The 2.7 Kernel: Back To The Future For Linux

Anonymous Coward writes "Now that the Linux 2.6 kernel has been released and is being worked into distributions, many in the open-source community are turning their attention to the next development and test kernel, known as the 2.7 tree. To get an early glimpse at some of the thinking going into the next kernel, key vendors that aid in shaping the Linux kernel helped eWEEK last week put together a long-range wish list for 2.7."

Comments

would you believe?

[whovian]

There's a Microsoft ad on that page! Something about linux TCO.... Um, thanks. Don't need an ad for that.

MPPE?

[Malc]

Is there any reason why after all these years we don't have MPPE in a stock kernel? I always have to get a specially built kernel so that I can use pppd to connect to a MSFT/Windows VPN server. I use somebody else's build (deb http://www.vanadac.com/~dajhorn/projects/debian-pp tp woody main) which makes my life much easier, but it's not released as fast as the stock kernels.

Monolithic kernel and Unix philosophy?

[lawpoop]

Does it strike anyone else as strange that everyone keeps dreaming up more stuff to throw into the kernel? What happened to the unix philosophy of small, independent programs that do one thing well?

I'm aware of projects such as The Hurd -- this seems to follow closely the unix philosophy, but it's a ways off from general usability. Others have noted that it's usually easier to debug a monolithic program than to debug communication problems between small unixy programs. (Maybe there is some way to make a communications chart of said small programs, so that it looks like monolithic code? )

Discuss.

Re:So...

[hcg50a]

Don't get mad at Microsoft; get mad at eWeek for placing the the silly ad where they placed it.

I thought it was hilarious for the ad to be completely surrounded by the article about the Linux Kernel release.

Almost makes you wish SCO was in the news business....

just some SATA support

[ducman]

After a frustrating weekend trying to get a High Point SATA card working in my Linux server, I'm putting better SATA support on the top my my wish list!

One has to wonder

[krammit]

With so many people with their own agendas pushing and pulling at the kernel, and Linus being the steadfast leader he is, I can't help but think Linux may be headed for a fork in the not so distant future. Unless there is a way to make the kernel truly enterprise class as well as a responsive, low latency desktop system and a near real time embedded platform all at the same time.

I'm amazed (in the good way) the kernel devs have made it as versatile as they have to this point. Hats off to them and here's to hoping they can keep it up.

Re:Focus switching to the desktop

[Anonymous Coward]

That was quoted from an Oracle Exec. let's not get ahead of ourselves. I'm not sure how 2.6 is considered a "server" release. The Kernel is the kernel. 2.6 will be the default kernel on desktop installs in a few months I'm sure.

A lot of the patches in 2.6 benefit both the server and desktop camps equally. The scheduler and VM improvments and XFS. I believe RedHat backports those patches to the 2.4 kernel for the ES/AS/WS versions.

If you haven't tried 2.6 yet, you really should. I noticed a considerable increase in X response time with it.

Clutering Finialy

[silas_moeckel]

Not much infomation in the article but I must admit it would be nice to start having SAN/Cluster filesystems as part of stock kernels. People realy dont understand the power of these filesystems to provide security and scaleability. With modern cluters inconnects being able to serve up fiber channel multigigabit ethernet and low latency interconnects it gets easier and easier to make pure diskless compute nodes that are for more than just number chrunching.

Think about only needing a single copy of your web server image mounted read only to the web servers themselves.

Setting up CAD farms that all utilize direct attached storage in a shared method leaving network bottlenecks behind.

Low end systems like firewire may even be able to attach single disks between multiple machines with similtanious access (have to check on multi initiator firewire looks posible never seen a definate though) in a safe manner.

not actually true

[Anonymous Coward]

The 2.6 kernel included a lot of stuff to give better consumer hardware support and make interactive applications more responsive. Characterizing it as a server release is Just Plain Wrong, despite the enhancements for scalability.

Besides, if virtualiation is the big feature of 2.7/2.8, that is much more of a server feature than a consumer feature. Sounds to me like 2.7 will be the server development series, and 2.6 the desktop release.

Re:Maybe I'm not smart...

[smittyoneeach]

Vague, yes, but boring?

