Linux v2.6 Begins Testing

xose quotes Linus from the kernel list: "the naming should be familiar - it's the same deal as with 2.4.0. One difference is that while 2.4.0 took about 7 months from the pre1 to the final release, I hope (and believe) that we have fewer issues facing us in the current 2.6.0. But very obviously there are going to be a few test-releases before the real thing. The point of the test versions is to make more people realize that they need testing and get some straggling developers realizing that it's too late to worry about the next big feature. I'm hoping that Linux vendors will start offering the test kernels as installation alternatives, and do things like make upgrade internal machines, so that when the real 2.6.0 does happen, we're all set." You all know what to do ;) Update: 07/14 17:49 GMT by S : OverNeith writes "Joe Pranevich has done it again! He's written another summary document on what to expect in the new and upcoming 2.6 Kernel!"

Comments

Clear this up for me

[Neophytus]

What is it I have to do? Send loves and kisses?

Re:Clear this up for me

[Surak]

Flowers. Don't forget the flowers. And PLEASE remember to pickup a gallon of milk and a loaf of bread on the way home, okay, honey?

Sorry

[keesh]

There isn't an ebuild yet, and I'm too lazy to do it the old way...

Re:Sorry

[presroi]

There isn't an ebuild yet, and I'm too lazy to do it the old way...

Just double-click on the kernel.msi button right next to the Explorer.

Re:Sorry

[Surak]

Actually, it *is*. It's in the portage tree under sys-kernel/development-sources development-sources-2.6.0_beta1.ebuild

Hmmmmm...

[Arthaed]

I know if I start downloading and compiling this kernel I am going to have a 'case of the Mondays' real quick.

Difference?

[KillerHamster]

For us newbies here, what are the relevant differences in the new kernel? Better performance? New hardware support?

Re:Difference?

[pe1rxq]

The biggest change for normal users is the preempt patch, it will make your system very responsive to interactive tasks (ie a graphical desktop) also the new schedulers should help here.

Jeroen

Re:Difference?

[Libor Vanek]

And better USB support with easier way for writing drivers for various USB gadgets.

Re:Difference?

[Wiz]

The best reference I've found is Dave Jones' website..... Linux 2.5 core updates.

Re:Difference?

[inode_buddha]

Yes, IPSEC is already in 2.5.xx, along with the NSA's SELINUX hooks, IBM's JFS filesystem, and SGI's XFS filesystem. Lots of VM and block I/O work, too.

Re:Difference?

[caluml]

Yep, it's called mkreiserfs.

Re:Difference?

[Miles]

They're using the new NTFS drivers. Check out:
http://linux-ntfs.sourceforge.net/

Re:Difference?

[Urchlay]

The new anticipatory scheduler seems to make a much bigger difference than the preempt patch did in 2.4.

My test box is a Duron 750 with 384M of RAM, running Apache 1.3, Tomcat 4.0 (with Sun 1.4 JVM), MySQL 4.0, X11 + Windowmaker, usually running Opera and Mozilla.

With 2.6.0-test1, I can run the load average up to 3.6 or so, and Mozilla is more responsive than it ever was on 2.4, even with a completely idle system. In fact, it's almost as responsive as the ancient Netscape 4.7 on this same system (compare Netscape 4.7 with any Mozilla 1.x release, if you don't know what I mean).

I'm doing all this junk at once:

- Recompiling the kernel in a `while true' loop
- Recompiling a 100,000 Java project in a `while true' loop
- Playing mp3s with mpg123
- Untarring a kernel tarball, then deleting it, in another loop
- Using Mozilla to hit locally-hosted Tomcat servlets, which make heavy use of the local MySQL server, which has pretty large tables (biggest 2 tables are 1.6G and 400M)
- Reading /. in Opera :)

I can't make the mp3s skip, and virtual desktop switching is instant. In 2.4, even with the preempt and lowlatency patches, either Mozilla or mpg123 will freeze up, and/or Tomcat/mysql will lag badly (of course, preempt/lowlat isn't supposed to help much with background server daemon processes anyway). 2.6.0-test1's performance under load also beats the 2.5.6x and 2.5.7x kernels I tried on this machine, though most of the 2.5's were an improvement over 2.4.

It helps that all this activity doesn't cause much swap usage (hovering right around 200Kb of swap used).

BTW, if you're already able to run recent 2.5 kernels, you should be able to just throw 2.6.0-test1 in and have it work (no need to upgrade anything you haven't already, to support 2.5).

Executive summary: I'm a happy camper... If you're able to do so, you should try out this kernel on a spare box & see how you like it.

