Slashdot Mirror
NetBSD 2.0 RC5 Tagged
ulib writes "NetBSD 2.0_RC5 has now been tagged. Changes since RC4 include fixes to various COMPAT_ emulations, IP Filter backward compatibility fixes, XFree86, pax(1), rsh(1), hp300 boot blocks, pthread fixes for amd64 and i386, documentation updates. Binary snapshots of NetBSD 2.0_RC5 are available in the daily builds directory on the main FTP site."
Posting works. It's morning EST; the nerds slumber. I for one am pumped about 2.0. I'm a recent convert from Linux and I like NetBSD's installer as well as its bloat and crap-free default software arsenal.
With Fluxbox, the GNU coreutils, and bash, my P133 makes a reasonable desktop workstation. Though Linux would work, the low hard-drive footprint of a NetBSD install is what makes the installation trouble-free. Comparable modern Linux distros seem to me to take time to whittle down to a sub-300MB install. With NetBSD, the core system with XFree uses only 290.
Does NetBSD have any plans to migrate? Or do they have a good reason for sticking with XFree86?
Licensing aside, most of the development seems to be happening at X.Org and that is a good reason to migrate.
any other reason why I would want to use netbsd instead of freebsd in an x86 machine?
THanks for the tips
Sigs are dangerous coy things
- If you need top of the line security, use OpenBSD.
- If you need a small footprint, use NetBSD.
- If you need to run on an odd architecture, use NetBSD.
- If you need a great BSD operating system, and know that everything you'll use is in NetBSD's package system, use NetBSD.
- If none of the above meet your criteria, use FreeBSD.
Also, FreeBSD has better hardware support than the other BSDs as far as number of supported devices. FreeBSD works well, and ports is pretty good, but the cost of that is bloat (although it pales in comparison to what Linux users regard as bloat. A bloated FreeBSD is comparable to a modestly slimmed down Linux installation, although if you really bite down hard, Linux can be made a little smaller than FreeBSD, while still not being able to compare to NetBSD). Basically, if one of the other BSDs will work for you, use it instead of FreeBSDLastly, in time, DragonFlyBSD will edge out FreeBSD. Dfly is going to have far better SMP and clustering, and it's being done so well that performance on a UP machine isn't suffering.
Just curious (as I've never installed or maintained any *BSD system): Suppose you have a FreeBSD and lets say a Mandrake workstation setup, complete with XOrg, GNOME, KDE, OpenOffice, etc. Are we talking a substantial bloat increase with the Mandrake setup just because it's "Linux"? Is it more libraries, bigger binaries, more software package dependancies...? Or are you talking a default install of *BSD versus a typical default install of a modern Linux distro?
I'm sure "SlashdotMedia" will improve on all the wonders that Dice Holdings blessed us all with
Mandrake
I would recommend FreeBSD over Mandrake, if you're just looking for a non-Windows OS. FreeBSD's configuration takes place via direct manipulation of the text files that control the OS. It's generally not hard to find information on what to twiddle and how, and you end up learning more in the process. Mandrake (and Redhat, the last time I used it) uses GUI frontends to change these text files, and often uses its own nonstandard text files, so that anything you learn won't carry over to another Linux distribution. Furthermore, the last time I used Redhat, I had choices of several different GUI frontends to many different text files, some of which were changed in different ways for different purposes. Without X, I couldn't configure my system.
Is it more libraries, bigger binaries, more software package dependancies...? Or are you talking a default install of *BSD versus a typical default install of a modern Linux distro?
It's typical distribution size. Almost every Linux distribution has a large sprawl involved with what it's including these days. A default Linux From Scratch install contains expect and Perl, for example. The BSDs are built with a focus on taking away everything that isn't necessary. Linux distributions for the most part don't even try.
As for the system binaries, it depends on how they're built. You could build in the library routines they use, or compile them so that they need dynamic libraries, or you could use something like busybox to decrease the size dramatically. It's not a cut and dry situation, but most Linux distributions manage to be larger than the BSDs anyway.
As always, it comes down to what you personally need and what matches your skill level. For Linux, you're better off using Slackware or Debian or Gentoo because they're less likely to damage your ability to learn Linux deeply than something like Mandrake. Today, even though I have no trouble managing FreeBSD or Linux From Scratch, I'd probably have trouble managing a Redhat or Mandrake installation, but I could probably pick up Gentoo or Debian's way of doing things in a few days.
A commercial Linux distribution tends to install more programs and libraries by default than *BSD. On a *BSD it's more common to actually add the different applications you want from the ports system.
