How not to sell embedded Linux

> Mr. Ready is perpetrating the fallacy that, to build an embedded system with Linux, one
starts with the various components and integrates them all by hand. If a company were to take
that path, it might well incur the high costs that Mr. Ready warns about. Creating your own
distribution - and maintaining it over a product's life - is, indeed, a difficult and
expensive job. 


I've seen it done.

It's tough coming from a PC-Linux background and seeing people who are experienced developers
otherwise deal with trying to roll their own Linux system when they are just trying to get
started.

It's absolutely a great learning experience and for that that approach can be valuable.. but
they don't quite understand that Linux systems available from other organizations  don't have
be taken at face value.. the benefit of Linux is a flexible system, but you do not have to do
it all by yourself  from scratch to get that.  But these developers, I think, generally only
see things from two perspectives... source code and proprietary binary distributions. That
is.. its either something you have to do yourself or you buy in a packaged deal from somebody
else. 

It's the middle ground between those two extremes is were the value is at. Take what you can
directly from distributions or vendors and benefit from their experience and widely used and
tested systems... and then modify it to suite your needs only when you have to. 

Each modification you make is more work for you. 

Linux distros provide the 90%, you have to do the 10%. 

That is the sort of thing that embedded developers often don't 'get' right away. Once they
start down the path of total customization then they can get trapped in it and it just makes
things more and more difficult and expensive as they diverge with aging code bases, most of
which ends up being a in-house fork and the rest of the Linux world moves on.

I think this sort of problem is more common then is apparent. For us experienced in PC or
Server stuff it's a no-brainer. For embedded developers it's not so obvious.

I guess he is not trying to scare customers from their services but scare customers from
replacing them. Montavista obviously adds value to their products, the problem is that
companies with good engineers simply won't need it. 

As comments and original post noted, we are not talking about rocket science here and these
companies' bigger competition may be their own customers. As Linux evolves (for instance a
great deal of RT patches and other stuff that helps embedded Linux users is being gradually
merged in the 2.6 line) it becomes increasingly easy to roll your own Linux (from a free
distribution) modified for your needs. And thus avoiding buying Montavista's or other
company's solutions.

They obviously want companies to use their services instead of build their own.

Almost related to the topic, not long ago I got an unsolicited sales call from Redhat (!),
with the classic fud-based strategy used against Debian:

http://www.hanno.de/blog/2007/fud-in-the-linux-world/

One problem the embedded Linux vendors have is they are still trying to behave in proprietary
ways - and with Linux at the core of your embedded project it becomes increasingly clear the
vendors are trying to lock you into them.

The last embedded project I did with Linux (in fact the first I did) we held the option of
going for vendor supplied kernel open in case we needed harder real-time support. In the end
it turned out we didn't need that so all they had left to offer was fancy (proprietary) GUIs
and distro's which would have tied the project to the vendor if we had used them. Unless you
are going for a heavily configured PC type application rolling your own distro isn't that hard
and it means you have a good understanding of what is in your system and why.

I suspect as more and more RT and embedded patches make it into the mainline kernel
(windriver's 81 patches seeming rather poultry) the value proposition of embedded Linux
companies will look increasingly weak. There is a role for them in the space somewhere but
they need to prove their worth.

"by Dan O'Dowd, Green Hills Software"

While it is clear to embedded.com readers, lwn.net readers might not be 
aware that Green Hills is a company that sells a proprietary embedded OS.

So it's just marketing as usual.

Take a look at Wind River or Montavista's web pages, they're so full of glossy marketing stuff
and buzzwords. Instead of saying the truth, "for some cash, we can save you a few weeks of
work in bringing up your Linux board", they're trying to build up some kind of hype. 

I think a lot of people decide to roll their own system (and with the help of projects like
Buildroot it can be quite painless, if you're lucky). In part this is because they're trying
to save money, but it's also an uncertainty on how much value these commercial vendors really
add. You can compare the situation to the desktop Linux world, where many people (me at least)
prefer upstream fixes over distribution-specific add-ons. 

Going from a modern PC Linux distribution into the embedded Linux world can sometimes feel a
bit like going back in time 10 years. It can sometimes be a bit rough around the edges, but
for an engineer that's also part of the fun. The joy of bringing up a board for the first time
and seeing the boot messages scrolling by over the serial port, priceless!

that the point he was trying to make was "here's this security critical 9-line patch, which is
entirely dependent on these other patches, which are in turn heavily dependent on these other
patches, and by the time you get back to the major release, you're at 800kloc."

Whether that's a reasonable approach to take is open to question.

But it's at least not blue sky.