I got it before the /.ing

[Anonymous Coward]

And ive just compiled it. I was quite surprised I managed to get it to boot without it panicing. I'm even typing from the new kernel now. But there is a word of warning though. The layout of the /dev folder has been rearranged. As a result some of my programs have broke.

For example. /dev/hda, /dev/hdb/, /dev/hdc now become /dev/discs/disc0, /dev/discs/disc1, /dev/discs/disc2. So you will need to edit /etc/fstab to reflect the changes.

Re:I got it before the /.ing

[caluml]

That's devfs. If you don't use that, they'll all be normal (hda, sdb, fd0, etc).
At least it wasn't mandatory as of 2.5.69 anyway.
Why isn't devfs the default now - it's been working fine for ages - for me anyway.

Re:I got it before the /.ing

[bumby]

For example. /dev/hda, /dev/hdb/, /dev/hdc now become /dev/discs/disc0, /dev/discs/disc1, /dev/discs/disc2

That is called devfs, and as far as I know is an optional thing. At least it was in 2.4-series, and I really really doubt it isn't in 2.5 and will be in 2.6. So just skipp the CONFIG_DEVFS_FS and CONFIG_DEVFS_MOUNT and use your old nodes.

Re:I got it before the /.ing

[Waffle Iron]

Devfs is an over-engineered solution to a non-existant problem..

That's exactly what I've thought for a long time now. I've come up with a much simpler solution that I call "drvlttrd". I'm going to submit the patch as soon as I do a little more cleanup. Basically, the devices get short convenient names that can be used like URL prefixes. Example:

/dev/fd0 -> A:
/dev/fd1 -> B:
/dev/hda1 -> C:
/dev/hda2 -> D:
etc...

Re:I don't know what to do - really

[pe1rxq]

You connect another computer to the serial port and use it as a console...
Or use multiple monitors, one for X, one for the console...
(with the serial solution you can automagicly log it and don't have to type anything from a screen)

Jeroen

Re:This is a big deal.

[avalys]

You don't have to wait - pretty much all of big stuff has already happened in the 2.5 series. 2.6 is the next stable series, which (usually) means no big architectural changes. What's going on now is testing to ensure that the 2.5 series is stable enough to be considered for a release as "2.6.0".

Does Linux 2.6 permit decent video capture?

[Sheetrock]

I've been tweaking my cheapish bttv-style card and no matter what I do I can't seem to prevent any of the video recording solutions for Linux to capture a steady stream of frames. The easiest test is to tune the card to a news network and watch the moving text on the bottom bar -- it skips in a regular pattern under Linux, but works smoothly under Windows (i.e., it's not a slow computer issue, although there could be magic in the driver that isn't being duplicated under Linux). From what I've read, it's an issue with Linux timing, and perhaps a real-time kernel would work better, but I was wondering if they've been addressing this.

Should mention that the sound capture seems to cause the problem -- without sound, the capture is smooth under Linux, but adding either ALSA or OSS to the mix guarantees problems.

Re:Does Linux 2.6 permit decent video capture?

[Josh Booth]

It's not the driver, it's just that Linux Kernel 2.4.x is not preemptible. In other words, the kernel system calls, like read and write, are made to be interruptable in 2.6.x. Basically, it adds some code that allows something else in the kernel to run if one thing is waiting, say, for some sort of input, which improves latency for short periods of time. Linux Journal, May 2002 edition explained this and they had a latency graph of a system playing music under a load. Without the preemptible patch, there were huge spikes in latency (basically audible as gaps in the music) but with the patch, there were minimal changes in the latency. Just today, I noticed something similar to what you talked about. I was ripping CD's and then encoding them as Ogg. Apparently, cdparanoia does system calls that have high latency and since they can't be interrupted, working in X is slow. However, since Ogg encoding is mostly userspace, X was much faster even though more of my processor was being used. Recently, Slashdot linked to an article outlining the changes in 2.6. People have also been making patches for 2.4 for a long while to improve latency, and I think there is a backport of the low latency patch to 2.4.

Compiled, tested, working.

[indigo78]

Downloaded, compiled and installed. Working since 4 hours on a Slackware-9.0-current, asus L8460K notebook (p3/1000, 256mb ram, i440bx, S3 savage/MX, ess allegro) and quite standard compilation options (acpi, alsa, pcmcia, usb, netfilter, no ipv6, preemptible kernel). Applied patch as seen on LKML (see here) for vfsmount.
Happy testing!

Re:This is a big deal.

[sfraggle]

I personally can't wait to skim the change logs.

Kernelnewbies.org has a page which usefully summarises the new stuff in 2.6.

