Cycles and Simplicities (Linux Journal) [LWN.net]

Doc Searls writes about the tendency for companies to become mired in the tracks of their own success. "It's strange to think of Google and Facebook as old, but Dave's right. They are. Search is old. Advertising is old. Online social communities in a big walled garden is old. You can look at it this way: Google fixed Lycos's problem. (And Infoseek's, and Hotbot's, and AltaVista's.) And then it fixed the yellow pages' and classified advertising's problems. And it used the proceeds from both to start fixing many other problems too."

to post comments

Cycles and Simplicities (Linux Journal)

Posted Dec 6, 2008 1:17 UTC (Sat) by cpeterso (guest, #305) [Link] (9 responses)

Linux is a good example of the "The Innovator's Dilemma" mentioned in Doc's article. Linux replaced Solaris in many many server rooms, even though it was technically "worse". But it was better in other ways that mattered to many people. Eventually Linux gained traction and reached parity or superiority to the more mature Solaris features.

Linus made a comment once (that I wish I could find) about the kid who will write the Linux Killer. Like Linus in his dorm room, some kid might look at Linux and extract some nugget of simplicity and wisdom from Linux's accumulated baggage.

Linux Killer

Posted Dec 6, 2008 2:02 UTC (Sat) by dark (guest, #8483) [Link] (8 responses)

Hmm, but how often does it happen that a dominant free software project is replaced like that? I honestly can't think of any cases. Creating a Linux Killer would mean finding an advantage that is compelling enough to merit replacing Linux, but not compelling enough to merit redesigning Linux. A free software project is in a good position to absorb all upcoming competitors. Like with Fortran, whatever ends up replacing Linux will still be called Linux.

What I expect to see in the free software world is what I call the "Apache model". There is a dominant project which is the default choice for its task. It is the one that everybody hacks on and it is insanely configurable. There are a few specialized workloads for which a conceptually different or radically simplified design works better; small projects spring up to fill these niches. The relationship between the projects is mutually beneficial: the dominant project does not have to break its design to accommodate the niches, and the niche projects do not have to generalize.

I can imagine a radical change of the "Linux Killer" level if technology changes so that one of those niches becomes the norm for that task. But not otherwise.

Linux Killer

Posted Dec 6, 2008 2:27 UTC (Sat) by smitty_one_each (subscriber, #28989) [Link] (2 responses)

You're taking a purely technical view.
In the postmodern world, the Linux Killer needs to be reasonably good on the technical merit (especially shiny new hardware, say, optical storage) and just have a fantastic marketing approach.
Style over substance, man.

Linux Killer

Posted Dec 6, 2008 12:23 UTC (Sat) by drag (guest, #31333) [Link] (1 responses)

Maybe.

But there are very better ways then Linux. There has to be. It's just that nobody has figured it out yet.

------------------------------

One thing that I like is Samba 4. Over 50% of the source code for that program is created by other programs. A lot of that is swig bindings for Python scripting, but also it's due to auto-generated code to build support for RPC protocols and such that is built from IDL description files.

It maybe that a small team of people could end up describing a better OS kernel in a text file and have the tedious details of actually programming the kernel be taken care of by other programs.

Linux Killer

Posted Dec 10, 2008 23:01 UTC (Wed) by man_ls (guest, #15091) [Link]

Interesting. But keep in mind that whatever is used to build the program is in itself another program, only that it might be more understandable than the target code. A description (such as an IDL) can be clearer than a step-by-step program.

Just for fun, surely there has to be a better way to program device drivers than C code. Maybe even some kind of IDLs since a driver must talk the device's low level protocol? Who knows...

Linux Killer

Posted Dec 7, 2008 5:39 UTC (Sun) by proski (guest, #104) [Link] (3 responses)

I suggest that you actually read The Innovator's Dilemma. Even a free software project would not absorb changes that make it less valuable for its existing users. Disruptive technologies are not created by following demands of the customers of the existing technologies. They are created for different users and different uses, and then mature and capture new (and old) markets.

If you take a broader look, whatever replaced Fortran is not called Fortran.

Linux Killer

Posted Dec 7, 2008 18:11 UTC (Sun) by JoeBuck (subscriber, #2330) [Link] (2 responses)

While a free software project isn't going to absorb changes that make it less useful to its existing users, anyone can grab the code and fork it, in a way that makes it less useful to the existing users of the original code, but opens it up for a new use.

Linux Killer

Posted Dec 8, 2008 21:07 UTC (Mon) by proski (guest, #104) [Link] (1 responses)

As far as I know, C compilers were not made by forking Fortran compilers, even if the copyright wasn't an issue. Free software may be more resilient to disruptive technologies, but it remains to be seen to which degree.

Linux Killer

Posted Dec 9, 2008 2:49 UTC (Tue) by Ze (guest, #54182) [Link]

A fortran compiler was made from a c compiler though :p

Gfortran is part of/based on GCC.

Linux Killer

Posted Dec 8, 2008 23:42 UTC (Mon) by iabervon (subscriber, #722) [Link]

I doubt that anything could kill Linux directly; it's too adaptable, maintainable, and configurable. A direct competitor would probably just become a configuration option or be a fork in the first place.

I think a real Linux killer would actually be something like Gnome. Some project that's complementary to Linux could reach the point where it's not necessary or even worthwhile to write things for POSIX and Linux, because the complementary project provides all the services you want better anyway. And people would still be using Linux, but they might as well be using Solaris because all of the interesting stuff would be elsewhere.

I can envision a future where there isn't anything interesting to do in the core parts of Linux and it's shipped installed in ROM on motherboards instead as the BIOS. On your bulk storage drive, you've got driver modules for the critical stuff, and userspace daemons for the interesting stuff. And the programs people use directly run in a filesystem namespace where the only mounted filesystem is the Gnome VFS FUSE module. And sound is done with the PulseAudio CUSE device factory. And the interesting work is in dynamically configuring programs to be able to search each other's filesystem metadata across virtual machines running on a NUMA cluster within the user's desktop case as well as extending into their household storage cluster and to some network services they use. And Linux would be part of the stack, in much the same way that the PCI bus is part of the stack, but it's not a big deal.