBSD, anything not listed in the kernel config file is automatically compiled as a loadable module. This means, if you later add or change hardware, you just have to load the module for it. No kernel recompile needed. (You also have the option of choosing whether modules are built with the kernel, with the world, or not at all.)
That’s the major difference in the two.
I still am not conviced by your reasoning, but I’ve nothing more to add ATM. Thanks for your input however.
Hi
So with freebsd you cant choose whether you dont want it all. you will have to have it as a module anyway?. is that what you are saying.
if i use linux i could easily compile everything as a module and have the same state as freebsd. this doesnt seem to me anything big.
regards
Jess
Okay, you twisted my arm
linus didnt manage or code nfs. why do you attack him personally?
Because he’s the big guy, and he like me, is stubborn.
whats your problem with them. they are not visible to the end user at all. its they way the kernel does its tasks internally. i dont understand your issue with them.
Better technology is always preferable to me than the most popular, or the merely good enough. Linux is not IMO the better technology, but perhaps with prodding by assholes like me, the devs will take it in a direction that would cause me to have fewer issues with it, and every one of it’s users would be better for it.
Hi
“Because he’s the big guy, and he like me, is stubborn. ”
yes. Linus is pretty thick headed but he also admits where he is wrong and is always willing to accept a superior thing once its proven.
“Better technology is always preferable to me than the most popular, or the merely good enough. Linux is not IMO the better technology,”
a virtual fs doesnt in anyway make linux the “merely good enough” technology. you havent presented your arguments on whats wrong with the vfs.
“but perhaps with prodding by assholes like me, the devs will take it in a direction that would cause me to have fewer issues with it”
fortunately or unfortunately only code matters. if you can code better and present a solution its likely to get accepted. there are already a lot of whining assholes in lkml. they just get ignored.
i think you are too close to freebsd to understand the differences or why a particular approach was taken in linux
the excellent digest at kerneltraffic.org is very useful if you want to keep track of whats going on. keep using freebsd but dont bash other operating systems.
instead of telling me linux is crap just tell me certain particular features are better in freebsd. we are likely to agree on that
regards
Jess
i think you are too close to freebsd to understand the differences or why a particular approach was taken in linux
Not correct, FreeBSD does a lot of things that piss me off as well, but you asked me about Linux
To be clear, I do not think that FreeBSD is the better technology either, it’s just less irritating to me personally.
Hi
“Not correct, FreeBSD does a lot of things that piss me off as well, but you asked me about Linux
To be clear, I do not think that FreeBSD is the better technology either, it’s just less irritating to me personally.”
cant argue about personal preferences. have fun
jess
Please, can you start to use the <i.> and </i.> ( italics) for the quote instead the quote (“) in your future replies? It’s kind of little hard to read in your replies with the quotes. Thanks! 🙂
BTW: Remove the ‘.’ dot above for the italics HTML tag.
😉
Hi
Yes. I will try that. thanks for the clue
regards
Jess
>So with freebsd you cant choose whether you dont want it >all. you will have to have it as a module anyway?. is that >what you are saying.
No, you can specifically specify to make.conf not to compile everything not listed as modules, but in general it makes more sense to compile unused things to modules to avoid the necessity of having to recompile should hardware configurations change.
>if i use linux i could easily compile everything as a >module and have the same state as freebsd. this doesnt >seem to me anything big.
FreeBSD’s kernel configuration is a hell of a lot easier than Linux’s. You can easily jump around in FreeBSD’s kernel config, set only the thing you want/don’t want, and be done with it in 10 minutes. Linux instead takes 30 minutes prompting you with every nitty gritty detail and even fails to do dependency checking. Meaning you could enable cdrom drives and not even load atapi/ide drivers and other idiotic things.
Better technology is always preferable to me than the most popular, or the merely good enough. Linux is not IMO the better technology, but perhaps with prodding by assholes like me, the devs will take it in a direction that would cause me to have fewer issues with it, and every one of it’s users would be better for it.
Although you probably think otherwise, Linux has dozens of developers, including Linus who know more about “UNIX”, operating systems design and theory, software development, project management, etc than you ever will. No offence to you at all because we’re talking about very bright people.
Prodding by assholes like you will cause the devs to killfile you and that is about all. The arguments you presented against sysfs and nfs filesystems, for example, are rambling and incoherent and basically amount to “Damn Linus and all his virtual filesystems”.
