Slashdot Mirror

http://lwn.net/Articles/351058/

Basically, he can not see any BFS performance improvements, on this box (Dual Quad-core).

Since it may not be obious to everyone what hugepages are, here's a link that may work out for you:

http://lwn.net/Articles/188056/

For personal reasons, I do not browse the web from my computer. (I also have not net connection much of the time.) To look at page I send mail to a demon which runs wget and mails the page back to me. It is very efficient use of my time, but it is slow in real time.

RMS, 2007

Why bother with useless junk like browsers?

Last numbers I saw said there were 78 people actively working on the Linux kernel, and not all of these full-time

Where did you get that? LWN's development statistics for 2.6.31 (subscriber-only for the next few days) say that there were 1,146 distinct developers whose patches got accepted into this particular minor release (i.e., over the last three months or so). The stats for 2.6.30 (publicly viewable) show 1,125. Granted, most of these are probably touching only a tiny portion of the code and might only get a few changesets accepted, but they could still fairly be described as "actively working on the Linux kernel" (even if far from full-time). Also, there are an untold number of people who "actively work on the Linux kernel" but don't submit their patches, or submit them but don't get them accepted.

So maybe there are 78 people working on the Linux kernel according to some very specific, narrow definition, but you should really provide the definition if you're going to throw around such small numbers.

For what it's worth, I doubt that either Apple of Microsoft devotes anywhere near the man-hours to their kernels as get devoted to the Linux kernel. But a lot of the extra man-hours that Linux get contribute to extreme specialization, like dozens of supported CPU architectures (Windows: 3, Mac: 1), dozens of supported filesystems (Windows, Mac: But I'm not a kernel hacker or anything close to one, so take this post with a grain of salt.

Last numbers I saw said there were 78 people actively working on the Linux kernel, and not all of these full-time

Where did you get that? LWN's development statistics for 2.6.31 (subscriber-only for the next few days) say that there were 1,146 distinct developers whose patches got accepted into this particular minor release (i.e., over the last three months or so). The stats for 2.6.30 (publicly viewable) show 1,125. Granted, most of these are probably touching only a tiny portion of the code and might only get a few changesets accepted, but they could still fairly be described as "actively working on the Linux kernel" (even if far from full-time). Also, there are an untold number of people who "actively work on the Linux kernel" but don't submit their patches, or submit them but don't get them accepted.

So maybe there are 78 people working on the Linux kernel according to some very specific, narrow definition, but you should really provide the definition if you're going to throw around such small numbers.

For what it's worth, I doubt that either Apple of Microsoft devotes anywhere near the man-hours to their kernels as get devoted to the Linux kernel. But a lot of the extra man-hours that Linux get contribute to extreme specialization, like dozens of supported CPU architectures (Windows: 3, Mac: 1), dozens of supported filesystems (Windows, Mac: But I'm not a kernel hacker or anything close to one, so take this post with a grain of salt.

The tickless/dynticks Linux kernel patches first appeared in 2.6.21 for the x86 platform. amd64 tickless patches appeared a bit later in 2.6.24. It isn't just a matter of changing the HZ either - it looks at what timers are set to go off and idles until the soonest timer is set to go off or an interrupt arrives rather than firing wakeup events and then realising there is nothing to do (so it's not just switching between fixed frequencies).

There may have been issues using tickless kernels with virtualisation solutions but that might have been when using tickless kernels as hosts rather than guests (it's not something I've looked at) and in all likelihood the problem has been solved by now.

Sad to say, it isn't just Microsoft that photoshops out the blacks from their promotional material. See here a kernel hacker group photograph, and every single person of African decent, replaced by white people.

Outrageous I say!

I run the results of wget through a custom Perl script and then parse the results and feed image URLs back through wget and into libjpeg. Why do I need a bloated web browser when I have such an elegant Unix solution?

