Adam has pretty nicely summarized my point of view. My interest is in workstations and servers. I'm interested in taking pretty much stock PC hardware finding the minimum distance hack that will make booting nearly bulletproof.
If the motherboard EEPROMs were commonly 512KB or bigger then the LinuxBIOS approach would be incredibly cool. It seems the most common size supported EEPROM size is 256KB. It seems Linux does not(??) fit comfortably in this space. It looks like LinuxBIOS is less practical than I'd like. The Disk-On-Chip and CompactFlash adapters are interesting but too single-source and more suited for projects hacking hardware. I'm imagining a boot sequence something like: 1) If no network, boot from disk. 2) If network and no disk, boot from network. Where (2) would effectively use the local disk as a cache. Now from browsing around the embedded Linux web pages, it seems that fitting a kernel into 256KB (compressed) is not impossible, but then I have not gone through the exercise. OTOH if a stripped down Linux kernel and a startup program can easily fit into a 256KB EEPROM, then we're back to cool :). From: Adam Agnew > I'd like to use the linux kernel as a bootloader too. But there are > problems with that. > > A lot of people who want to use LinuxBIOS are not thinking about clusters. > They have desktops and servers and want to replace their BIOS and still > proceed in desktop and server functionality. > The Linux kernel isn't fitting in 256k anytime soon. It's a tight fit in > 512k. Commodity consumer motherboards aren't shipping with 1Mbyte eeproms > any time soon. > > People don't want to swap out their prom for a disk on chip. They don't > want to patch or compile a kernel. They don't want installing LinuxBIOS to > take all day. They want simple functionality in a small amount of time. > The linux booting linux scheme isn't delivering that for the > desktop/server user right now. > > Linux booting linux works great for cluster nodes. The admittadly small > but annoying investment on a DiskOnChip for every node is worth it. > Spending a day getting just the right functionality out of the linux > kernel is worth it. People running clusters are experianced and > knowledgable so it's not a big deal. > > But when some system administrator wants to use LinuxBIOS I want to be > able to point him to our rom-o-matic type web site, a place to download a > working bootloader, and a nice little utility to make it all come together > so he never has to think about it again. LinuxBIOS gains popularity more > quickly and maybe some chipset makers or motherboard manufacturers take > notice and start to make our lives easier. > > I like the leaving everything to linux philosophy too. And if all > motherboards shipped with big generous eeproms we wouldn't even be having > this discussion. What can I say? I need a way to boot LinuxBIOS on 256k > for desktop and server machines.