I understand the OpenBSD has its origins in NetBSD. I've been using OpenBSD for some time now and was curious if anyone knew any pros/cons of one versus the other. Thanks.
It's more than just base software choices actually. The BSDs have huge amounts of useful base system software, and depending on your needs you can get an amazing lot done without touching a single out-of-tree package. This includes relatively comfortable software development thanks to nvi, gcc/g++/etc, gdb, BSD make, and so on.
The difference comes in when you look at how bloated the software itself is. All the BSDs have libcs that are tiny (a couple of minutes to compile on even my slowest machines) and do everything a C library should, including full networking and everything. The GNU libc, which you'll find on every Linux system by default (there's a diet libc out there, but it isn't recognized), is a HUGE package that takes a very long time to compile and results in a hefty binary in the end. What does all this bloat go towards? Most say it's all because of its attempt at being completely internationalized, but this is hardly enough to warrant about a 10x size increase.
The same idea applies to all other software that wasn't imported from GNU. If you can do the same things smaller and more efficiently, do it that way. There's no point in having 90% of your source appeal to minor features few people will ever use. There's also a strict adherence to tradition where possible - nvi is kept instead of some stripped-down vim-alike (which would have more convenient features, for instance) because people coming from a BSD system a decade ago won't get culture shock. But all the same modern software is a 'make install' away on any of the BSDs.
The bloat difference in the Linux kernel and BSD kernels isn't even worth discussing. It's just not funny any more. Linux has inflated a LOT in recent times. I remember back when some 2.4 was about 25 megs tar.bz2, now look at it - 2.6.9 weighs in at 35 meg. Is it really 40% more functional? Nowhere near. If anything it should be getting smaller, since they insist they're refining to simpler algorithms that should work faster and take less code.
The NetBSD 2.0RC5 src/sys source compresses (bzip2 -9) to 20M, smaller than Linux 2.4 was. Compared to 2.6, it includes most of the same drivers, the same functionality (plus good security), the ability to run Linux binaries natively (and FreeBSD, SVR4, and some others I forgot), a network stack known to be better than Linux', and oh so much more. This source INCLUDES the ports of NetBSD for which Linux needs EXTERNAL patch sets to run on, meaning that this source tree is even more portable. We all know it's more stable, too. Where is the gargantuan (~175%) size of Linux going? It's all pure bloat. And I challenge even one person to come up with something Linux does that NetBSD can't do, and that takes up 15 meg when bzip2-9'd.
Sam ty sig.
Actually, before some clueless newbie posts saying it's all file systems, take a look at this:
/usr/src/linux
thor linux # tar -c fs | bzip2 -9 > fs.tar.bz2
thor linux # du -h fs.tar.bz2
2.8M fs.tar.bz2
...) they have in common. Sorry folks, file systems doesn't answer for it.
thor linux # pwd
Not to mention NetBSD supports a few file systems Linux doesn't in-tree (LFS, portalfs, nullfs...), and that many (NFSv3, UFS [NetBSD supports more of it anyway], ext2fs, SMBFS,
Sam ty sig.
RC's aren't a huge deal (in my opinion anyways), compared to -RELEASES. I'm sure the 2.0-RELEASE will be on the frontpage of slashdot for surely!
OpenBSD 3.6 src is only 15MB: :)
-rw-r--r-- 1 root root 15656593 Sep 20 19:09 sys.tar.gz
And that's just a plain tar.gz, not a bz2...
It has less features than Linux, but it's got all the ones I need. I even run it on my desktop machine.
click here [planetquake.com] to see the ideal case mod for a *BSD system!
Exactly, see? Good systems don't need code bloat. OpenBSD is yet another system which does what it does very well and obviously efficiently.
Linux doesn't have security, in 2.6 it doesn't even have stability, and performance seems limited to synthetic cases. Why it gets all the attention is beyond me.
BSD: The cathedral versus the bizarre.
Sam ty sig.
Excuse me? While Linux is probably almost triple the size, it supports more architectures than NetBSD, is more scalable than FreeBSD, supports close to as much hardware out of the box as windows, has numerous journalling filesystems (of which the free BSDs have none).
You're right. I should join in and help make Linux a bigger, more 'forward' project.
/usr/src/linux
...and then email the diff to Linus who'll happilly include it. Seems that's how the coding is done, anyway.
cd
dd if=/dev/urandom of=l33t-new-functionality.c bs=1k count=10k
Face it man, getting bigger doesn't mean getting better. It still suffers from very significant issues in stability, networking capability, security, and root@life knows how dirty it is in every way, even dmesg. That's what you get when you hand a tosspot around to thousands of developers.