Ah ha! I always suspected that RMS read Slashdot. Now my suspicions are confirmed. :-)

Just read LWN, Corbet has been writing these for a couple of years already.

See http://lwn.net/Kernel/Index/ and search for "Contributor statistics"

Just read LWN, Corbet has been writing these for a couple of years already.

See http://lwn.net/Kernel/Index/ and search for "Contributor statistics"

Valerie Aurora wrote an interesting article about the paper at lwn: http://lwn.net/Articles/270081/

As you suggest, you have to provide the source even if you have not made any modifications to the GPL code if you are doing commercial distribution - it is not enough to point to an upstream source.

This issue pretty much happened a few years ago. The MEPIS Linux distribution was "forced" to comply with the GPL by distributing the source code itself rather pointing to an upstream distribution (found via linux.org).

He did indeed claim this

Just having mmap_min_addr and setting it to a page or more above 0 isn't good enough. It also depends very much on the exact kernel version you're running. 2.6.30.2 had a problem with both SELinux and personalities making it possible to get around this. 2.6.30.3 fixed both I believe.

See http://lwn.net/Articles/342420/ for more about which versions are vulnerable and why (and, yes, I'm the same Athanasius linked to in the "This change is not enough for some users, who have requested the ability to turn off the personality feature altogether. " bit, if I could get my arse more in gear I'd have coded up a sysctl/personality patch by now).

It turns out that it's not simple to map NULL, but unfortunately (from a security standpoint) not impossible. It's normally disallowed, but since there are programs that sometimes need it (namely domemu-style programs). See this LWN article describing the previous NULL-dereference exploit. Presumably the seriousness of this NULL pointer exploit inspired others to look for exploitable NULL pointer bugs as well.

SELinux might actually hurt in this situation, since this is a NULL pointer exploit. The LWN article Fun with NULL pointers from a few weeks back has an interesting explanation.

You can find less eccentric (and less easy targets) than RMS among those who support the GPL.

Yes -- RMS is quite the eccentric, and pretty much the antithesis of what you'd want to have in a spokesman.

Part of the problem is that he insists on taking "ownership" of the GPL, and frequently acts as though he's a spokesperson for the entire open source community. The GNU/Linux naming "war" that he's been waging for some time now is outright embarrassing for all parties involved.

He's also responsible for turning GNU from an open-source software collective into a pseudo political advocacy group. GNU's philosophy page reads like some sort of paranoid rant -- It's virtually impossible to take it seriously.

Perhaps the smoking gun is this mailing list post RMS made two years ago, admitting that he hasn't yet embraced hypertext. He browses the web through a HTTP-Mail gateway, and strips out the HTML. (On the flipside, GNU appears to have finally hired a competent web designer. Stephen Fry on the homepage is a rather nice touch)

Torvalds is a much better role model to follow. He keeps quiet, and insists that Linux is a community effort, of which he is only a small part.

'The proponents of GPL like to tell people that the world only needs one open source license, and I think that's actually, frankly, just a flat-out dumb position,'

Yeah, well I think that's actually, frankly, just a flat-out fabrication. Could we have a source for this assertion please?

I think it's actually the (a?) purpose of the "additional permissions" language, to make GPLv3 flexible enough for anyone to use.

Anything beyong one simple license that we can clearly explain the use and restrictions around open source software fails the future use and growth of the adoption of such software.

I've also seen calls to have only 4 licenses (BSD, LGPL, GPL, AGPL).

Wouldn't the drive benefit from a real understanding of the filesystem for this sort of thing?

There is no need as a standard ATA TRIM command exists by which the OS can tell the device when a block is no longer in use. LWN wrote about this almost a year ago.

The KDE team has been code-naming its releases with words starting with C for at least a little while, and this isn't the first time they forced a word's spelling to fit. For example: KDE 4.2.3 was called "Cuagmire" instead of Quagmire.