Sam Greenblatt, senior vice president and chief architect of Computer Associates International Inc.'s Linux Technology Group, in Islandia, N.Y., agreed about the need for virtualization technologies. "Right now, you can have multiple Linuxes virtually operating, but we would love to see that expanded so that you could power other operating systems, whether that be Unix or Windows,"

Now, I don't know if they mean something like ReactOS or not, but if you had a Linux that could deftly boot a RedmondOS, such that you could tap into all of its drivers, how cool would that be?
I end up running XP so that my HP6110 driver can give me double-sided, four sheets per page printouts of those endless .ppt-gone-.pdf lectures for school.
Don't get me wrong, I love my RH9 and all, but the pragmatist runs Linux for love, and Redmond products when it makes sense.
Such a potential capability in Linux must be soiling laundry in the State of Washington even as you read this.

What I would like to see

[ChiralSoftware]

I would like to see less things in the 2.7 kernel than in the 2.6 kernel. Getting device drivers, network drivers, etc, out of the kernel core and into modules was a step forward, but I think the next step forward would be to get these things out of the kernel entirely, and into userland. That would give Linux a huge advantage over Microsoft Windows. Installing and un-installing device drivers would become much easier for users. Manufacturers would like this too because then there would be less concern about GPL and device drivers. It would be easier to release binary-only drivers.

Re:What I would like to see

[Sloppy]

I think the next step forward would be to get these things out of the kernel entirely, and into userland. That would give Linux a huge advantage over Microsoft Windows. Installing and un-installing device drivers would become much easier for users

Maybe you should try HURD. I don't think Linux is going to go in the direction that you want.

Re:What I would like to see

[mandolin]

Playing devil's advocate here. I'm sure you're already aware of most of these points.

Is insmod so difficult?

First, you'd really want modprobe. Second, for the few not using their distributions' modules, the point is that it is still more difficult than running an executable. Usually because the module in question needs to be compiled against your particular kernel, which is much less backward/forward compatible than glibc.

So, any other good reasons why you'd want userland drivers?

It should be more robust. You're not subject to kernel limitations (C language only, fixed 8k stack to play with, etc.) You can use more standard APIs, which are better documented and which also lead to better portability, if you can modularize your code well enough. There's a reason the XFree86 drivers aren't completely in-kernel.

Are those reasons good enough to offset the additional overhead that this would incur (additional context switching,etc)?

For low-throughput devices like serial ports, keyboards, and mice, it's quite possible. At least it makes for an interesting thought experiment.

The new layers of indirection that would have to be added?

Need to be more specific here. The amount of indirection depends on the driver to be "converted" and the way you approach it.

My wish

[chrysrobyn]

I want filesystem priorities. A background task that is grinding the hard drive, should only do so when a high priority task isn't using the drive, or when its data is adjacent to the high priority data the head is next to anyway.

Better security would be nice

[unoengborg]

Today windows is plagued with viruses, trojans and worms. If Linux usage becomes more wide spread among users with little knowledge in computers, networks and security, we might see similar problems in Linux in the future. The fact that Linux is a much better acrhitecture than windows will probably not be enough to protect Linux from incompetent users.

To prevent this, it would be nice if some kind of sandboxing technology was implemented. E.g it could be based on digital signature technologies, where applications could be given capabilities depending on who signed them. That way an adminstrator could allow only applications signed by approved vendors or himself to run on the system.

Apart from raising the security in Linux it could, provided it is done right, also make it clear to users that DRM technologies a la Microsot is there to provide security for the content provider, not the user. That is if Linux in reality was just as secure as MS technology is in theory, nobody would accept hardware locking users out of their own system.

Re:What would be a great "desktop focus"

[NeoThermic]

>Yes, I am speaking about windows-program-clones.

Well, after my switch from Windows to RedHat, I found this helpful:
equivalents / replacements / analogs of Windows software for linux.

NeoThermic

Re:Dear Linus,

[swillden]

how is it possible to change a user on x without logging out?!

There are many ways, depending on exactly what you want to accomplish. I suspect the one you're talking about is "Start New Session" -- on KDE 3.2*, just click the "K" menu and pick "Start New Session".