Sam ty sig.
It does not support more architectures in-tree. You need external patches for a great many of them. A person with a clue would know this. NetBSD supports them all in-tree and with mutual inclusion.
LFS is a journalling file system, which NetBSD has. A person with a clue would know this. If you looked at my self-reply post, I pointed out that the difference the Linux file systems makes is roughly "jack shit".
There are also few drivers Linux has the BSDs don't. These are mostly only the 'evil' new cards made by NVidia and co, for which you often need external Linux drivers anyway. What's your point? None of them take up 15 meg bzip2'd anyway.
More scalable doesn't require larger code, if done right. If you knew ANYTHING about programming you'd understand this. You're coming off as a clueless troll and I can only conclude this is in fact what you are. Go read a book before you try to write one.
Sam ty sig.
It is worth remembering that recent scalability test shown NetBSD the second most scalable OS.
After Linux kernel 2.6.
Now, notice that NetBSD beat both FreeBSD and OpenBSD.
OpenBSD came last on the test, and did rather poorly.
I like OpenBSD, a nice OS but NetBSD is more polished, faster and more scalabel as an OS.
NetBSD is also very secure.
Furthermore, NetBSD holds the TCP/IP record for amount of data transferred...
AND
it was the fastest in a PPPoE test between
Free/Open/Linux24
MIT uses it and NASA uses it, and VUW in Wellington, New Zealand.
NetBSD is the finest OS I have ever used.
Oh, and LFS isn't a journalling filesystem. It's a log structured filesystem. The idea of LFS is that nothing is deleted (until you run out of space). When you save a file, it writes the changes at the end of the disk. This means that nothing you do will corrupt the disk in the event of a power failure (as with journalling, the only thing you'll lose is the last update), and that you can easily undo changes to a file (since the original is never overwritten). Additionally, the write performance of LFS is close to the theoretical maximum of the disk (since the head never has to seek for a write). The disadvantages are that it requires a lot more disk space, and disk reads of frequently modified files are slow. Additionally, when the disk becomes nearly full it is necessary to compact the log (i.e. erase bytes that have been overwritten). Log structured filesystems were a hot research area about a decade ago, but the disk space requirements were such that they were not considered feasible for general use. That will probably change in the near future due to larger disk sizes (except for Linux users, who won't be happy unless they have 500 text editors installed...). Tanenbaum proposed an interesting variant of a log structured filesystem that was implemented in Amoeba. In this version, there was no modify operation. Files were created and written to, then became read-only. Modifying a file was a matter of reading the entire file, modifying it in-core, and then writing the new version. This gave similar write-performance to a log structured FS, but also gave very good read performance (since files were always contiguous). While this design is not feasible for everything (very large files, frequently updated files), it would be ideal for many things.
I am TheRaven on Soylent News
Well log-structuring is said to be the 'more complete journalling' so I didn't figure the distinction would matter as much. Thanks for the extensive description though, that really depicts the differences. Even the online documentation on NetBSD's site didn't go into that much detail.
We need more people like you and less like great grandparent (fellow talking about Linux being the only one with journalling file systems and hence being so bloated, etc).
Sam ty sig.
Beautiful. The first 100% troll post I've seen in ages. Usually trolls have had at least one thing that was partially true, but not this one. Pure shit :)
A work of art, even. Thankyou sir, you give us something to mark other stupidities by.
Sam ty sig.
Hey, you. You're a NetBSD advocate.
Make them release 2.0 already. I bought CD-R's specifically for this and I'll be damned if I don't get a chance to use them soon!
-If God wanted people to be better than me, he would have made them that way.
I don't have that kind of power, nor would I do it if I could. You can install a 2.0RC5 system with a couple of floppies and FTP, or failing that, an unofficial ISO. You can then track (via cvs) netbsd-2-0 past its release and on to -stable. This will work very well for you and not require an official release, which, at this stage, could mean a flaky system for some other users (since there are still issues holding it back).
A lot happened to make 2.0, and given it's edging NetBSD out of 'old slow deprecated system' into 'holy sh*t-fast modern system', getting the first release Right is well worth the wait. The functional difference is about as much as FreeBSD 4 into FreeBSD 5 (sans some things like Project Evil), but with a performance gain instead of performance loss. You really have to try it to believe it.
Sam ty sig.
You're right. I should join in and help make Linux a bigger, more 'forward' project.