How's the must-fix list going?

[Bollie]

Last time I looked at ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/ must-fix/ there were still some showstoppers. It seems like they were updated about a month ago, so I guess progress must have been made on them...

The biggest problem I have with the newer kernels is probably some ACPI/IRQ routing bug in my board. It's a common problem with the NForce2 chipset (APIC doesn't work, so you have to boot with pci=noacpi or acpi=off). It's not the biggest inconvenience, but it causes half of my unused USB slots not to work...

I must say the snappiness of 2.6 is great! I'm looking forward to beta-testing. AFTER I backed up my drive, of course!

SCO vs linux

[VEGx]

I hope they added a few lines like "This is not SCO code" :-P

OK, So what if I'm a troll?!

Re:This is a bad idea....

[stinky wizzleteats]

All I see is badness coming from this.

Linux is changing. The average Linux user of today doesn't recompile their kernel. What's wrong with Mandrake or Suse offering a clearly labelled "testing kernel"? One of the problems Linux development is having right now is that the testing community is so closed that they aren't getting a good cross section of production machines during testing. The end result is that the rubber doesn't really meet the road until the kernel goes "live".

BIO

[zonix]

If I remember correctly there's a new Block IO (BIO) layer included too, which should enable IDE CD burning without the need for SCSI emulation. Should speed things up somewhat.

I'm not exactly sure if this is correct - I believe I heard it a the Linux Forum in Denmark back in march. The speaker was Jens Axboe, the current cdrom subsystem maintainer.

z

Re:BIO

[kill-1]

From Dave Jones' write-up (link in the post above)

