Embedded Systems Study Rebutted

Gabba writes "LinuxDevices.com has a rebuttal to the Microsoft-funded report purporting to show Windows nearly 4X more efficient than Linux for developing embedded systems. The rebuttal shows the study to be full of flaws in both design and execution."

Comments

Define Embedded

[spoonist]

First, let me be honest. I just skimmed the LinuxDevices article and didn't read the Microsoft article.

One thing I've noticed among PHBs is an ever-broadening definition of "embedded systems". I've seen more than one project go down the road of using a cPCI system running Windows NT 3.51 (yes these are current systems running this old version) on a harddrive. These systems are calling themselves "embedded".

This has been especially in systems that had serious size, weight, and power needs. Had I designed the system, I guess I would've used something like QNX or Linux on a much smaller processor, compact flash card, etc.

I guess my point is that these days it seems like general-purpose computers are being called "embedded" when I see embedded as much, much smaller (e.g. no moving parts, a microcontroller, etc...).

I dunno, I'm rambling...

This Study *is* Flawed

[0x0d0a]

Okay, a lot of pro-Linux studies have their own problems (frankly, I don't put much stock in "studies" any more, especially vendor-funded ones).

However, the numbers this one used are *ridiculous*. Total Cost of Development?

Okay, let's see. Of the parties surveyed, cancelled Linux projects cost more than CE projects. This Jerry guy (he's got a PhD, so he must just be dishonest, not stupid) then uses this as a basis to claim that Linux is more expensive than CE. He's got to be kidding me.

By the same metric, all C++ software should be replaced by bash equivalents. Why? Because the average cost of a bash-based project is much, much smaller than the average cost of a C++ project. Of course, there's the little additional detail that the sort of projects one uses bash on are much, much smaller and simpler. That is, of course, the factor that makes the huge difference. However, you can conveniently ignore that tidbit.

Somebody tried to do the same study with Windows and some Sun servers back in the day to show that Windows made a far cheaper server. Well...yes, but most of the servers being used to average out Windows cost in the study were small, departmental servers that nobody was spending much on. The Sun servers were the far more powerful and capable systems for things like eBay's back end that had technicians swarming all over 'em. Sure enough, the Windows boxes had a lower average maintenance cost.

Average total cost is *totally useless* without some additional constraints so that you're measuring average cost of *similar projects*. If you took all PVRs with roughly equivalent feature sets and examined cost based on embedded OS, *then* you might have a useful study. The current one is totally useless other than for FUD use.