[linux-elitists] Kernel-mode Linux
Wed Dec 11 04:54:18 PST 2002
On Wed, 2002-12-11 at 13:33, Nick 'Sharkey' Moore wrote:
> On Wed, Dec 11, 2002 at 11:26:04AM +0200, Gilad Ben-Yossef wrote:
> > This may sound crazy for desktop/server systems but for embedded devices
> > it might be very useful - in many embedded systems where the machine is
> > only doing a single thing anyway you might be willing to sacrifice the
> > protection for performance.
> So, remind me, why are we embedding a multi-user time-sharing
> preemptive-multi-tasking Unix-ish OS in such a device anyway?
Think router. Obviously, Linux is perfect for such an application. You
leave all the general purpose stuff (web interface, telnet etc.)
untouched but you have one very specific task (no, it doesn't have to be
a single process) where speed and overhead are very important - say it's
some VoIP NAT application. You execute that specific task under KML.
Sure, if you've got a bug in that task you bring down the whole machine
instead of just re-spawning a process. As I said there's a trade off
involved. You can cut your losses but having a hardware watchdog and
you've narrowing down booting time and you also decide to write really
good code :-)
Now, you *develop* your VoIP NAT mangler as a user space app, which is
much easier then writing a kernel module (just the tough of having a
kernel module to parse ASN1 gives me the creeps...) and when it's mature
enough you move to running it under KML.
Is this the choice I would make every time? nope. But it's a viable
option. In fact, most RTOS do exactly this but they also come with a
expensive per device license fee. ;-)
Gilad Ben-Yossef <email@example.com>
"Denial really is a river in Eygept."
More information about the linux-elitists