Using that plus vnc you can even make your X session relocatable -- again supported in point-and-click fashion out of the box by KDE 3.2 (called Desktop sharing). Wife is using the computer in the den? Just pull up your still-running desktop on the machine in the kitchen. Do that with XP!

The coolest way, though, is this one. This guy dropped two video cards into his machine, hooked up two keyboards and mice and set things up so that both he and his girlfriend could use the machine at the same time. Granted, this isn't something that can be done out of the box (it requires running two different X servers, one patched), but it's a very cool hack.

* I'm sure GNOME has similar features, since KDE isn't actually doing any of the multi-session heavy lifting, that's part of XFree86. KDE just puts a pretty interface on it.

Re:Dear Linus,

[FreemanPatrickHenry]

"Universal Plug an Play" [slashdot.org]

here with the Lindows [lindows.com] distribution

porn browsers such as Pornzilla [netscape.com].

It's things like these that let you know the parent post is copied from another /. post.

One Userland Improvement

[Bruha]

I would like to see something in the nature of a area where all executable commands for any user software get put into.. Many programs today install theirselves into various /usr /usr/share /usr/local it just goes on and on. Reguardless of where the program installs itself I think a top level directory /usr/software where all programs put in a link back to it's working directory and main executable for all programs..

That way all users know that their programs reside in /usr/software and it makes it easier for plugin/mod authors to know where things are.

Either way if this is not feasable then it's time to standardize where things are going.. Windows has it's Program Files which went a long way towards fixing user confustion :) where people now know their programs (With very few exceptions) now end up.

Re:One Userland Improvement

[JahToasted]

That's not exactly the kernel's job. It's the guys that put together your distro (Redhat, Debian, etc.) that make that decision.

Each system has its advantages and disadvantages. By putting all files in C:\Program Files\program_name, yeah it keeps the apps nice and organised in their own directory (well not really) but it makes the command line pretty much useless. You would either have to add every subdirectory of Program Files to the path or type in the full path of the programme you want to run everytime. ugh

With the unix filesystem layout you only have to add /usr/bin, /usr/local/bin and possible /opt/bin to your path to be able to run any app from the command line. Shared libraries can easily be used by any programme by just looking in /usr/lib or /usr/local/lib.

Besides which, not everything can be organised in their own subdirectory, even in windows. shared libraries for example. Ever notice C:\Program Files\Common Files ? also in windows programmes create entries in this thing called the Registry. So if you try to delete a programme by deleting its subdirectory in Program Files you won't get everything and it may actually cause errors (because of the registry). So you have to use a programme to uninstall things for you (add/remove in windows, apt-get or whatever in linux).

The windows way looks simpler at first glance, but really it has a few disadvatages (can't use CLI) and no real advantage (either way you have to use a package manager).

Re:One Userland Improvement

[RDPIII]

This has nothing to do with the Linux kernel per se.

Still, the file system hierarchy is basically fine the way it is defined in the Filesystem Hierarchy Standard, LSB, etc. FHS has been around for a long time, at least eight years, as far as I can recall.

If I compile software that isn't already part of a distro myself, I tend to configure those packages with --prefix=/usr/local/stow and then use stow to install symlinks under /usr/local. That's pretty close to what you're suggesting, no?

If it was informative maybe

[bluGill]

Sure, but it wasn't really informative. About 4 ideas were suggested, and a lot of effort was put into making sure we knew who the players were. I still haven't figgured out what amazon.com wants, but they get a couple paragraphs.

What is going to happen? I still don't know after reading this. Well I can make a couple of guesses. Some clustering support. a couple other things. Not everyone wants all of the above.

I gaurentee that a lot more will go into 2.7 than the above. This gives me no clue as to what though. It was a waste of time reading that artical.

