On Mon, 9 Dec 2002, Andrew Kohlsmith wrote:
Now in quantity (i.e. clusters) I think that LinuxBIOS is probably cheaper in the long run (I'm sure that Ron's done all of these analyses and so forth since it appears it's for government use) and I really am glad that Ron and the toher contributors have done what they did. It's always embarassing to write off something as being fringe and then a year or so later find yourself asking questions about it because you need to use it yourself. :-)
Actually lots of people have found uses we would not have thought of, far outside our domain of use in clusters. I have been quite surprised.
The biggest complaints I hear about the proprietary BIOSes (these are not my opinions, as a person at a USG-funded lab I am not allowed to have such opinions) - very poor code -- tons and tons of assembly that is hard to figure out - limited control -- the bios does what it is going to do, and if your hardware doesn't quite fit a PC model, you have to work around it. Example: you have something in a DIMM slot that looks like memory but is not. You're going to have to design that hardware in a special way to make sure the BIOS doesn't configure it as memory. With LinuxBIOS, that is not an issue -- you can fix this kind of thing trivially. - very high overhead costs -- initial flat fee + ongoing per-year - high unit cost -- on a $50 motherboard, you can spend a huge chunk of that on bios license. So if you get rid of the BIOS cost, you either clear more money as profit or get to be more competitive by reducing unit costs
Again, I don't endorse these claims, as I am not allowed to endorse things; but I have heard them many times from various vendors and I find them interesting.
ron