I challenge you to lecture Linus, Alexander Viro, Dave Miller, et al about how “it should be done” on the linux kernel mailing list, if you think you can do it so much better.
The fact that it has grown very large in size and added not quite so much new functionality as one would expect for that dramatic increase. Sure FreeBSD (for example) doesn’t support every filesystem that Linux does, but it’s largely equivelent in functionality, and a fraction of the size. Heck, for a few tens of megabytes more, you have a complete system with a BSD. Mind boggling.
When you take out all the arch code (Linux has about 20 supported architectures), all the drivers (Linux has far more than FreeBSD), all the filesystems (ditto), the core code is comparable. A Linux kernel binary can be very small quite easily (I’d say tighter than FreeBSD). In fact, there is a simple patchset that allows one to boot the 2.6 kernel with 2MB ram.
The Linux 2.6 kernel is also far more advanced in terms of performance than the FreeBSD kernel.
I do know my way around a good design (and am quite handy at identifying flaws I might add)
Heh, yeah like the “everything requires a kernel recompile.” flaw that you identified in Linux. Tell me more about that one would you? How does FreeBSD work around this problem?
Oh, one more thing:
As for the backwards compatibility argument you made, it’s not so much that FreeBSD takes a more evolutionary approach to software design, but the fact that they have very well maintained compatibility libraries that allow one to use older binary only software under modern versions. Linux has NOTHING like this.
OK, I can run libc5 a.out binaries that are 10 years old on my modern Linux system (2.6 kernel). So what exactly does FreeBSD have that Linux doesn’t? I suggest maybe you were talking out of your lower mouth. Please refrain from doing that in future, it has very bad breath.
I’ve also mentioned above that I didn’t think of BSD as the better technology. Perhaps that escaped your causual browsing either?
But then again, like so many Linux zealots, you’ve no room in your head for anything from someone who is not a rabid Linux fan.
And though he is good, Linus is far from the best. I’ve met better.
And I wasn’t up in arms about the fact that you made a prediction, it’s that you have the balls to be so sure about it; something that you have absolutely not enough information to base it on.
Obviously I can’t speak for everybody, but from the people I know, it tends to be recent converts and refugees from the Linux world that still think that everything requires a kernel recompile.
Can you? As fun as it is lowering myself to be on the same level as you Linux tits, I think I’ve blown a month’s worth of innane babble here.
Where are those Sun posters when you need them. They are quite effective at cutting you fuckers down to size, even though you’d never realize it.
You fucking tit, I can take alot, but man, I already explained that one.
Yeah, it was just an unfounded outburst of FUD that you happened to get caught on, and had to explain your way out of.
I’ve also mentioned above that I didn’t think of BSD as the better technology. Perhaps that escaped your causual browsing either?
You seemed to think so until people started asking why.
But then again, like so many Linux zealots, you’ve no room in your head for anything from someone who is not a rabid Linux fan.
I’m not the one starting the personal attacks and spreading fud.
And though he is good, Linus is far from the best. I’ve met better.
I didn’t say he was the best. But no he is not “far from the best”. He architected and has done a lot of coding for one of the most advanced OS kernels out there. Who have you met who is better, by the way?
And I wasn’t up in arms about the fact that you made a prediction, it’s that you have the balls to be so sure about it; something that you have absolutely not enough information to base it on.
No, you are the one with the balls to state things as fact on topics you have no idea about, not me. My information that I base it on is 1. Linus is one of the best in the world, and 2. You seem to be a rambling imbicile.
From these two pieces of information, I conclude you’ll never be as good as Linus.
And just another reminder, so you don’t skirt the question again: Who have you met who is far better than Linus at OS design / software architecture?
>And just another reminder, so you don’t skirt the question >again: Who have you met who is far better than Linus at OS >design / software architecture?
Gee, man, you can’t even imagine how many extra-talented folks out there who are not known to the public. Linus is talented, no doubt, but putting the crown on his head is just childish and manifestation of linux-zealotry
You dumb beer swilling goat bugger. That is not be contradicting myself, that’s me knowing when a crude technique has a small chance of working in my favor (as in here) and when it does not (as in on the lkml).
What is it with Linux zealots and being toatlly fucking brain dead?