FHS

[krmt]

I don't get it. Isn't this what the FHS already solves? I know Debian follows the FHS as part of policy, and so everything basically has a set place where it goes. The basics work like this:

• Binaries meant for normal users go in to /usr/bin, unless they're part of the base system, in which case they go in to /bin. If they're part of XFree86's special playground, then they go in to /usr/X11R6/bin, but that's really an ugly holdover more than anything.
• Binaries for administrators go in to /sbin or /usr/sbin
• Shared libraries go in to /usr/lib or /lib, depending on how close to the base system it is. Sometimes they put their own subfolder in /usr/lib, but not as often.
• Executables meant just for the app and not the user, as well as images, sounds, etc go in to /usr/share/appname
• Documents go in to /usr/share/doc/
• System-wide config files go in to /etc

This is all really well established, and I'd be surprised if all the major dists didn't follow it. It's not really that complex, especially when normal users really only have to know about /bin and /usr/bin. It's also not very complex ultimately, since once you start working with it things are exactly where you expect them to be, and besides, the packager package manager (or port-type manager, does emerge's type of soft have a general term?) should be managing these things for you. Next time you're on a Debian system, try checking out /usr/share/doc/packagename for whatever program you're interested in. You'll find tons of good info.

Re:So...

[Dick Faze]

Think so? I'm not trying to start an argument, but I think this is a case of "Don't attribute to malice that which can adequately be explained by stupidity". Microsoft probably DID have some general requirement that their ad be placed next to articles specifically containing Linux subject matter, however, the reason for this was probably to target senior IT staff considering using Linux for specific projects. The Kernel article isn't going to be read by this demographic, its going to be read by dedicated linux heads (and those who endeavor to be). The management types will read the first three lines, realize they're reading a 'techie' article, and go off in search of higher-level material....and Microsoft has just wasted a few thousand $$$$ preaching to those least likely to listen!

Re:So...

[WhiteDeath]

I actually read the ad....
I wonder - did they use people who had no experience with windows to compare against the support costs for people who had no experience with linux?

Given that a windows desktop server can cost several thousands of dollars to buy software for, before you pay someone to actually install and configure it, are they saying it cost them several thousanddollars to get the linux server working?

Takes me less than a day to get a working, configured server linux server... (two if I download all the software).

Ongoing costs? Yes, they did have to read the manual for the linux software... But i'd have to read the manual for the Windows software if I wanted a non-default config.

As for the "case studies" I wonder how much it cost M$ to send someone out to walk them through the changeover? Might not have cost that customer, but It sure didn't come out of Bill's pocket!

Interesting Facts: Giga Research is a wholly owned subsidiary of Forrester Research, who changed their policy on paid-for product comparisons as a result of at least a similar study, if not the one touted in the advert.

In their defence (or perhaps not), Forrester did find that MP3s are good for the music industry...

Meta Group will say anything: (not that I don't like the idea, but wouldn't you try to "correct" a firm saying this about you?)
By 2006 or 2007 Linux will be running on 45% of new server
again on eeek (I notice that has a HP ad on it) er, eWeek - but I like the typo better :-)

IDC - well.... IDC: Microsoft breakup would benefit the industry and a quote from here
"IDC has also published research in the past that shows some companies replacing Unix systems with Linux can save twice as much as those that move from Unix to Windows".

Re:Advertising in the article

[metasyntactic]

Nice selective quoting. Here's the full paragraph:

"Amazon, which has been running Linux since 2000, has been steadily moving its infrastructure from Sun Microsystems Inc.'s Unix servers to Hewlett-Packard Co. ProLiant servers running Linux. The company said in a 2001 Securities and Exchange Commission filing that Linux cut its technology expenses by $16 million, or 25 percent."

They cut costs 25 percent over Sun servers (presumably running solaris). It makes no mention of Windows 2003.

Wish List

[Anonymous Coward]

Hot Swappable processor support (obviously on 2+ processor systems), hot swappable memory support (obviously on systems with more than 1 stick of memory). Better system-within-a-system support (currently you can run Linux under Linux). Clusters would be nice. At some point, SOC systems will come along, at which point support for that will need to be put in. Selectable schedulers would be nice too (although that is too much bother for most people).

OT: Re:would you believe?

[rcamera]

in windows, put this file in 'C:\WINDOWS\system32\drivers\etc'. for linux, it goes in '/etc'

the real trick is changing the 'action canceled' message into a plain-white (or your favorite bgcolor) page so blocked ads show up as a color patch instead of a text message...

credit where credit is due, i did not create this file. i downloaded it pretty much how it appears (but from where i can't remember)