/usr/src/linux ...and then email the diff to Linus who'll happilly include it. Seems that's how the coding is done, anyway.
cd
dd if=/dev/urandom of=l33t-new-functionality.c bs=1k count=10k
Yeah OK man. As I said, you keep telling yourself that.
It does not support more architectures in-tree. You need external patches for a great many of them. A person with a clue would know this. NetBSD supports them all in-tree and with mutual inclusion.
Really?
Linux, 19:
alpha, arm, cris, ia64, ppc, sh, sparc64, x86_64, h8300, m32r, mips, ppc64, sh64, i386, m68k, parisc, s390, sparc, v850
NetBSD, 13:
alpha, arm, parisc, i386, m68k, mips, ns32k, ppc, sh, sparc, sparc64, vax, x86-64
What's more, Linux supports more _relevant_ CPUs. For example, all of 3 people care about running NetBSD on VAX, but they don't support ia64 or ppc64, which are the two fastest general purpose processors in existance today. And NetBSD doesn't support systems without MMUs - excluding it from a large and growing class of embedded CPUs.
Newsflash for you: system platforms != CPU architectures. Linux doesn't count the platforms it supports, they are far less interesting than supporting a new CPU - usually consisting of nothing more than some chipset and bus drivers. But if you want to count them up yourself, you'll probably find something comparable to netbsd.
LFS is a journalling file system, which NetBSD has. A person with a clue would know this. If you looked at my self-reply post, I pointed out that the difference the Linux file systems makes is roughly "jack shit".
LFS, isn't a journalling filesystem as someone alraedy pointed out... funny you should mention the phrase "jack shit"...
There are also few drivers Linux has the BSDs don't. These are mostly only the 'evil' new cards made by NVidia and co, for which you often need external Linux drivers anyway. What's your point? None of them take up 15 meg bzip2'd anyway.
Whatever. The core linux kernel code without drivers or architecture specific code, including the ipv4 stack takes up about 5 megs uncompressed. Which would be less than 500K in bzip2 format.
How about we use a more relevant metric. The Linux 2.6 kernel can be configured down to run a busybox in 2MB RAM, and have a few hundred K leftover. With network stack.
More scalable doesn't require larger code, if done right. If you knew ANYTHING about programming you'd understand this. You're coming off as a clueless troll and I can only conclude this is in fact what you are. Go read a book before you try to write one.
It does. It requires more elaborate algorithms. It requires more sophisticated locking, use of per-cpu data and lockfree algorithms.
For example, Linux 2.4's basic scheduler was pretty simple and not very scalable (because it didn't have to be, at the time). It's 33K. Linux 2.6 uses an O(1) algorithm and per-cpu runqueues, and has so far scaled up to 512 CPUs. It is 118K of code. (Although it compiles down to about 35K of object code, 15K when SMP support is not compiled in).
And in case you have any trollish comments about the Linux 2.6 scheduler, FreeBSD reimplemented it for their next gen SCHED_ULE (which they can't get right, but that's another story).
However, Linux actually often has a smaller footprint where it does count in terms of efficiency and scalability, though. For example, on i386, NetBSD requires 64 bytes for every page of physical memory in the system, while Linux requires only 32 bytes. The difference gets much larger on 64-bit architectures.
Yep.
FreeBSD, however includes soft updates which solve the same problem as journaling but without the performance hit of journaling and with fewer buzzwords.
Actually, FreeBSD's softupdates are not nearly so good as journalling. For a couple of simple reasons.
First of all, the concept is basically the same as journalling - write out metadata in the correct order, to give a consistent disk image.
However, with journalling, one can write out large swaths of metadata information in linear writes to the journal, and later lazily write them out in an optimised order to the actual metadata blocks. With softupdates, all the little synchronous metadata writes go to different areas of the disk. Performance often isn't as good as with journalling, despite what zealots will say.
Second, with journalling, a crash just means a journal replay or rollback. Softupdates can still get into an inconsistent state on crash - it is just structured so those states are recoverable, but still take long, performance sapping background fscks, which are often not acceptable in high availability situations.
Third, softupdates has no possibility to give fully ACID semantics at the filesystem level, like journalling theoretically can (see for example, Reiser4).
FreeBSD also supports background fsck for all except the root partition (which is usually
Heh, this isn't a feature. See above: journalling filesystems don't *need* this hack.
FreeBSD also supports one particularly useful feature which, I believe, Linux only supports for XFS - filesystem snapshots. You can mount a snapshot of a mounted filesystem, and continue to access it. This allows you to back up the filesystem as it was at a fixed time point, without having to worry about writes during the backup, and without having to unmount the drive.