It's strange, but that's how they roll, apparently.

http://lwn.net/Articles/144681/

Linux has had kernel level support for TPM for a while but most F/OSS developers have an intrinsic aversion to the concept (as I said in the GP, the identity of the TPM principals doesn't exactly give me a lot of confidence) so it's not widely used as far as I can tell.

A wonderful response from the F/OSS community would be to build a version of TrueCrypt that uses TPM to authenticate the BIOS and MBR against the known good versions.

A licence is a contract. It's a permit with clauses that must be accepted before the permission is given, which enters the parties into a contractual obligation.

I'm sorry to have to tell you, but you are wrong. http://lwn.net/Articles/61292/

Don't tell us you're using LVM for critical data such as backups.

LVM does not implement file system barriers.

Betcha didn't know that, did you? Scary, ain't it?

Heck, might as well google for why barriers aren't enabled by default in Linux....

You seem to be new to linux so I'll explain it. The $80 is a subscription to the Red Hat Network support service, it is not a purchase price for Red Hat Enterprise Linux.

In fact there have been multiple distribution forks/clones created from the Red Hat Enterprise Linux source files. You download the source files, remove the Red Hat trade marks and logos, compile and distribute. :)

So with that bit of knowledge under your belt you can now intelligently discuss Red Hat's position in the F/OSS community and join the rest of us in celebrating this acknowledgement of Red Hat's business success based on open source.

If it weren't for Mozilla's anti-patent zealotry and their wide market share, HTML 5 would have had a standardized video codec two years ago. It's not as if they can't afford licensing fees at this point.

When compiling a kernel driver for BSD or Windows, you will never, ever, ever see anything even remotely like "FATAL: modpost: GPL-incompatible module foo.ko uses GPL-only symbol 'usb_register_dev'".

There are countless other examples of the GPL stifling innovation, and it's the end-user that really loses out in the end. This would be perfectly fine if GPL advocates weren't influencing policy-makers, or using their market share in part to push their crusade.

Software raid on Linux?

If you care about your data, don't do that. LVM does not respect file system barriers!:

There is another hitch: the ext3 and ext4 filesystems, by default, do not use barriers. The option is there, but, unless the administrator has explicitly requested the use of barriers, these filesystems operate without them - though some distributions (notably SUSE) change that default. Eric Sandeen recently decided that this was not the best situation, so he submitted a patch changing the default for ext3 and ext4. That's when the discussion started.

Andrew Morton's response tells a lot about why this default is set the way it is:

Last time this came up lots of workloads slowed down by 30% so I dropped the patches in horror. I just don't think we can quietly go and slow everyone's machines down by this much...

There are no happy solutions here, and I'm inclined to let this dog remain asleep and continue to leave it up to distributors to decide what their default should be.

So barriers are disabled by default because they have a serious impact on performance. And, beyond that, the fact is that people get away with running their filesystems without using barriers. Reports of ext3 filesystem corruption are few and far between. ...

Anybody can turn on barriers if they are willing to take the performance hit. Unless, of course, their filesystem is based on an LVM volume (as certain distributions do by default); it turns out that the device mapper code does not pass through or honor barriers.

... send mail to a demon which runs wget and mails the page back.

1995 called. It wants its flamewar back.

But seriously. In another thread earlier today, I came across this gem from RMS in 2007. Even more hilarious is the response to his message with an "ASCII Ribbon Campaign" banner at the bottom of his message. Hadn't seen one of those in nearly a decade.

http://lwn.net/Articles/262570/

It sounds like he doesn't use a web browser in exactly the same way that Knuth does not use an email client. He finds the usual http interface of a web browser cumbersome, and prefers his more old-fashioned tools, much in the way Knuth prefers his old-fashioned paper.

Stallman also says no to web browsing.

There is (was?) a problem with Intel floating point precision. The register internal precision is higher than that specified by the IEEE standard, but when a value is stored it's precision is truncated to meet the standard.
-This might not seem like a problem at first glance, but this can happen unintentionally if your code is interrupted by kernel code or HW interrupt.
Therefore you can get two slightly different results depending on whether you were interrupted of not. This is unacceptable.

