Slashdot

If 90% of what a user does is web browsing and email, that sounds like a good bet. If you push "on" and have it up and running in a few seconds, who would bother going into Windows? You'd only need to boot to Windows when doing some office work or the like, and that boot option would be a quick-click icon. If you primarily do office work with it, then you'd want a full-blown "regular" laptop anyhow instead of a netbook.

However, I imagine that Microsoft will find some way to sabotage multi-OS-boot options via screwy licensing and pricing games.
         

From Chrome and Chrome, What is Chrome? [cringely.com]: "The most interesting part for me will be Microsoft's response. This strikes at the very heart of Redmond's business success and Microsoft will not take it lying down. Expect thermonuclear warfare."

It seems if you are aiming to have a very narrow and specific design to your system, a general purpose Unix work-a-like is overkill. Wouldn't a minimal POSIX-ish system with some graphical operations be sufficient. It's great to use something familiar and actively developed like Linux. Just for the device drivers alone it is pretty valuable. But after digging into the Plan9 kernel, I realize that most of these drivers are not really that complicated if you can accept a basic level of functionality and less than optimal level of performance. (like the nvidia drivers in Plan9, it's only one short .c file, and just enough to get 9wm up and going). Even something like L4 is overkill, a lot of the cool abstraction it offers is probably not necessary if you can just wedge it into a library.

Many of us on here have hacked together little pseudo-kernels. Glorified Hello World bootloaders really. If you had a TCP/IP stack, using an existing one like KAME or uIP, or a new implementation (I don't care which) and a filesystem that is more like a simple memory mapped key-value pair database (using critbit, hash table, b+tree, whatever). it seems to me that would be enough to get something like WebKit going.

What value would a custom kernel/OS have over a specialized Linux? Well I think you could focus on implementing abstractions most suitable for a browser instead of trying to fit a filesystem or sqlite library to your design. Mostly I suspect you could optimize the boot of a very primitive system pretty easily. And you could do things where isolation of the browser in memory can be done in a way much finer grain than the Unix scheme of dividing everything into a user process or kernel mode thread.

Perhaps the browser would be more like a root user, but individual tabs would have permissions controlled by a kernel or hypervisor that would be in isolation of one another. One page may not be able to hijack the rest of your browser or access cookies or passwords unless specifically authorized. And it could be done in such a way that is still relatively fast and low overhead, but more secure than current schemes.

Imagine if plug-ins like flash and video codecs had to run through a socket or some fast IPC messaging scheme. where you could just close it to force the process on the end to shut down.

Why don't I implement it you ask? Well assuming I have the skills necessary to do a good job, and the ambition to complete such a task. I'm too old school to accept the idea that a system where the only application is a browser is useful to me personally. Maybe when kernel development becomes browser based?