On Sat, 21 Apr 2001, Charles Steinkuehler wrote:
> Well, I generall think almost EVERYTHING should be modules. You can regain
> IDE support for booting by adding the modules to the initial ramdisk (the
> linuxrc mods I posted a while ago for my SCSI-RAID support do this).
Yeah, Oxygen does the same, and I was planning on incorporating that into
anything that I cranked out to begin with.
> I'm envisioning a single kernel (or perhaps 2, if we don't want the module
> hooks for RAID, IPSec, and various other higher-end functions taking up
> space on a single floppy firewall), along with a small number of pre-rolled
> initial ram-disks (floppy-only, IDE HDD, IDE HDD+CD-ROM, Flash/DOC support,
> etc). Anyone wanting a really funky or custom boot system can roll their
> own initrd (or make a custom root.lrp if we stick with the initrd-archive
> patch & use a tar.gz file).
Now this will take some pondering.
Time to do some good old-fashioned "market classification" here. We have
two base-level types of people using LRP:
1. People who want to have a firewall/router that will let them share IP
addresses and don't want to spend the money on a commercially available
one.
2. People who want to tinker, and as such have a fair bit of knowledge.
I know this is severe overgeneralization, but bear with me. Anyone who
would be using this for a commercial project, or would be setting it up as
a server, is someone who wants to tinker with it. This ranges from the
sysadmin that gets the 'net connectivity project for the company tossed at
them because, "hey, he's the tech guy right?", all the way up to those of
us on this list that're crazy enough to do the devel work. The former will
want something he or she knows, and the latter is already doing stuff with
us.
So, if we've got two targets, with two differing styles and needs, why not
make two kernels as you suggest?
The first one is the bare router kernel. It'll do firewalling, NAT,
perhaps DMZ stuff, and be pretty much as we know LRP or EigerStein today.
The second will be much more powerful, with all the goodies that any geek
could want, everything from ext2 support on up through the PCMCIA devices
and Disk-On-Chip stuff. (Ham Radio, anyone?)
The problem then becomes where we draw the line between the two. For
example, more and more people are getting USB DSL Modems foisted off on
them, and so the USB stuff becomes necessary for the basic kernel, even
though it's still somewhat on the bleeding edge. Advanced stuff like WAN
Router, QoS, and packet radio can be reserved for the beefier kernel.
IPSec/PPTP/VPN stuff should go for Basic, because a lot of people want an
IPSec kernel to connect to their office LANs from home. The reason why I
say that we can go with the more complex stuff for advanced users is,
simply, these are the folks that are gonna be willing to give it a go with
the CD-ROMs, CompactFlash, or even hard drives. The space will be
available to them in general, one way or another.
Any thoughts or ideas? I'm thinking that trimming the fat off of this
stuff, combined with UPX, might be enough for us to go glibc 2.1.x or even
2.2.x for base router images. At least then, it would be easier to
transition from the basics to the fun stuff.
--
George Metz
Commercial Routing Engineer
[EMAIL PROTECTED]
"We know what deterrence was with 'mutually assured destruction' during
the Cold War. But what is deterrence in information warfare?" -- Brigadier
General Douglas Richardson, USAF, Commander - Space Warfare Center
_______________________________________________
Leaf-devel mailing list
[EMAIL PROTECTED]
http://lists.sourceforge.net/lists/listinfo/leaf-devel