The Fortran libraries, at a significant overhead, keep storing floats into memory to get consistent repeatable results while the C code did not.

Fortran still came up ahead.

If you want to read more about stuff you have to take into account when optimizing matrix operations you can take a look at Ulrich Drepper's long article at lwn.net
http://lwn.net/Articles/250967/

You must be kidding right? There's tremendous code sharing and abstraction done in the kernel. Hardware specifics are placed in their own files but the rest is shared. Most hardware drivers work on all architectures and share common code. Here's one example, of many, where code was consolidated. i386 and x86_64 merge Code sharing works because of well thought out abstractions.

Integrity Management Architecture

Contributor: IBM

Recommended LWN article: http://lwn.net/Articles/227937/

The Trusted Computing Group(TCG) runtime Integrity Measurement Architecture(IMA) maintains a list of hash values of executables and other sensitive system files, as they are read or executed. If an attacker manages to change the contents of an important system file being measured, we can tell. If your system has a TPM chip, then IMA also maintains an aggregate integrity value over this list inside the TPM hardware, so that the TPM can prove to a third party whether or not critical system files have been modified.

From the recommended article, the key dilemma:

There are clear advantages to a structure like this. A Linux-based teller machine, say, or a voting machine could ensure that it has not been compromised and prove its integrity to the network. Administrators in charge of web servers can use the integrity code in similar ways. In general, integrity management can be a powerful tool for people who want to be sure that the systems they own (or manage) have not be reconfigured into spam servers when they weren't looking.

The other side of this coin is that integrity management can be a powerful tool for those who wish to maintain control over systems they do not own. Should it be merged, the kernel will come with the tools needed to create a locked-down system out of the box. As these modules get closer to mainline confusion, we may begin to see more people getting worried about them. Quite a few kernel developers may oppose license terms intended to prevent "tivoization," but that doesn't mean they want to actively support that sort of use of their software. Certainly it would be harder to argue against the shipping of locked-down, Linux-based gadgets when the kernel, itself, provides the lockdown tools.

OK, maybe this is overdramatic, but trading freedom from third-party oversight through trusted computing for the security of first-party oversight through trusted computing seems a little like:

"They that can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." - Benjamin Franklin

But I can see both sides. Pondering... what are your thoughts?

Several approaches are being investigated; see
http://code.google.com/p/chromium/wiki/LinuxSandboxing
http://lwn.net/Articles/332974/
http://www.imperialviolet.org/2008/11/27/sandboxing-on-linux.html

I'm pretty sure that linux isn't designed to be fully audited either though. In fact, last I read the kernel rejected patches to allow it to be (on the grounds that it would be abused by DRM advocates).

http://lwn.net/Articles/333825/

One catch in performance is that it sure is faster to use RAM for data, but there is also a lot of useless data floating around in RAM, which is a waste of resources.

RAM is cheap these days.

I thought we were talking about performance? Adding more memory for a slow app does not necessarily make it faster when we're involved with parallel architectures. Maybe that ought to be a law of its own. See http://lwn.net/Articles/250967/

All memory is not created equal. As CPU speeds have risen, it is becoming increasingly expensive to access main memory - you need to keep things in the CPU cache(s). As NUMA architectures increase in importance, that will become even more true. Today's supercomputer is tomorrow's high end server is next week's common desktop box.
Anyway, Ulrich Drepper put it better than I can in the link above.

Storage devices are still slow and the most interesting ones have a finite (Though still large) number of writes.

That part is true, however absolute read/write speed may not necessarily have anything to do with absolute performance. The basic idea is that you need to move less rarely accessed data to more expensive (in terms of access time) storage and keep the most actively accessed data in the fastest cache/(on|off node)memory as you can.