Err, cool... so Linux supports snapshots. That's great.
I'm a NetBSD user but don't follow all the news about latest developments at this level. In a nutshell, what's responsible for such a large performance boost? The biggest new feature I'm looking forward to in 2.0 is native threading. I'm going to recompile a bunch of my apps to take advantage of it.
Constitutionally Correct
Native threading is one very big thing (surprisingly big thing even), but generally they seem to have optimized everything and it all adds up to blinding performance, especially in how well it uses hardware. You won't see any hardware NOT doing the best it possibly can, is what I mean to say. And the hardware support itself is what I would call "clean like Linux, but with sensible messages". Hotplugging and everything is kernel-level unlike FreeBSD, which I rather like.
More interesting changes will come in AFTER 2.0 is released, since some new stuff is happening in -current that has to wait until after release. On this topic, the 2.0 release HAS BEEN TAGGED, so it's a matter of days before the release is official! The wait is over!
The power to serve - on the platform of your choice.
Sam ty sig.
Ahem, is this thing on? Excuse me, why hasn't anyone replied? Oh it is on... so that means I have TAMED THE RABID BSD ZEALOTS! Bwaahahahaaa!!
Now in all seriousness - I forgot one other thing which makes journalling really nice compared to softupdates.
You can journal to a really fast non volatile ram drive. This can give you huge speed boosts, especially on metadata intensive operations, but still with all the protection of a journalling filesystem.
No way in hell softupdates can do anything of the sort.
"We all know it's more stable, too."
Really? My Debian and Slackware boxes never, EVER crash. They're rock solid. Perhaps comparing NetBSD to, say, Mandrake -- you might be right. But there's barely any difference in terms of reliability with proper, tested distros.
Oh, and here's one thing Linux can do: be supported for years. Certain distros such as RHEL have a whopping FIVE YEARS of security and bugfixes -- a great help on production servers when you don't want to upgrade every 12 months or so.
Otherwise, I mostly agree with what you say. And I'm certainly looking forward to NetBSD 2.0.
Oh, and binary updates. I can't stress this enough. OpenBSD could be the most secure system on the planet, but when I have to:
- read advisory and download patch
- apply patch by hand
- have a compiler toolchain installed
- rebuild affected files
- copy them into place in the filesystem
As opposed to:
- apt-get update
It's a royal pain. One command vs serveral steps (and having a compiler and toolchain on a server is unwise anyway). Give the BSDs a clean, fast and simple binary update system and they'll be so much more pleasant to administer.
All this attacking linux is probably making you guys thirsty, you don't have an inferiority complex do you?
I wish people would stop whinging and just accept that different things a better for different things.
i.e. there are cases where netbsd, openbsd, freebsd or various versions of linux excel at what they are designed to do.
For example; out of the above, linux is "best" on the desktop. From speed of setup, hardware detection, open and closed drivers etc. If you want to run a desktop with some games, 3D acceleration, flash, realplayer etc. And you don't want to spend an age setting it up. Things like fedora core 3 leave the others in the dust. I can have a fedora core box set up in about 20 minutes with everything working for the above tasks. It simply leaves your command line installs and hardware problem OSs in the dust. That is it's PURPOSE.
BUT, for a networking purpose, security needed, embedded/low disk space etc. it is likely that netbsd will absolutely thrash fedora core.
It's possible that they can do each other's jobs, it's just they have emphases. Which makes me think the guy running fedora on his desktop and running netbsd on his special server is the true techy. And the guy running fedora on both or netbsd on both is the zealot.
I know a number of phds in comp sci who use fedora because they are sick of messing with command line stuff they don't need to (even though that CAN, they dont NEED to). That is a massive feature which cannot be priced in your "eliteness" worlds.
It's gentoo syndrome all over again. 1% speed boost.....AFTER hours and hours of tweaking which cancel out the speed boost by many orders of magnitude. Gentoo is a platform to LEARN on, not to use as a desktop.
Everything has it's place and it's purpose. Anyone who claims they have a one-size-fits-all is peddling snake oil.
So netbsd is very impressive for what it specialises in. But SHUT THE FUCK UP to all these trollish wankers who constantly claim their OSs are appointed by God. It goes for linux zealots to BSD zealots.
The BSD guys always claim to be more mature, but all I see of them is constant jealousy. And the linux tardates are always stirring shit by "bsd is dying" it.
All of you just shut the fuck up you fucking nerds. Wipe the orange cheeto dust off your fingers and just discuss the pros of netbsd and what is good about it.
You are all falling into troll traps.