CD Recording.
~~~~~~~~~~~~~
- Jens Axboe added the ability to use DMA for writing CDs on ATAPI devices. Writing CDs should be much faster than it was in 2.4, and also less prone to buffer underruns and the like.
- Updated cdrecord in rpm and tar.gz can be found at *.kernel.org/pub/linux/kernel/people/axboe/tools/
- With the above tools, you also no longer need ide-scsi in order to use an IDE CD writer.
- Ripping audio tracks off of CDs now also uses DMA and should be notably faster. You can also find an updated cdda2wav at the same location.
- Send good/bad reports of audio extraction with cdda2wav and burning with the modified cdrecord to Jens Axboe
- Currently only 'open by device name' works in cdrecord. cdrecord -dev=/dev/hdX -inq
- More info at http://lwn.net/Articles/13538/ & http://lwn.net/Articles/13160/

Re:devfs?

[rjw57]

Actually devfsd does (see here). Most distros use DevFS + devfsd these days (notable exeption off the top of my head is RedHat).

timeslice and 'hyperthreading'??

[elwinc]

Two possibly dumb questions (but this is slashdot, after all). (1) Can you change the scheduler default timeslice (10 msec seems a bit long for a multi-GHz CPU). (2) does it do the right thing for hyperthreading? (for hyperthreading, the scheduler needs to understand that one of the CPUs is sorta crippled, so jobs should flop back & forth between both CPUs).

Re:timeslice and 'hyperthreading'??

[nilsjuergens]

Replying to point (2):

The scheduler in 2.6.xx is hyperthreading-aware.
It knows that switching a process from one hyperthread to another on the same cpu is less expensive than switching to another physical cpu (becaus both first- and second-level cache reside on-die), but it also tries to balance load on physical cpus.

While >=2.4.19 supported hypterthreading up to a certain point it happend that two processes were running on the same cpu while the other (physical) cpu was running idle. This does not happen with the new ht-aware scheduler.

Look here for a (compressed) version of the initial discussion.

Re:timeslice and 'hyperthreading'??

[groomed]

The HZ value which determines scheduler granularity has been bumped from 100 (which gives 10ms granularity) to 1024 since 2.5.low-twenties or something. You can change the HZ value yourself on 2.4 kernels right now in fact.

Haven't heard much about scheduler/hyperthreading interaction.

Re:timeslice and 'hyperthreading'??

[TeknoHog]

You can change the HZ value yourself on 2.4 kernels right now in fact.

I think it requires the CK patch to change it. The patch also includes other low latency features which can be quite useful.

Product release cycles

[mfh]

Technical achivements aside, the most amusing thing about the 2.6 series of kernels is seeing all the large corporate entities with vested interests deal with the release schedule.

That is to say, there isn't one. I especially liked the quote from Torvalds I recently saw in a CNet news.com that basically said, "it'll be done when it's done - deal with it".

Re:Product release cycles

[dotwaffle]

This quote may sound quite arrogant and that Linux isn't suited to the business world - but think about it for a minute - what he said was "We're going to do it our way, making sure it all works, rather than release it before testing properly". I'm glad someone has the sense to do this. How long was it before Microsoft Apps get patched? Something like hours after release? I'd prefer something stable (or as stable as possible) on release, and I thank Linus (and Alan, and Dave) for taking their time. I'm pretty sure we all can't wait until these new features come out - but I'd rather wait for them to become stable instead of risk my precious collection of... well... you catch my drift. =)

Re:majjor new features?

[Anonymous Coward]

LINUX 2.6 KEY CHANGES

Faster, more predictable performance and new APIs are on tap
[Yay]

Desktop improvements
[Whatever that means]

Universal Serial Bus 2.0 and production Bluetooth support
[Yay]

Pre-emptible kernel with low-latency kernel patches for more user responsiveness and better multimedia performance, even under heavy loads
[Like Windows XP!]

Server improvements
[Whatever that means]

Updated I/O and memory subsystem for faster throughput and scalability
[Yeah, like blast processing]

Faster, more scalable process scheduler
[Pfft]

User-mode Linux to allow multiple system images running on the same box to aid server consolidation and application separation
[Sounds like the minutes of a business meeting]

Asynchronous I/O and completion events--a big improvement for Web servers and databases
[I'll take your word for it]

Support for disks larger than 2 terabytes and for SGI's XFS enterprise file system
[OK]

Faster, POSIX-compliant threading library
[Redundant]

took me a while to make it work...

[C0vardeAn0nim0]

and it still shows nothing on screen if i pass vga=normal during boot, and it took me several atempts before I relized that regular ps/2 keyboard can be left out or compiled as a module. well, this kind of changes were expected.

after i managed to get it working (booting, with keyboard, framebuffer console, et. all) surprise... no DRM on X.

happens that for some reason X doesn't detect working agp when a Radeon 8500LE in inserted in my kt266 based mobo. even with agpgart and radeon modules loaded.

so here's a few sugestions:

leave ps/2 kboard selected by default for x86 architectures, same for a way to display the console on text mode vga and check this radeon issue.

except those minor stuff, the new kernel is great. really fast for regular use.

How to install?

[KillerHamster]

I know Slashdot isn't a support forum, but could someone point out a good tutorial for compiling and installing a new kernel? I'd like to give 2.6 a try, but I don't know where to begin.

Re:How to install?

[caluml]

Very rough old notes.

Should help though
http://gk.umtstrial.co.uk/~calum/2.5-kernel/
Might update it if I get a few hits.

Let's call it LINUCS

[DeadVulcan]

In the GNU tradition:

Linucs Is Not Using Code from SCO

Re:Works, but no nvidia

[defMan]

Take a look at minion.de. They have patches for getting NVIDIA's driver going.

word of warning

[Maimun]

According to Alan Cox, there are security issues with 2.5.* (and thus with 2.6-test1)

Last time I checked there were remote DoS attacks and local root attacks present in 2.5.7x

See:

Re: Linux v2.6.0-test1

The whole thread is here Linux v2.6.0-test1

Re:Have they fixed SBP2 yet?

[Cthefuture]

Well, assuming that's a legitimate question and you're not just being a smart-ass (hard to tell)...

I used rfstool.

Still the same problems since 2.5.68

[hacker]

2.6.0-test1 is MUCH slower than 2.4.21 or 2.4.21-preempt-rml here. I see that the timing issues are still not fixed in 2.6.0-test1, and haven't been working since 2.5.68. I've reported this at least a dozen times to the appropriate people, with no fixes eminent yet.

To test this issue out, run Sawfish, and bind a key like Ctrl-Alt-B to a black-background xterm. Launch X, and run Sawfish. Hit Ctrl-Alt-B once and see what happens. It's consistant here across about 6 machines, all different hardware.. a 3-4 second delay, then anywhere from none to 4 xterms will open up. On 2.4.anything, it opens the xterm instantly, and only opens one of them, not 3, not none.

The other issue is that there's some underlying change in the TCP stack/net drivers that cause rsync and anything running over ssh/ipsec to fail with weird dropped-socket errors from the applications using them. Again, on 2.4, it works flawlessly.

It's very annoying, and both of these are blockers for me and most of the machines I'd be running this on. It happens with anything that involves keyboard shortcuts; menu accels, launched applications, keybindings, everything.

Changing to the different schedulers does not help; deadline, as, or cfq. 2.5.68 worked perfectly, and didn't have these anomalies, but every single kernel since that time, has had it. I've diffed, and I can't tell which of the dozens of changes actually broke this.

If anyone has a solution, I'm all ears.