Each guest needs its own kernel, so you need to allocate memory and disk space for all these kernels that are in fact identical

I'm pretty sure VMWare can detect when the same block of the same file is mapped into multiple guests, and share the physical page. Plus, the kernel's memory image is small compared to, say, the database server you're running on it. I guess there's overhead like an extra set of page tables (either nested page tables managed by the guest, or shadow page tables managed by the host). Overall a small effect I think.

TLB flushes kill performance. Recent x86 CPUs address the problem to some degree, but it's still a problem.

Any context switch between two userspace programs in a non-virtualized system needs a TLB flush too (BSD jails included). Or, if you're using a processor that has a tagged TLB, you don't need to flush it, but your virtualized guest gets the no-TLB-flush benefit too.

A guest's filesystem is on a virtual block device, so it's hard to get at it without running some kind of fileserver on the guest

Again I don't think this is a huge deal. Aren't there drivers to allow a host to see inside a guest's block device and/or filesystem?

Memory management is an absolute clusterfuck.

In a naive hypervisor, yes. In more mature hypervisors, not really. See the following articles for solutions on fully virtualized and paravirtualized guests, respectively: http://www.usenix.org/events/osdi02/tech/waldspurger/waldspurger_html/node6.html
http://lwn.net/Articles/198380/

It used to be a lot easier to be a good assembly programmer. Just consider the implications of modern memory models. Just consider that Ulrich Drepper had to write a seven-part series just on memory.

(It's like NFS getting a bad reputation--especially locking--because Linux's implementation sucked for so many years.)

Huh?

fcntl/lockf works fine over NFS so long as both sides are either using NFS4 or both sides are running NLM (the NFS locking daemon -- i.e. rpc.statd)

Dot-locking also works over NFS, though open(..., O_EXCL|O_CREAT) does not work, nor does flock. Have these worked on other systems, or started working under Linux?

Its been tried:

UserLinux, United Linux....

... because apps that work fine on every other filesystem should not need to be re-written specifically for Ext4

Not quite. I believe XFS and JFS behave the same way as Ext4. Here's a good article and thread on the subject. http://lwn.net/Articles/322823/

I nearly lost my whole filesystem. It's a good thing I had a backup core system on reiserfs to boot from and run fsck. from what I understand, it's a problem with the ext4 journaling system and metadata. this link has info on the journal problem...which may have already been patched in the current kernels. http://lwn.net/Articles/284037/ wiki page for ext4 - bottom has a fix for the problem: http://wiki.archlinux.org/index.php/Ext4 essentially, mounting and ext4 filesystem with option "data=ordered" helped my system out. since I have enabled this mount option, my filesystem is now stable even after hard reboots or power failures. Hope this helps out people as it did me! -Kamphor

Linux generally gets faster over time. Beyond that, you can customize your OS and make it really fast. Probably similar things can be said with the *BSDs. Open Source OS's are flexible, if you need to you can do all sorts of task-specific tuning, and the whole package is extremely light on resources. The latest version of Fedora has similar system requirements to XP while providing a much more responsive user experience, and minimalist distros like PuppyLinux will be like screaming lightning on modern hardware.

Microsoft is the second largest software company in the world (depending on how you count things, IBM is bigger). Why can't they do better than linux? They spent six billion on Vista, why does it perform more poorly than XP? How does anyone find that acceptable?

Indeed. But Microsoft has a long way to go to catch up to linux in boot time.

I will prove that anything written in a higher-level language will not be as fast as my implementation of it in C. I leave this challenge out to anyone to take. (*)

Seriously, I'm sick of this crap. Bring it on.

That's what programmers said 50 years ago s/C/Assembly/. IBM proved them wrong with a FORTRAN compiler.

I think I could do the same thing you described[1]. I have every doubt that the folks sitting around me in adjacent cubicles could do so. That's kind of the point.

[1] Things aren't so simple now with the way CPU manufacturers are designing chips - http://lwn.net/Articles/250967/

I used the alpha. I am about to install it to my netbook's hard drive; hopefully it provides a way to upgrade. I would presume the answer is yes, but it's pretty rough.

I kind of like XFCE, though. I hope this UI can be disabled easily. The second important note is whether (since this is Fedora-based) the Fedora repos can be enabled without making the computer do bad crashy things.

The other point that the summary neglected to mention is that this project is the first real implementation of Arjan Van De Ven's work on fast booting. He's the guy that made his Eee boot in five seconds. Moblin can be expected to boot fast, which I think is necessary if we're going to recategorize netbooks from 'underpowered miniature laptop' to 'powerful internet appliance'.

A friend of mine bought a macbook a few months ago because she needed a computer that was extremely simple and user-friendly. Macs are somewhat better in that regard than PCs, but the computer is still pretty incomprehensible to her. This new UI is probably not for the slashdot crowd (anyone who can tell you why it's important that Moblin has a (relatively) standard Xorg server is not really the intended audience), but I think for the common Joe or Jocelyn it's perfect. Especially if you were ever wanting to make a $100 netbook...if the price point is sufficiently low to the point where it's clear you're not buying a Real Computer, and stick this UI on it, you can both set new expectations for what the device is supposed to do and sell a lot of toys to people that don't really have any use for a Real Computer.

Random points:
Webcam support is essential.
The social networking pane needs to be Facebook, not Twitter.
um, End of Line?

But why are Linux enthusiasts hoping for a future of Linux on the Desktop (TM)?

Because it would solve the russian guy's points 5 "Problems stemming from low linux popularity and open source nature: ", 5.1 through to 5.3 and 5.3.2. Read up on network effects, monopoly effects, and this LWN article from Januari 2007:
http://lwn.net/Articles/219791/

Personally (but I have stupid ideas often, and IANAE), I think it's a matter of Hysteresis, and Linux adoption is not yet catapulted to the steep bit of the bottom slope.

And if software patents are finally abolished everywhere (USA I'm looking at you!), also the illegality of points 5.4 and 5.5.

BTW, point 5.4 should read: "5.4 It's illegal to watch Blue-Ray movies. ", and whose fault is that?

I don't think solving 5.3.1 serves any purpose. Why should I buy a printer that doesn't work with my OS?

Device support is offered, since Januari 2007: http://lwn.net/Articles/219791/ but apparently the offer is not being taken up. Write to your music hardware manufacturer sending them this link :-)

captcha: trapped

I've found several articles about the Extremadura LinEx project but I have been unable to find any information about exactly which schools were migrated.
One article (http://lwn.net/Articles/41738/) from 2003 states one computer for every two children were installed so I would say a substantial amount of all schools were running LinEx by June 2003.
An other article (http://www.osnews.com/story/12611) from November 2005 states that 66000 computers in schools and education centers and an additional 14000 in public administration buildings.
According to a third article (posted August 2006) on debian.org, all schools were migrated to Linux during 2004 ((Article in Swedish) http://www.debian.org/News/2006/20060803.sv.html)

In goverment...
* 1000+ in French parliament : http://blogs.computerworld.com/node/4060
* 11000 at German Foreign ministry.
* 14000 in Munich.
* 13000 at The Federal Employment Office of Germany
* 80000+ in Spain 2003: http://lwn.net/Articles/41738/
* 90000 at France's national police force in 2007

In education...
* "Germany has announced that 560,000 students in 33 universities will migrate to Linux."
* "Russia announced in October 2007 that all its school computers will run on Linux."
* "9,000 computers to be converted to Linux and OpenOffice.org in school district Geneva, Switzerland by September 2008"

In business...
* "Peugeot, the European car maker, announced plans to deploy up to 20,000 copies of Novell's Linux desktop."

Read more about adoption of Linux at Wikipedia: http://en.wikipedia.org/wiki/Linux_adoption